Mo Bitar’s hilarious video painted the current state of AI hype in both tech and AI Twitter/YouTube, where the only way to keep your job was to sell hype to your employer.

Make it known that you’re the expert in AI and find ways to replace others before they replace you, so that your trigger-happy employer doesn’t add you to the layoff list.

You have to sell that you’re all in on AI, even when you’re a skeptic of it, just to be in the good books of your employer.

You have to constantly stuff AI buzzwords down people’s throats that you don’t even know how they work, just to stay relevant.

But, what if there’s another way to use AI in your job that keeps you safe without becoming another obnoxious co-worker?

You don’t need to know everything about AI

Being AI-native was never about using every single feature that launches every day.

The engagement farmers will constantly tell you that you’ll be left behind if you don’t try them, but it’s just a major distraction.

Most of the models and new tools that are hyped online won’t be relevant in the next few months.

I’ve never tried a Ralph loop before, and I don’t intend to, until I’m sure that I can give clear instructions to the AI to achieve the outcome I want.

Falling into the AI Tool Trap puts you into another hamster wheel that you can’t escape:

• A new feature gets released

• Everyone hypes it up on Twitter and YouTube

• You fall for the hype and try using it

• You spend more time troubleshooting than doing anything meaningful

• You get frustrated and revert to your old workflows

Which is ultimately just a huge waste of time.

So ignore all the buzzwords that everyone is throwing around.

Harness, context engineering, model quantisation, none of that matters to you now.

Instead, just focus on this:

Build practical workflows with AI to automate the tasks you hate doing

While automation seems like a dream that everyone wants to achieve, we don’t want to give AI the wheel so it runs our lives entirely.

There are still things that we enjoy doing, and AI’s main aim is to automate the tasks that we absolutely hate.

The soul-sucking, draining, repetitive tasks that we do day in, day out, that are mindless, and we wish someone could do them for us.

Those are the tasks that we should be outsourcing to AI, while we have more cognitive load to focus on the tasks that we enjoy doing.

So instead of building a flashy tool with the sole purpose of impressing your employer (who couldn’t care less), build with AI to achieve these 2 outcomes:

You prove your value by making or saving more money

The more directly tied to revenue your job is, the more valuauble are to your employer.

If you can prove that whatever tool you build with AI makes or saves money for the business, you will be safe.

And it all starts with this:

Find one painful bottleneck that stops the company from making or saving more money in your workflow, and build a solution with AI to remove the bottleneck.

There’s no need for any hype when you have substance and real outcomes (instead of burning tokens and raising the expenditure of your company).

What if you could become AI-native in less time?

You can build practical workflows with AI, even if you have zero technical experience like me.

After spending the past 4 months with the sole purpose of getting good at AI, I have built a system that helps me automate all the tasks that I hate doing.

This is what I’ll be sharing more of in The AI-Native Sprint, a 90-minute crash course that gives you a clear path for busy 9-to-5 professionals to become AI-Native.

If you’d like to join us, sign up for the waitlist here:

Join the Waitlist

But I want to help you avoid going through all the frustrations as I did.

2 months ago, I was excited to build my OpenClaw agent.

I spent a week researching the perfect setup and model to use with it. But when I bought my Tencent Cloud VPS, I had no idea of the challenges I faced.

It became a nightmare just trying to get everything to work, where I didn’t even reach the stage of creating meaningful outputs with it yet.

And that’s when I saw the hype over Hermes, and decided to switch over too. Everyone was saying how it was better and the setup was more beginner-friendly.

So I bought another VPS (Hetzner). The setup was definitely easier (likely because I was more experienced after the OpenClaw nightmare), but there were still parts that kept breaking.

Of course, I’m not saying that it’s a bad technology. I’m still amazed at how I could pop into Discord, chat with my agent to carry out a task, and it just gets it done.

But at this stage, for beginners like myself, it’s a major time sink and distraction. You’ll likely spend more time getting things to work than doing anything meaningful with it.

I recently shut down my Hermes agent for this very reason,

and here’s what I’d recommend doing instead of building an agent:

The learning curve is still too steep

While installing OpenClaw, there were so many things I had to learn along the way:

• Reading and understanding the documentation (it’s still not beginner-friendly yet)

• Navigating the Terminal setup

• Finding files within the explorer

Most YouTube videos talk about how easy it is to set up OpenClaw, but they glaze over all the important details that a beginner would need to know.

And of course, a bunch of them were just shilling Hostinger as the VPS provider because of the affiliate commissions, even though it was one of the costliest plans.

Even though Tencent Cloud provided a one-click install for OpenClaw, I had absolutely no idea what to do next after launching the VPS.

And their documentation was horrendous too.

I knew nothing about SSH and the ability to connect to my VPS via the Terminal.

I didn’t know that VS Code had an extension so I could access all of the files within my VPS.

I didn’t know that I should not have installed OpenClaw on the root account.

There was no clear guide that went through all of these specifics, and I had to figure it out all by myself.

Using OpenClaw or Hermes via a VPS is not the only way, and the other method was running it on a local computer?

But what about those who do not have a spare MacBook lying around?

Because you wouldn’t want it anywhere near your main computer:

They are still not secure yet

I don’t fully trust either agent with my API keys or main accounts.

Not after what happened when I tried to create an onchain trading bot. I specified that each trade should be $20 as I was still testing out the trades.

But my mistake was giving it $400, and it decided to make a trade with my full amount.

It’s possible that an agent could do something irreversible to your finances or accounts, and the only thing it’ll tell you is: “Let me be honest, I made a mistake”.

At that point, there’s nothing else you can do except regret giving your agent full access to your accounts.

Because there’s no one else you can blame at that point in time.

Memory is still a major issue for most agents. Some can continue to execute the tasks, until one day, it decides to forget everything.

Are you willing to risk ruining your entire life to automate some of these tasks? I’m not, especially when it comes to my finances.

So the ideal way to set up your agent is by giving it completely separate accounts that are not linked with yours.

Though now, its capabilities will be limited and you can’t automate tasks that require your credentials.

Most APIs still lack scoped access and guarded controls, so it will take a while before agents become more useful.

But at this point, it’s more like an all-or-nothing scenario:

Either you give it full access to your account (which is risky), or none at all (which limits its usefulness).

Most of the features are just shiny tools

Unless we’re business owners who have extremely complex workflows, most of what we see online doesn’t apply to us.

Yes, they are cool and flashy. They make you feel productive by setting it up.

But in the end, do we really have a need for them?

These were some comments I’ve received on my YouTube videos that talked about Hermes:

Some people know how to set it up, but they don’t know what to use it for.

Others were saying how they are just shiny toys that don’t do anything useful.

I know this too, because I spent days trying to get a multi-agent setup to work, without even learning how to run OpenClaw in the first place.

I was inspired by the hype posts that I saw online, and convinced myself that I should set it up now so that I don’t have to worry about the infrastructure.

But instead, I wasted time on creating new Discord bots, getting them to work in each channel, and having the gimmick of agents being able to talk to each other.

And at the end of the day, I had no idea what I should be using my 5 different agents for.

I have shiny object syndrome and feel that the only way to keep up with AI is by trying new tools. But you’ll just be stuck inside the AI Tool Hamster Wheel of starting from scratch and having nothing to show for it in the end.

I no longer touch my OpenClaw agent, and I completely shut down my Hermes agent once I finally understood this.

You can’t automate what you don’t know

Agents seem fantastic because you could tell it to make you 6-figures every month, and in theory, it will help you to do just that.

But because you (and me) have no idea how to create a concrete plan to make these 6 figures, the agent will spew out average advice that anyone can give you, so there’s no edge at all.

Without proper guidance and context, because you’re not even sure of the workflow:

You can’t expect AI to do anything useful, since it’s an incredibly intelligent intern but is completely lost at execution.

That’s why I completely switched my focus to building Effective Skills:

These are repeatable workflows in your .claude or .agents folder that are invokable by a slash command, and you can run an entire workflow with it.

Here are some examples of the Skills I built or adapted from other Skill libraries.

Before using Hermes and OpenClaw, I was building out an Obsidian + LLM tool (which I call my Portable AI System), where the core of it was my:

• Context (who I am, what I do, how I think)

• Skills (repeatable SOPs)

And now, I’ve completely switched back to this system.

Yes, I lose out on automation and cron jobs. Yes, I will lose out to anyone who builds insane workflows and runs everything on autopilot.

But I’d rather spend my time nailing the gold standard of my workflow now, instead of stressing over Hermes or OpenClaw messing up everything because my instructions weren’t clear.

So I’ve made a promise to myself that I will only touch agents again once I’m confident that my AI-native workflows will consistently produce the output that I want.

Build workflows first before agents

It will be tempting to jump straight into building agents, because of how easy everyone online makes it seem.

But once you’re in, you could be spending hours on fake productivity, troubleshooting all the connections without producing real outputs from it.

So instead, focus on building Effective Skills:

Turn the current tasks that you have into an AI-native workflow.

Determine what tasks you hate doing, and find a way to outsource them to AI.

Test out the workflow through multiple runs (because you won’t get the right workflow immediately), before moving to full autonomous automation.

After spending 4 months learning how to get good at AI, I’ve settled on a system to build these workflows that can be integrated into your job (and even your daily life).

This is what I’ll be sharing more of in The AI-Native Sprint, a 90-minute crash course that gives you a clear path to becoming AI-Native for busy 9-to-5 professionals.

If you’d like to join us, sign up for the waitlist here:

Join the Waitlist

What vibe-coded site can I build with all the fancy features that will wow my employer?

What multi-tool and agentic workflow can I build, so I can show I know how to use all of these tools?

But all of these are just distractions that waste your time.

Especially if you’re new to coding (like I was), and suddenly the world is your oyster. Where you could literally build anything so long as you describe it to AI.

As someone who fell into that trap (and eventually got laid off because I had nothing to show for with my AI usage except for burning unnecessary tokens), I should have known this instead:

Being AI-native is about building something useful

Your employer doesn’t care about the number of tokens you burned or the number of tools you integrated into your workflow.

But this is the only thing they care about:

Does this site, tool, or workflow help to make the company more money, or save more of it?

If whatever you built doesn’t do either of these, it’s just a poor ROI in the eyes of your employer.

They are paying you a salary to figure things out with an AI subscription they paid for, and will be disappointed if you have nothing to show for it.

Especially if you took time off from your actual work to build it.

So rather than focusing on something flashy, this is what I’d do:

Start with one problem

AI is excellent as a mindless task executor, or a knowledgeable but clueless intern. If you give it a clear SOP (Skill) and context of what task it needs to complete, it can give you an output that is close to what you expect.

Of course, this takes many iterations (and time) to reach that stage, but I ask myself this question when integrating AI in my workflow:

What is the single most annoying, soul-sucking, repetitive task that I absolutely hate doing inside this workflow, and how can I outsource it to AI?

Usually, this involves a lot of mindless clicking.

One task I absolutely hated was the copying and pasting of every image I had inside my article draft (written in Notion or Obsidian) to the publishing platform (Twitter, Substack, Medium).

Each image involved 5-10 seconds of copying and pasting, and you could imagine the time I wasted doing this for 20 images and repurposing across all 3 platforms.

So I built a workflow with the help of AI to convert my Obsidian markdown file into an HTML file with all the embedded images.

Instead of copying and pasting every image, I could now do this in 5 seconds per platform.

And it all started out with the task that I didn’t want to do anymore, and it was perfect for AI to execute it for me.

I’m sure that in your own workflows, there are tasks that you hate doing too.

The ones that drain so much of your time and you wish someone else could do it for you.

Now, you don’t need to hire an intern, and just outsource it to AI instead.

Being AI-native starts with understanding your workflows

Instead of trying to build something fancy, find ways to make or save money for your employer with AI.

And all it takes is:

• Understanding the workflows you do in your 9-to-5

• Working together with AI to find an automated solution

After spending 4 months learning how to get good at AI, I’ve settled on a system to build these workflows that can be integrated into your job (and even your daily life).

This is what I’ll be sharing more of in The AI-Native Sprint, a 90-minute crash course that gives you a clear path to becoming AI-Native for busy 9-to-5 professionals.

If you’d like to join us, sign up for the waitlist here:

Join the Waitlist

And who knows if your company will be next.

It’s frustrating that your job has never felt so unsafe. At any moment, your employer could pull the plug and decide to go all in on AI.

And that leaves you with the unhappy process of finding yet another job, in a market where most employers are questioning why they should hire someone else instead of just using AI.

But, I don’t think this will happen to most of you, unless you fall into this group:

The mindless task executors are unsafe

I got laid off in Feb. At first, I felt that it was unfair.

But then, I finally understood why I was no longer valuable to my company:

I was excellent at following orders, but had zero agency. I was too afraid of making mistakes, and I became an executor, rather than a strategist.

As a marketer, I followed whatever that my manager planned to do. Write this article, create this graphic. And whenever I was asked to give ideas, I had none.

Because I was too scared that my ideas sounded stupid.

Since AI could do my job better, faster, and cheaper, it made sense why my employer would swap out my salary for a $100/month Claude plan.

For those who fall in the same category as me (a mindless task executor):

There could be a higher chance that AI replaces you.

AI is excellent as an execution engine. So long as you give it the direction, the expected output, and a clear workflow:

It will do that task almost to perfection.

And those who survive the next wave of layoffs are those who possess this skill:

Being AI-native starts with mastering Skills

Everyone talks about the high-income skills that you need to survive the AI era.

To me, the only one that you should focus on is building good Skills:

These are repeatable SOPs that you give to Claude, ChatGPT, or any other LLM, and they execute them according to that workflow.

Skills are the most effective when they tell LLMs to execute tasks that you hated doing every single day:

• Searching old docs and chat messages to find something that your teammate said previously

• Filling in claim forms manually for your claimable expenses

• Formatting reports so they look presentable

When employers want you to be AI-native, it’s not about how many vibe-coded projects you can produce every week.

Instead, it’s finding ways to integrate AI into your workflows:

What steps can be automated by AI, while which ones should a human still do?

When you free up your time and cognitive load by outsourcing the right tasks to AI, you become more valuable to your employer.

So if there was only one skill that I would master now, it would be to build Effective Skills with the SIGNAL Framework:

• S: Spot the bottleneck

• I: Integrate the stack

• G: Guide the LLM

• N: Nail the standard

• A: Automate the logic

• L: Loop recursively

So long as you can describe the problem and bottleneck in plain English, you can build a Skill without any coding experience.

You decide on the ‘what’, while AI finds the ‘how’ to execute it.

Become AI-native in less time by learning this skill

Building good Skills will be something that your employers will value, especially when it saves them time on the soul-sucking, boring tasks that you and they hate doing.

And after spending the past 4 months learning how to get good at AI, I perfected the SIGNAL framework to build any Skill to automate my work, instead of relying on another automation platform like n8n.

This is what I’ll be sharing more of in The AI-Native Sprint, a 90-minute crash course that gives you a clear path to becoming AI-Native for busy 9-to-5 professionals.

If you’d like to join us, sign up for the waitlist here:

Join the Wailist

• Claude adds a new feature

• A new platform releases its MCP

• ChatGPT launches a new model

Infographics like this go viral online, and you’re overwhelmed by the sheer number of new tools to try.

Someone proclaims on Twitter that this new agent changes everything.

A YouTuber warns that if you’re not using this feature, you’re already falling behind.

And if you’re constantly consuming content like this, it’s the easiest way to feel inadequate about yourself.

You know that you need to catch up with the latest tools to stay ahead, but you don’t have time to spare.

So you burn whatever free time you have trying to understand how it works.

You get frustrated when nothing seems to be working, and ultimately give up with nothing to show for.

But this tool-chasing habit that everyone tells you to adopt is the completely wrong approach to becoming AI-native.

Tools are just distractions meant to trap you in this hamster wheel

The platforms and models that you use today won’t be the same ones that you use in 3 months.

So it’s a huge waste of time comparing them, because most are already capable enough.

Time that you can’t afford at all.

I know this because I was stuck in this loop too.

When I first started learning how to use AI effectively, I was too focused on choosing the right model and tool for me.

I saw everyone shilling a new tool that supposedly automates my entire life, and I wasted precious time getting it to work.

Not forgetting the days that I burned just getting OpenClaw and Hermes to work, when I could have used that time to create real outputs.

I was constantly stuck on this hamster wheel:

• A new feature gets released

• Everyone hypes it up on Twitter and YouTube

• I fall for the hype and try using it

• I spend more time troubleshooting than doing anything meaningful

• I get frustrated and revert to my old workflows

Whatever you see online is meant to trap you inside this hamster wheel.

They use fear-mongering tactics to convince you that your existing setup is useless (when that’s completely false).

And then I finally realised that none of these matters if I wanted to get good with AI.

What if you could become AI-native in less time?

There’s no point in spending all your free time trying to master Claude Code, when it could be replaced tomorrow.

Instead of focusing on the tools, I’ve learned that building an AI system that automates all the work I hate is the fastest way to become AI-native.

Your employer won’t care if you’re using the latest tools, if you can’t give them the outputs that they need.

And after spending 4 months learning how to get good at AI, I’ve settled on a system that completely removes all the distractions and focuses on automating workflows instead.

This is what I’ll be sharing more of in The AI-Native Sprint, a 90-minute crash course that gives you a clear path to becoming AI-Native for busy professionals.

If you’d like to join us, sign up for the waitlist here:

Join Waitlist

Time is so tight, and it’s impossible for you to try everything that gets shilled on Twitter or YouTube.

As a non-coder, you think that you’re at a disadvantage because that’s the only way to become AI fluent.

But that’s completely wrong.

You don’t have to spend time learning code to become good at AI. Of course, it’s more beneficial for you to know it, but anyone can build anything with AI now.

If you know how to describe your problem in plain English, AI can create a solution for it and execute it for you.

While I’ve been unemployed over the past month, I’ve been using AI extensively to get better at it.

So I’ve gone through all the failures, getting frustrated when things don’t work, and spent all the time finding out what works for non-coder like myself to become AI fluent.

And here’s the framework I wish I knew before I started:

What AI is, and isn’t

Before we start, let’s get this point out of the way.

AI is not a magic pill that can drop 6 figures in your lap with just one simple prompt.

If that were the case, everyone would be millionaires already.

But after using AI extensively since January, this is what I realised about AI:

It is an amplifier of systems by automating the soul-sucking, repetitive tasks.

But if you don’t have a system or clear workflows, it will amplify that as well. And the result comes out as an amplification of a messy system that you’ll spend more time troubleshooting the system than getting real outputs you could use in your life or job.

Trust me, I’ve wasted weeks of my life just getting OpenClaw and Hermes to work, while accomplishing nothing.

There’s no point in building agents that automate every task for you, when you don’t even know your workflow well enough.

It’s impossible to hand off everything to AI completely. You still need to give inputs on what works and what doesn’t, based on your own taste and judgment, while AI is just the execution engine.

Quoting from Albert Romero, who introduced me to this concept:

AI collapsed the how, so the only thing that matters is determining on the what.

So to get good at AI, you need to get good at determining what to use AI for (because it’s great at some things, but not at others).

And this is how you can do it:

The FOCUS Method to get good at AI

With all the shiny new tools that are constantly being shilled on Twitter and YouTube, you’re likely overwhelmed.

And with every new feature that Claude or ChatGPT launches, you’re tempted to throw whatever that you’ve built away to start using this new feature.

But to become AI fluent, you just need focus.

Focus on the fundamentals, understand what AI can do for your workflows, and then try the tools afterwards once you have a better understanding.

With time as your most precious resource, we need to optimise it so you don’t waste it on actions that won’t help you become AI fluent.

This is the FOCUS Method that I use to learn AI:

• F: Filter the noise

• O: Optimise your context

• C: Concentrate on one win

• U: Use it daily

• S: Scale your Skills

And here’s each step in detail:

Filter the noise

AI Twitter and YouTube are the perfect places to make you feel completely inadequate.

There are so many proclamations of how everyone is:

• Running a fully autonomous company with AI

• Replacing a $120k/month department with just Claude Code

• Using 10 agents to get them to 100 million views

But 90% of what you see online is just noise. The ultimate skill to possess right now is skepticism, because nothing can be trusted.

It’s just a waste of time scrolling on the timeline, when you could be building something that solves a painful problem in your own life.

I found one way to solve this is by DM-ing the account that makes all of these insane claims. How did they actually do it?

If they can’t answer you clearly, then it’s likely just farming engagement.

It’s impossible for someone to become an expert at a new feature just 5 hours after it was launched.

So don’t fall into the trap of feeling bad about yourself by comparing with accounts that do nothing, yet are acting as they do.

I approach this by taking inspiration from what others post and see how I can add it to my current AI system.

Since the AI knows me enough after already executing multiple tasks for me, it provides a recommendation on whether this new feature is something worth integrating or not.

And it all starts with providing this to AI:

Optimise your context

AI works best when it knows who you are, what you do, and how you think.

And the easiest way to provide this is through your past work (blog posts etc.) or conversations.

Otherwise, it’ll just give you generic outputs when you don’t provide it with enough direction on what it should focus on.

So throughout your day, start compiling everything that you’re willing to share with the AI, so that it understands you better:

• Your tweets, articles, and meeting transcripts

• Your chat history with AI platforms

• Important documents about your personal life or work

The more you share, the more AI knows you better. But of course, there’s this fine line of privacy where anyone who gains access to your system knows everything about you.

So it really depends on how much you’re willing to share.

LLMs love markdown (.md) files as it’s LLM-friendly, so I spent my time converting all of my files into this format.

While it was still a painful process, I’m grateful that most of my work was done in Notion, so I could export my pages as markdown files.

For any files (.docx, pptx) that are not already in the markdown format, I like using this MarkItDown library to convert anything into a .md file that I can give to the LLM to understand me better.

This is a one-time task that can be time-consuming at the start, but it saves you the hassle in the future.

Once you have all of your context compiled, you can throw it inside any LLM to analyse and understand everything about you.

This also means that proper file organisation is crucial inside your AI system.

For mine, I organise it between Context and Tasks, so everything about me goes into the Context folder that is used by the AI to complete the Tasks that I want to complete.

Concentrate on one win

I have a bad habit of shiny object syndrome, where I have so many workflows that I want to automate with AI.

Even when one workflow is not fully fleshed out yet, I want to immediately move on to the next one.

And that could be a problem that you face too.

It’s better to fully flesh out one workflow that consistently gets you the output that you need, instead of building multiple AI-assisted workflows that require you to constantly troubleshoot them.

Something boring that gets the job done is way more valuable than a flashy workflow that you’re just building to get views on Twitter.

Everything else about AI doesn’t matter, like:

• What model you should use to achieve the task (because most are already capable enough)

• What tools you should be using (some of the ones that exist today won’t last in a few years’ time)

We don’t need to waste time worrying about the most cost-efficient model when we’re on a monthly plan and can swap models easily while finding the best one that meets our needs (some are even free too).

And this is only possible when you build a Portable AI System that lets you switch out components (like your models or tools), and it still gets the job done.

So instead of worrying about any of this, find one bottleneck that’s stopping you from doing a task better or faster, and work on using AI to remove that bottleneck.

This involves building Effective Skills, which are hyper-personalised SOPs that give you high-quality outputs, no matter what model you use.

And you don’t need to write them out at all, or even know the right script that automates the workflow.

Just by going through a dry run together with the AI, it has enough context to build the Skill for you.

I shared more about the framework I use to build these Skills here.

Use it daily

AI is not a tool that you can use once and get good at instantly.

It has to be something that you do every single day, just like practising the piano.

So commit to spending at least one hour a day playing and experimenting with AI.

This doesn’t mean you have to build out a vibe-coded platform every day.

Instead, work together with AI on finding bottlenecks in your workflows, and finding ways to integrate AI inside.

One of the first tasks that I automated was generating summaries for YouTube videos.

In the past, this was the workflow that I used:

• Copy the URL from YouTube

• Paste it into a YouTube-to-transcript site

• Copy the transcript

• Paste it into ChatGPT

• Paste my summariser prompt inside ChatGPT

• Generate the summary

This took about 2-3 minutes and a lot of mindless copy-pasting and getting frustrated when I couldn’t find the summariser prompt.

So instead, I built a Skill that helps me achieve this:

I type /learnings-youtube with the URL, and it captures the transcript and generates the summary as specified by my prompt without me doing anything else.

Everything is saved inside my Portable OS, so the LLM can retrieve it in future when I ask it to.

From 2-3 minutes of work being reduced to less than 30 seconds.

And all of this was done with the help of AI.

I gave it this problem and the workflow that I wanted to achieve (including where the files should be stored), and it came up with the solution itself.

I’m sure that many of your workflows can be automated by offloading all of the boring, soul-sucking tasks that you hate doing to AI.

And it all starts with:

• shaping your problems together with AI

• Solving them with automated scripts that AI generates for you

• Telling AI what went right and wrong

The more you work with AI, the more you understand its capabilities, and the more work you can outsource to it.

Scale your Skills

After building out a single Skill, it’s time to start scaling them.

You now have the knowledge to:

• Identify the bottleneck in your workflow

• Shaping the problem together with AI

• Directing AI to the ideal output

And this can now be used across every workflow that you hate doing.

But this final step takes practice. You have to keep playing with AI in this manner to get better at building Effective Skills.

Becoming AI fluent doesn’t mean getting another certificate

It is hard to learn AI when you’re already pressed for time every day.

Your time is precious, so don’t waste it on getting another certificate that means nothing.

Instead, what others are looking out for are real workflows that you have automated for yourself or your clients that save them time.

Being AI fluent doesn’t have to be anything fancy. So long as you can show a result for someone (especially saving time), your perceived value increases to your employer or future clients (just like what I’m trying to do here).

But of course, please don’t farm engagement like what most are doing on Twitter or YouTube.

In the end, the only two things that differentiate you from everyone else are your:

• Context: AI will continue giving generic outputs, unless you give it enough context about who you are, what you do, and how you think

• Skills: Repeatable SOPs that any model can follow to automate all the boring, soul-sucking tasks that we hate doing

Anything else can be swapped out if you build your AI system with portability in mind, especially when the AI cost war will make subscription plans worse in the future.

So to automate the tasks that you hate with a system you truly own and is not locked in with any providers:

Get the blueprint to build your Portable AI System here.

I’m trying to understand who currently reads Stack Signal and how I can make it more valuable for you.

If you have two minutes, take the reader survey here, and I’ll share what I find in a future issue.

I write about AI every week for non-coders who are overwhelmed by all the noise online. If you want to build workflows that automate the tasks that you hate doing, subscribe below:

Subscribe now

I get it, because that was exactly me back in Jan 2026 as a non-coder.

Everyone on Twitter is giving you the wrong advice:

• “Steal my prompt to make your first 6 figures with AI”

• “Use the latest models, or you’ll get left behind”

• “Learn prompt engineering so you don’t get replaced by AI”

I tried all the latest tools, watched all the YouTube videos out there, but nothing seemed to click.

I kept getting generic outputs that I’d never use.

I spent more time teaching AI how to perform the tasks I did with it a few days ago, rather than executing them.

After locking in for 2 months (while I was on hospitalisation leave), I realised I was approaching AI the wrong way.

To be truly successful with AI, you need to build an effective AI system.

And all of them have 2 things in common:

• They include strong context about the user

AI will continue giving generic outputs, unless you give it enough context about who you are, what you do, and how you think.

The more context you give it, the more it understands about you, and the less time you take to re-explain yourself with every session.

• They have Skills that automate their work

Models are just our employees, and we need a way for any model to understand how we work and execute tasks according to how we want.

Skills are these repeatable SOPs that any model can follow to automate all the boring, soul-sucking tasks that we hate doing.

It was never about the best model or prompt.

So long as you can articulate your workflow and problem clearly, AI will help you automate any task (even if you have no coding experience).

You don’t need agents or a complex setup

The more time I spend with AI, the more I realise how Hermes and OpenClaw are major distractions.

Yes, they are effective and can do amazing work. But only for those with clear systems and workflows that AI can understand and execute immediately.

While for the rest of us, it’s a waste of time building out these automations when we don’t even know what to automate yet.

There’s no need to use all these shiny tools. Instead, we just need a working system where we can use AI to free us from all the tasks we hate doing.

Over the past 3 months, I’ve been trying and failing with different setups, until I found the most adaptable way to build one.

I call it the Portable AI System, which is designed to:

• Understand your workflows and save you time by doing the tasks you hate

• Be flexible enough so you can switch model providers and still get a similar quality of outputs

In this course, I am focused on building out a system that saves you 2+ hours every week.

And all you need is a clear path to achieve that, without any fluff or over-sensationalised news.

Here’s what you receive that makes it very difficult to fail:

• The fastest way to set up an AI client, connect a free model, and run your first Skill in one session

• The fundamentals of the Portable AI System, so you know exactly how AI can automate the work you hate without wasting time

• A pre-built repo structure that gives you the starting point, which you can structure and re-organise as you wish

• Learn the SIGNAL framework to turn any annoying daily task into a repeatable Skill (no matter which model you end up using)

• How to turn your prompts that are found in 5 different folders into a system that works on any model you choose

I created this course for my past self.

This is the course I wish I had, so I didn’t waste my time watching YouTube tutorials or trying the latest shiny tool that added nothing to my workflows.

This is what I would do if I wanted to become good at AI in the shortest time possible, without worrying about the best model or prompt.

If you want to stop getting overwhelmed by AI and start building a system that is adaptable to any new change, now is the perfect opportunity to start.

Navigating the course

This course is a full newsletter with multiple sections. Depending on your device, here’s how you can get to your intended section:

• Desktop: A table of contents can be found on the left-hand side

• Mobile: No navigation, but your scrolling progress will be saved so you can go back to where you left off

The 8 components

The Portable AI System is made up of 8 components for it to fully function:

Read more

Models are more expensive now, rate limits are getting worse on coding plans, and us consumers are all trapped in subscriptions that no longer serve our purpose.

For those who don’t have any additional laptops or GPUs lying around (like me) and don’t want to buy new hardware just to run local models, the options of using AI are getting increasingly limited.

Within these few weeks, we’ve seen instances like:

• GitHub Copilot pausing signups for their Pro plan (just when I wanted to try it out)

• Claude silently removing Claude Code from the Pro plan and putting it back (likely after the backlash)

This is only the start, and other model providers and aggregators will follow the same route of nerfing their plans:

Because it no longer makes any sense for them to give out inference that costs more than what consumers are paying in their plans.

My first experience with AI was through Claude Code and I grew to love the Claude Code extension inside VS Code because of the great UI.

But when I wanted to switch from Claude and try out other models inside my system, it was complicated and I spent too much time troubleshooting it.

It’s likely easy for someone who understands code and the config.json files, but I didn’t want to continuously find different configs each time when I switched plans.

And that’s when I finally realised something about the state of models today:

Models are just commodities

Whenever a new model launches like Opus 4.7 or GPT-5.5, the whole timeline goes crazy.

There are many claims of ‘it’s so over’ all over the timeline because the new model is more capable and will replace jobs in many industries.

The posts go on to show the new benchmarks of the latest model updates and how it surpasses other existing models.

Everyone on Twitter tells you to upgrade to the latest models to get the latest benefits.

Most of it is usually engagement bait. But because of all the over-sensationalised news and fear-mongering, you’re convinced that the latest models are the only way to get work done with AI.

You drop everything and sign up for the latest Claude or ChatGPT plan, having to start again from scratch because the model knows nothing about you.

You waste time re-explaining everything to the model, because importing everything from your previous system is too time-consuming.

And all that time is wasted when you could be producing real outputs that go towards achieving your goals.

So now comes the most important question:

Do you really need the latest model to suddenly become good with AI?

If your work involves building and running applications that are used by millions, probably.

But for non-coders like myself, we don’t need the best model to execute tasks. The ones that can give us the output that we want and at a lower cost is more ideal.

As compared to running a more expensive model that gives us a slightly better output.

But maybe it’s because I’m using AI differently from others. Instead of consistently vibe coding (I did build a few projects but it’s not my priority), I’m using AI to automate all the tasks that I hate doing.

The soul-sucking, repetitive tasks that do not require a lot of thinking, but I still have to do them for my life to function.

And that’s where I see AI is most effective at:

I can outsource these tasks that I hate doing to AI, while I have more cognitive load to worry about the tasks that I enjoy (or those that require thinking).

Functionality beats aesthetic, and all that matters is getting a result.

One example I use AI now is to generate HTML content for my articles, so I can repurpose them in 1 copy-paste function.

In the past, I had a nightmare when I wanted to repost my article across Substack, Twitter, Medium, and any other long-form platform, mainly because of the images.

Copying directly from Notion or Obsidian didn’t work. My old workflow involved manually copying and pasting every single image into each section within the article.

And it was a mindless, tedious chore to do this same task over and over again across all of these platforms.

But now with the help of AI, I generate an HTML file that I can easily copy and paste directly into any long-form platform.

Though I still have to do it manually for Twitter, but it saved so much of my cognitive load.

This is just one example of how AI has become an executor of the tasks that I hate doing.

And back to the point, it doesn’t matter what model you’re using.

Most models are capable enough to execute daily tasks that we hate doing. Just tell AI the problem that you’re facing, the platforms you use to get the task done, and it will figure out how to do it.

At the end of the day, models are just commodities.

You can easily switch between any of them and still get similar results.

Of course, the frontier models will give you better results. But sometimes, a lower-quality output at a cheaper cost is good enough.

And I wanted to build an AI system that gives me that portability to switch between providers at any time:

Most of your AI system is easily interchangeable

When I first tried Obsidian many years back, I thought it was extremely bare bones with a steep learning curve (and decided to use Notion instead).

But when I tried Obsidian again to build out my AI system, I finally learnt how to appreciate its open system.

For my existing Notion system, I had to use an API or MCP for the LLM to understand and work with it, which was more tedious.

Meanwhile, Obsidian was just a collection of folders and markdown files that the LLM can easily read (.md files are LLM-friendly) and edit at any time.

While all my notes truly belong to me, instead of being held ‘hostage’ by another software company.

This goes along well with @kepano’s whole principle of self-guaranteeing promises:

One that is verifiable, non-reversible, and does not require you to trust anyone.

Anything within your current system can break, or better options will pop up.

Is your system adaptable enough to switch any of these components and still function?

Ever since I wanted to move away from Claude Code, I’ve been obsessed with building such a system, which I now call the Portable AI System.

Inside, I have 8 core components that are required for my system to function:

• *P*ersonalised Context: A file system that includes your context and tasks so any model knows who you are, what you do, and how you think

• *O*wning Your Keys: A BYOK provider that lets you easily switch between models

• *R*equests: Your input (voice or text) that gives commands so the AI runs tasks for you

• *T*hought Processor: The LLM acts as the brain that receives an input and provides an output

• *A*ctionable Workflows: Skills that are repeatable and can be invoked by any LLM

• *B*uilding Capabilities: Tools like MCPs or APIs that increase the capabilities of your system to external platforms

• *L*inking Across Devices: Syncing your portable system across multiple devices

• *E*diting & Managing: Using a markdown editor to write and manage anything in your system

Most of my system is interchangeable:

• Models will come and go, and I can switch to the most cost-effective one that gives me the outputs I want

• There are many BYOK providers that let me add my own API keys to use their platform (just that I like OpenCode for now)

• I can use any tool for voice requests (just that I bought a lifetime licence for Letterly)

• There are plenty of markdown editors (but Obsidian is still king)

• APIs and MCPs can be switched whenever I use a new platform

• GitHub is the best for syncing and version history, but a new platform may pop up at any time

At any time, I can switch any of these 6 components to something newer or better. And the system still continues to function while giving me the outputs that I want.

Though there are only 2 components that truly belong to me (and that’s how I won’t be replaced by AI):

Nothing else is more valuable than your context and skills

The best outputs from a model come when it has these 2 things from you:

• Your context (a file system that helps the LLM understand who you are, what you do, and how you think)

• Your Skills (repeatable workflows that you’ve built out for the AI to execute for you with clear instructions on your specifications)

If you ask AI to ‘write a viral tweet’, the output will be average because it knows nothing about the type of content you write, or the templates that you want it to follow.

But if you have a library of your past tweets that display your tone and your main ideas, and you have some tweet templates (or a prompt) that you want to follow:

It will give a tailored output that fits your needs, instead of something that is so generic.

And that’s something that becomes the most valuable asset of your entire system:

• Your thoughts, perspectives, and the type of content you consume

• The specific workflows that you use to get an output that you want

Everyone has their own way or workflow of doing the same thing, like writing an article or managing their to-do calendar. The steps we take to complete the same task will be different and cannot be copied.

And if you can articulate that workflow clearly to AI (via creating a Skill), it can execute it in exactly the way you want.

Which is why I don’t recommend downloading and using someone else’s skill, and I prefer adapting those skills to fit my own context.

The models you use are the commodity, while your context and Skills are the true assets that are irreplaceable and are the core for your system to continue functioning.

Any of the 6 components can be swapped out from the system, but context and Skills are eternal.

You need to have zero dependencies in your system

Most YouTube videos out there share all the amazing things with Claude Code.

But you can execute the exact same tasks in a system that doesn’t require you to just be locked in with Claude or any other ecosystem.

There are many models that will offer promotions or free trials, just like what OpenAI did for me.

Or you could get sign-up bonuses like Kiro with 500 free credits that you can use.

And because my system is flexible enough, I can switch between any IDE or model and still get the outputs I want (because everything is written inside of my Skills).

Especially when a model or coding plan gets progressively worse or more expensive.

The AI cost war will only get worse, and the only way to escape is by building a Portable AI System:

One that lets you automate the tasks you hate with a system you truly own.

So focus on building out your context and your Skills, while worrying less about what models you are using.

The answer to ‘what is the best model’ is ‘all of them’. Give them the right context and workflows, and they’ll execute tasks in the exact way you want them to be done.

Learn more about how to build out good Skills here.

If you want more guidance on building out your own Portable AI System, Signal Starter is a live session where I’ll be going through the entire process step by step.

In 90 minutes, you’ll:

1. Identify one painful workflow you actually want to automate (not what someone else tells you to)

2. Build the Portable AI System from scratch with the 8 components

3. Turn that task into a working Skill you can invoke anytime

No coding experience is required, and I’ll show you how to build your system your way and automate all the tasks that you hate doing.

The session will be live on 29 May and includes the full recording if you can’t make it:

I want in

The AI cost war is just getting started, and inference will no longer be as cheap as it currently is right now.

I’m facing this constant struggle of managing my usage very carefully to avoid being rate-limited, and it has been a pain (and somewhat stressful experience).

Those who constantly hit rate limits will be those who are in a constant loop of asking AI a question and it executes, which burns too many tokens.

Meanwhile, the ones who survive are those who know how to spend their tokens wisely. Instead of prompting simply every time, they build out repeatable workflows that have clear instructions for any model to execute.

With models now being a commodity (so many are good now), the only key differentiating factor becomes the workflow and instructions that you give AI to execute.

And today, I’ll be showing you that key component you need in your system to escape this cost war:

Models will only get more expensive from here

Vibe coding used to be fun. AI used to be accessible to anyone who subscribed to a Coding Plan.

We saw posts about how someone can vibe code a cool tool that helps them with their workflows, and I remembered no one complaining about the rate limits.

A few months back, I knew nothing about this term. I never worried about them because it seemed like I would never hit them with my paid subscriptions.

I just kept prompting simple requests and asking AI to make simple changes to my vibe-coded projects (which was the completely wrong way to do it).

But it’s not the same now. If I were to do the same actions, I’d likely hit my rate limits with just a few requests (when I could have made the edits directly via a code editor).

Models used to be cheap because model providers wanted to acquire more users. They cut costs and give usage limit promotions to lock you in with their annual plans. Most are not profitable, and that has to change.

Right now, most of our costs are still being subsidised by the VCs. But it’s only a matter of time before model providers are forced to make profits (they’d want a return on their investment).

And that’s when AI will no longer be for everyone. We start to face the real costs we incur when using AI, which are exponentially higher than the costs we have right now.

A simple chat with the AI will use 20% of your weekly limit. Soon, AI will only be accessible to those who can afford the higher-tier subscriptions.

While everyone else is left with the scraps of poor rate limits or low-tier models.

Rate limits get hit more easily, and providers decide to cancel third-party/agent integrations when it becomes too expensive to run them with a coding plan.

The $10/month Coding Plans will be pitiful, and we barely get any usage out of them. Or, they could just be like Z.AI and increase their subscription plan costs without informing anyone.

And those who use AI as a simple chatbot without building anything repeatable will be stuck in this cost war.

Most people are too focused on the models

Every now and then, a new model will launch with all the hype and engagement farming on the timeline, just like Opus 4.7.

Most talked about the benchmarks that they exceeded every other model.

But does that matter to you?

The ‘best’ model may not always perform the task in the way that you like or expect. In other cases, a ‘lower-tier’ model could give you the same output at a lower cost.

The ‘best’ model today won’t be the same tomorrow, since everything changes so quickly in AI.

Maybe it’s because I’m using AI differently from what others are used to.

I’ve stopped vibe-coding extensively, unless there’s something that I’d believe is truly useful to myself (like FreeTierModels).

Instead, I’m using AI to automate tasks that I hate doing myself. The soul-sucking, repetitive actions that I do every day can be offloaded to AI.

So instead of stressing over what model you should use, I’m focusing on building this instead:

You need to build a Portable AI System

I’ve been obsessed with building a system that lets me easily switch between any model or subscription, while still getting good outputs.

Everything needs to be modular and flexible enough so all of my data remains truly mine, while the model is just the executor that automates the workflow for you.

These are the 8 PORTABLE components of my system:

• Personalised context: A file system that includes your context and tasks so any model knows who you are, what you do, and how you think

• Owning your keys: A BYOK provider that lets you easily switch between models, even within the same chat window

• Requests: Your input (voice or text) that gives commands so the AI runs tasks for you

• Thought processor: The LLM acts as the brain that receives an input and provides an output

• Actionable workflows: Skills that are repeatable and can be invoked by any LLM

• Building capabilities: Tools like MCPs or APIs that increase the capabilities of your system to external platforms

• Link across devices: Syncing across multiple devices

• Edit and manage: Using a markdown editor to write anything in your system

Any of the external components (like the models) can be switched out and the system still works.

We are not locked in with one provider, while still having access to all of our data (instead of having to export it).

And the only way for the entire system to work with any model is through repeatable workflows:

Skills let you Build Once, Switch Anytime

Systems thinking is something that I didn’t care much about. But the more I looked at how any business runs, everything has an SOP.

Someone like McDonald’s, which I quote this idea that Ali Abdaal makes in his videos:

McDonald’s has built systems so easy and repeatable that they can replicate across any outlet.

Any employee can be hired, and they can immediately start contributing to the system because of the clear workflows and SOPs that have been built.

And it’s not just McDonald’s, any food outlet is the same:

• Someone takes orders

• Orders are passed to the kitchen

• The kitchen prepares the food and passes to the server

• The server gives the food to the customer

Everyone has well-defined roles and workflows with clear instructions on how to execute the tasks they are expected to.

We can build our AI system in the same way.

Our employees (models) can be switched at any time, while we get the same output because we gave them clear instructions on how to execute the task, regardless of the model we choose.

If you’d heard of Claude Skills before, these are repeatable SOPs written in .md files that tell Claude how to execute anything to get an output that you want.

In my current AI system, I’ve added 190 different Agent Skills (it’s not just limited to Claude) to automate all of the tasks in my daily life.

Some are ones that I built, others I’ve downloaded or adapted from skill repositories online, and these let me carry out certain actions like:

• Converting my long-form article into an HTML file that can be easily repurposed into Substack, Medium, and LinkedIn

• Fetching an onchain summary from DeBank from the wallets that I track

• Syncing between my lead magnets and email platform

• Generating colour palettes for different sites and my own brand

• Adding YouTube transcripts and summaries to my knowledge base

• Converting different files into markdown for easy reading by AI

• Full workflows that keep track of all of my articles and YouTube video scripts

These are just some of my examples, but there are many others that you can add to your own workflows.

But I’m not going to bore you with the technicalities of how a Skill works. What’s more important is using the right steps to build out a Skill that you can use with any model:

The framework to build effective Skills

You may be wondering, why even bother with learning how to build Skills?

There are so many repositories out there with Skills for almost any use case.

But what works for someone else may not necessarily work for you, because everyone’s workflows will be different.

And there’s a risk of prompt injection attacks when installing someone else’s public Skill, so I’d adapt my workflow from them but never copy it directly.

You have a unique way of carrying out a certain task to get to an output, so there’s no point in copying someone else’s workflow.

While it seems daunting to write out a SKILL.md file, it’s not that hard. You just need to run the entire workflow together with AI, and work on solving the problem together.

There’s no need to write out the entire SOP (which I hate doing). Once you get a ‘gold standard’ run that gets the output that you want, the LLM will write out the full spec while taking note of what they should avoid (through bad runs).

So there’s no coding involved either.

I’ve developed a SIGNAL framework that I use whenever building a new Skill, and here’s how you can use it too:

(I’ll be using examples from a recent Skill that I built with this framework):

• Spot the bottleneck

Instead of building out your entire system from day one, let’s start small and find one annoying problem to solve first.

Throughout your entire life, there are many tasks that are extremely repetitive and soul-sucking, and you just wish there were a way to complete them much faster.

This is exactly how AI can help you automate your workflow: you only need to give AI a command, and it executes the entire workflow for you.

First, spend some time trying to identify that annoying problem you have. What is something you wish you didn’t need to do every single day?

It doesn’t matter how small it is, it just needs to be something that you hate doing.

That gives you more motivation to find a way to solve that bottleneck.

It can be as simple as connecting a new provider to OpenCode. I didn’t want to constantly:

• Go back to their documentation

• Find the right page and give it to the AI to execute it

So I built out a Skill to lay out exactly how I want this action to be done (just like all of my other OpenCode skills).

If it’s something that you do repeatedly, it can be converted into a Skill.

And this saves so much time because you can just invoke it with a slash command.

Frame your problem by describing exactly what you want to achieve, and what bottleneck you face.

The more details you give the AI, the better it can understand your problem and suggest a solution.

I’d suggest using a voice transcription tool (I’m using Letterly) to describe your problem via speaking instead of typing.

Here’s a recent example of mine when building out a Skill:

• Integrate the stack

After telling AI what the problem is, and if you give enough details to it, the AI should come up with a comprehensive plan on how to automate that workflow for you.

Some skills don’t require complex tools. They can just exist as a simple SKILL.md file with written text on what to execute.

Others will require connecting to APIs or MCPs (if you’re using external platforms), or Python scripts to execute tasks (like getting all of the recent tweets I posted via Typefully).

Depending on what platform you use, the LLM will recommend a way to connect with them so you can carry out the task.



That’s why Skills are more versatile than prompts that just give you text-based outputs.

• Guide the LLM

I saw a recent Reddit post that complained about how confused an intern was when executing tasks that the OP instructed him to do.

A similar concept can be applied to AI:

They are just an execution tool, and if you don’t define your instructions clearly:

They’ll come up with extremely poor outputs.

Ones that you don’t expect because you didn’t guide them enough.

So we have to walk through the entire workflow with AI step by step. We can’t expect it to know what we want, especially if we don’t articulate it clearly enough.

That’s why after framing the problem, I’d always want to run the entire workflow once to see how the AI understands and executes the task.

It will never be perfect during the first pass, and the only way to build the Skill is through failing repeatedly:

• Nail the standard

I can guarantee that the AI will make mistakes when running through the workflow.

I’ve experienced it so many times, but it’s actually good that they make the errors now.

By telling it what went wrong and what it shouldn’t do, it has more context on the wrong executions and what to avoid.

That’s why I like failures at the start:

We get to iterate on the workflow together with AI and get it to clarify anything that they’re still unsure (which could be something we didn’t think of too).

Constant iteration is the only way to build a good Skill. It’s tedious, but once you get the ‘gold standard’ that gives you the output you want:

It’s finally time to turn it into a Skill.

• Automate the logic

This is probably the easiest step, after getting the output that you want.

Claude has a skill-creator Skill that outlines exactly how a Skill should be created and how to reduce the token burn through progressive disclosure.

There’s no need to understand how it all works.

Once you get a successful run, tell the AI to convert the entire session into a Skill, including all of the mistakes made and bad runs with the skill-creator Skill.

While this Skill is focused on creating Skills inside the .claude folder, it’s also applicable to any Agent Skill. All of my Skills are in the .agents folder to be invokable by OpenCode, and they were all created by the same skill-creator Skill.

It has rich context on what worked and what didn’t, so the Skill becomes more personalised and specific with a higher success rate.

I’d also be more conscious about naming my Skills. When I want to search for them inside my .agents folder, it’s easier to find them if I organise by their domains, including:

• core-*

• learnings-*

• article-*

This is entirely optional, and you can choose to name your skills however you like.

But that’s not the end of the entire Skill creation process:

• Loop recursively

Whenever we run a Skill, there’s likely a new problem or output expectation that we have.

Sometimes it could be as simple as changing the file directory.

Skills are not static, and they will constantly evolve.

So whenever there’s something wrong or new when you execute them, just ask the AI to update the Skill to fit the new workflow.

Consistently iterate on the Skill until it gets the outputs that you want.

Exit the AI cost war with Skills

Skills let you be truly portable with your AI system:

They are clear SOPs and instructions given to any model that can execute them and get similar results.

And you won’t get locked in with any provider because you built the Skill once that can be repeated at any time with any model.

Of course, that’s not always the case, especially when asking the LLM to write good content (that doesn’t sound like AI slop).

The more expensive models may produce better outputs, but for other workflows that have been converted into reusable scripts, AI is just there to execute them.

So it’s possible to use a cheap model like GLM 4.7-Flash, and you’d still get the same results.

To start building out your own Skills, I’ve included both my SIGNAL Skill framework and the skill-creator Skill inside this folder.

If you want more guidance on building out your own Portable AI System, Signal Starter is a live session where I’ll be going through the entire process step by step.

In 90 minutes, you’ll:

1. Identify one painful workflow you actually want to automate (not what someone else tells you to)

2. Build the Portable AI System from scratch with the 8 components

3. Turn that task into a working Skill you can invoke anytime

No coding experience is required, and I’ll show you how to build your system your way and automate all the tasks that you hate doing.

The session will be live on 29 May and includes the full recording if you can’t make it:

I want in

I’ve been a heavy user of this free tier ever since I found out about this ‘good deal’, and it’s starting to get worse with the lower rate limits and greater congestion.

So before you use it, here are some key considerations and caveats about OpenRouter’s free models from my experience:

The ‘completely free’ tier is extremely limited

If you want to use OpenRouter completely free, you’ll be severely disappointed with how quickly the requests get used up every day.

The ‘truly free’ tier is limited to only 50 requests per day (different from token-based usage).

Depending on how complex your request is, it is possible to use multiple requests within the same prompt.

So if it’s very complex, you could reach this limit quickly.

To get more free requests per day, I would recommend topping up ≥ $10.50 to your account.

This now gives you 1,000 free requests per day, which is much more generous.

Right now, I’m comfortably below the rate limits every day, where I’ll send at most 500 requests a day.

Though I don’t use OpenRouter’s free tier for all of my tasks with AI, so it doesn’t paint the full picture.

While 10 credits cost you 10,OpenRouterchargesa50.80 minimum fee if you top up via a credit card).

The fees are definitely much more expensive than other model providers, and I’m just focused on making this one-time top-up so I get access to the higher rate limits.

But I’ve already spent $1.50 with this mistake:

Your credits may get used unknowingly

I added my OpenRouter API key to my Hermes agent, with the main aim of using the free tiers for simple tasks.

But what I found was that my API key was being used for the paid version of Gemini 3 Flash, and I spent $1.50 without intending to.

Though the good news is that you don’t need to maintain 10 credits in your account. So long as you have topped up 10 credits, you’ll gain access to the 1,000 requests per day.

If you’re just intending to use the free tiers, set a usage limit of 0 and you won’t face the same problem as me.

I likely incurred paid model usage on OpenRouter because of this:

You get rate-limited when servers are overloaded

Hermes may have fallbacks when the free tiers are overloaded, and that’s why Gemini 3 Flash was used instead.

Since we’re on the free tier, our requests will be the lowest priority.

There will be certain timings when the demand for compute is extremely high, and our requests will take much longer to be completed.

You’ll likely face rate limit errors like these.

Sometimes, when the model performs multiple steps before giving an output, you could be rate-limited during the entire process.

So I’m only using this free tier to execute tasks that I don’t require the result to be instant.

The tasks that require more thinking will be outsourced to my paid subscriptions instead.

Some of the models won’t be rate-limited as much, but that comes at a price:

Your data can (and will) be used for training

At the time of writing, NVIDIA’s Nemotron 3 Super is the most popular free model on OpenRouter.

I’ve used it before and it’s rather stable, but it comes with a catch:

NVIDIA will use whatever data you send to the free version to train its models.

If you’re conscious about privacy, this is a key consideration before you select the Nemotron family of models.

Other free models do not have the same disclaimer, but it’s possible that they may use your data too (there’s no such thing as a free lunch).

I’m fine with this disclaimer since I’d outsource more of the monotonous tasks to Nemotron instead of ones that heavily involve my personal data.

Models will change frequently

I wouldn’t get too reliant or comfortable with any of the free models.

Without warning, they could remove any model from the free tier, just like what they did for Qwen 3.6 Plus.

When a new model launches, they’ll give it away for free on OpenRouter as a marketing stunt. For Qwen 3.6, it was only available for a week.

I was a heavy user of the Step 3.5 Flash free tier, because it was fast and could execute most tasks.

But a few days ago, it was removed from the free tier with no announcements at all.

There will be some, like Arcee, that disclose when they’ll be removed, but I wouldn’t count on it.

After the Step incident, I’m showing less loyalty to any models on the free tier. They will come and go, and I’ll just enjoy the free inference while it lasts.

It’s getting worse, but OpenRouter is still worth $10.50 (for now)

Since I started using the OpenRouter free tier a month ago, the quality has dropped significantly:

• Fewer free models to choose from

• Servers are getting overloaded constantly

I see the OpenRouter free tier as a backup plan, and not my main model provider.

The free inference is great, but it won’t last forever. Sooner or later, OpenRouter will nerf this tier further once the costs of giving away free tiers are no longer profitable.

So I’m only using it for executing tasks where I don’t need the outputs immediately and don’t require thinking capabilities.

Tasks like updating my calendar or fetching my latest Twitter posts from Typefully.

That way, I’m able to save those inference costs on my main subscription plans (I’m currently using ZAI and OpenCode Go), so I don’t hit the rate limits that quickly.

As a one-time payment of $10.50, I still think it’s worth it for now, so long as you build up good skills that give clear instructions to the LLM on:

• What the desired outcome is

• What are the steps required to reach that desired outcome

To learn how to set up OpenRouter’s free tier with your AI system (via an API key), the full steps can be found in my YouTube guide.

That’s why I’m focused on building a Portable AI System where I can switch models at any time, and still get the same quality of outputs.

If you want more guidance on building out your own Portable AI System, Signal Starter is a live session where I’ll be going through the entire process step by step.

In 90 minutes, you’ll:

1. Identify one painful workflow you actually want to automate (not what someone else tells you to)

2. Build the Portable AI System from scratch with the 8 components

3. Turn that task into a working Skill you can invoke anytime

No coding experience is required, and I’ll show you how to build your system your way and automate all the tasks that you hate doing.

The session will be live on 29 May and includes the full recording if you can’t make it:

I want in

I’m lucky that I didn’t get stuck in one. But with the recent Claude policy changes for OpenClaw, many on Reddit and Twitter are trapped.

Anything can happen in 1 year, and you’re left frustrated with lower rate limits, limited integrations, or a plan that’s completely unusable.

So to avoid getting locked into any provider, I made 2 key rules for my AI system:

• Only buy monthly plans

• Make my system completely portable with different providers

Here’s the exact setup I’m using to switch providers without losing output quality:

You’re paying more than what you save

Most model providers want to lock you into annual plans. They dangle the perception of saving 20% or 2 months from your plan, which makes it seem like a good deal.

While deciding which AI plan to go with, I was tempted to go with annual plans for these extra cost savings.

But a Reddit comment made me change my thinking completely:

Once you lock in with them, you’re essentially trapped. So go for the monthly plan instead.

Anything can happen during this 1 year, especially when things change so quickly in this space.

Will Claude or ChatGPT still be the best models in 2027? Maybe, but it’s hard to say.

Other providers are getting better each day. Every new model claims they are smaller than Opus, but with almost the same capabilities.

Open-source models are competing with frontier models, and they can significantly reduce your costs (if you already have the right hardware).

While you save money with the annual plan, you’re incurring costs in other ways.

Right now, all of our coding plans are cheaper than direct API calls.

OpenAI did amazing PR by resetting their rate limits twice over the past 2 weeks.

Why would they do that when it incurs more costs for them?

We’re at the stage of hyper-competition in the AI space, as every model provider rushes to acquire new users:

• Free tiers with generous rate limits

• Heavy discounts on their annual plans

• Attractive pricing for new business seats

All of these marketing tricks are aimed at building loyalty, so you stay inside their ecosystem.

But once the cost of acquiring a new customer no longer makes sense, things will change for the worse.

Most of our token usage is subsidised by VCs.

OpenAI just announced its latest funding round of $122 billion, which is just insane.

It’s all the same for tech companies, just like how Uber underpriced their rates to get more users.

But eventually, the funding will dry up.

The costs of large-scale inference will increase to meet the demand for more users. And VCs aren’t a charity, they expect these AI companies that they’ve so heavily invested in to start turning in profits.

Once that happens, you get trapped.

The rate limits decrease rapidly. Typing in one simple request can cause a spike in your usage.

Companies start removing integrations because their infrastructure can’t handle the sheer amount of requests.

We’re at the tail end of the golden era for AI.

I have many regrets about not starting earlier before Claude Code blew up, as that would have given me more freedom to play around with AI.

The free inference, generous rate limits, and any other perks will soon be a thing of the past.

And those who are locked in are the ones who are trapped the most:

You need a Portable AI System

The best model that works for you today, won’t be the best one that works for you next month.

And when you’re locked in with one model, it’s so hard to make the switch a few months later.

That’s why I’ve been so obsessed with building a Portable AI System for my workflows, which needs to meet this criteria:

• I have full control over my data and workflows (it’s not locked into one provider)

• I can easily switch between any model (my system can connect to almost any model provider)

• I still get a similar quality of outputs, but at a lower cost

The key characteristic of this system is adaptability. I’m only locked into a plan for 1 month, so I can try it and switch to another one (if it doesn’t meet my needs).

My data, my context, and my skills are all fully owned by me. And because I build highly specific workflows (Skills) through iteration, I can get almost any model to execute the same workflow with similar results.

Inside this system, there are 5 key components required for it to run successfully:

• A file system that includes my context and Skills (in the .agents or .claude folder), all in the .md format

• A monthly LLM coding plan or API (this will change frequently)

• A Bring Your Own Key (BYOK) interface that lets the LLM access my files

• An editor platform to do all of my writing

• A syncing platform so all of my work is accessible across any device

I’ve been trying multiple setups, and I’ve finally settled on a Claude Code-Obsidian hybrid system:

• A simple folder on my Mac desktop for my context and Skills

• OpenRouter for free tier models (1,000 requests/day after a $10.50 deposit) and other coding plans as a backup (I’m currently using @Zai_org’s GLM)

• @opencode to connect to these models (it’s available on both Desktop and Terminal)

• Obsidian for editing

• GitHub for syncing between my desktop and laptop

I’ve moved on from the OpenCode Desktop app (because it became too laggy with the number of chats that I have).

Right now, I’m using the Terminal because I can run multiple windows at the same time, each for one specific task I want to complete.

Though it’s messier than I would have liked, and I should have stuck to at most 2 windows for my setup.

You may be wondering, why is there no OpenClaw or Hermes at all?

There’s too much noise right now with everyone bragging about the complicated workflows on Twitter, making you feel so inferior.

Instead of focusing on what to run with these agents, build out high-quality workflows first.

Once you get the outputs that you want and like, then it’s time to go for ‘always-on’ automation.

Not everything you build with AI has to be impressive or make you 7 figures.

All you need is getting AI to solve one painful problem in your life, and that’s enough to change the way you see AI.

Don’t get trapped by convenience and take action instead

You don’t have to get locked into one ecosystem just because they offer you a seemingly good deal.

Yes, it’s so tempting to just stay with a Claude or ChatGPT plan. It feels so convenient and easy. But that only benefits them, and not you.

Anything can happen inside the full year that you lock in.

So instead, you need something portable enough so you can switch out models to whatever you like and still get the same outputs.

If you’re done being locked into one provider, start with this free setup video guide.

If you want more guidance on building out your own Portable AI System, Signal Starter is a live session where I’ll be going through the entire process step by step.

In 90 minutes, you’ll:

1. Identify one painful workflow you actually want to automate (not what someone else tells you to)

2. Build the Portable AI System from scratch with the 8 components

3. Turn that task into a working Skill you can invoke anytime

No coding experience is required, and I’ll show you how to build your system your way and automate all the tasks that you hate doing.

The session will be live on 29 May and includes the full recording if you can’t make it:

I want in

Everyone on the timeline makes you feel inadequate by flexing the crazy workflows that earned them 6 figures with agents.

But for most people, this is too complicated.

The fastest way to get good with AI is to start with a free setup that helps you understand how to integrate AI into your workflow.

AI Twitter is showing you how amazing the latest models are and the high-quality outputs that it produces.

Instead of being overwhelmed by fear and analysis paralysis, it’s time to start taking action.

I’ve been searching for the easiest and simplest setup that anyone, even complete beginners with no coding experience, can build and start using AI more effectively.

And I finally found the perfect combination that achieves this:

The Claude Code + Obsidian system, but with any model provider

Before I built out my Hermes and OpenClaw agents, I started out with a Claude Code + Obsidian setup inspired by @ttorres and @carlvellotti’s YouTube videos.

I call it my Signal OS, a personal operating system that compounds my social and onchain workflows while saving me time on soul-sucking tasks.

Back then, I had the misconception that Claude Code was only for coding.

But I finally understood that it can be used for almost anything, once you create a repeatable workflow that the LLM can execute (or Skills).

I enjoyed this setup because I could create Agent Skills to perform tasks for me with Claude Code, while I do all of my writing in Obsidian.

This system knows who I am, what I do, and how I think because of all the context within markdown files that the LLM can access.

Which lets the LLM give more personalised outputs, rather than an unreliable memory system that is found in the chatbox.

Even then, Claude Code may not be for everyone.

The subscription’s pricing can be expensive to some, especially when they change their rate limits at any time.

There are other LLMs that provide comparable quality of outputs while at a much cheaper price with higher rate limits, and I’ve been exploring the most cost-effective setup for my own needs.

That’s when I realised that I can’t rely on Claude Code alone:

You need a Portable AI System

You don’t want to be locked into any ecosystem, not even Anthropic’s.

Imagine some other company releases a new model that outperforms Opus/Sonnet/Haiku, and you can’t bear to switch over because you bought an annual plan.

Things move too fast in this space, and our operating system should not be stuck with one provider.

We need to freely switch between models, depending on what best fits our needs, while getting the same high-quality outputs that we expect.

This all boils down to creating high-quality Skills that are clear, repeatable workflows that any LLM can pick up and execute right away.

I’ve tried switching between models inside the Claude Code extension as I felt that it’s the most beginner-friendly way for anyone to get started.

While it had some success when I switched over to my GLM-4.7 subscription, I faced errors like this when using OpenRouter’s API key.

So I couldn’t recommend this system yet because it wasn’t ‘open enough’, until I finally found the best setup:

The OpenCode-OpenRouter combination

One of the biggest gripes I had with OpenClaw and Hermes was when I needed to switch models.

When my subscription hit the rate limit for the week, I had to change the model provider within the .json file.

But this led to (a lot) of troubleshooting and testing to ensure that the connection works. It usually doesn’t on the first try, and I burned lots of tokens just trying to get everything working.

I tried switching models in Claude Code and Codex to use another Anthropic/OpenAI-compatible endpoint, and these led to many errors too.

So after playing with more CLIs and applications, these are the only 2 platforms that I’d recommend for complete beginners who want to start for free:

OpenCode to access any API

I like OpenCode because it lets me connect to any provider that I currently have a subscription with, including:

• ZAI Coding Plan Lite

• Alibaba Coding Plan Lite

• @MiniMax_AI API

• A bunch of free-tier subscriptions

For OpenClaw and Hermes, I had to verify the Base URL (it’ll be different for the coding and API plans), and I wasted a lot of time troubleshooting to ensure that everything works well.

While for OpenCode, most of the popular model plans (both from the providers themselves or aggregators) are already preconfigured, and we just need to add the API key inside.

I’ve been using the CLI first, until I realised that OpenCode has a desktop app too, which makes it even more accessible to beginners who don’t want to use the Terminal.

I can switch between models that I’ve configured for each subscription, while each individual chat allows me to select a specific model (so I can route requests for each specific task).

In my case, I’m using MiniMax-M2.7 to build out the article outlines (like this one) with me as it requires more thinking capabilities, while I use Step 3.5 Flash to review my article with a checklist (after writing everything out).

It gets even better when you pair it with a platform that offers free inference:

OpenRouter to access free models

I chose OpenRouter as the main choice to route my requests to free models because:

• It has > 28 models in their free list (though I face issues when connecting to some)

• The setup is straightforward when generating an API key

• We get free 50 requests a day

OpenRouter even has Tier 2 models (a similar tier as Claude Sonnet) that you can access for free, including Nemotron Super.

Though you’ll be sharing data with them if you choose the free tier.

While 50 requests per day is a great way for any beginner to start building out their Signal OS, we can bump it up to 1,000 requests a day with a one-time deposit of 10.50(asa10 credit top-up + 5% fees).

I’ve been using the free tier extensively to carry out most of my execution tasks, and I haven’t hit 1,000 requests a day (yet).

So I believe that this is generous enough to try out this new way of building out your personal operating system.

But of course, these models are not good enough to code out anything. I would still use a paid subscription if I want to vibe code or create Skills, while I use the free models to execute tasks that I’ve already created comprehensive workflows for.

I shared more on how to get access to these free models here.

You just need a system that works for you

The Simplest AI Setup lets you experiment with AI in a smarter way through better context engineering.

Most memory systems in the chatboxes have a limit to how much they know about you, until they start replacing old memories with new ones.

And they will forget some of the tasks that you asked the LLM to do before.

But if you design your system in the right way, the LLM will never forget who you are, what you do, and how you think.

All it takes is building out a markdown-based system that routes the right context to the right request.

So if you want to use AI the right way instead of a generic chatbox that forgets everything, get the full (free) video guide to build your Simplest AI Setup.

I will guide you through the entire installation process and model setup for both OpenCode and OpenRouter, while giving you the full prompt to build your Signal OS:

Get the free guide

Anyone who can access your file will have full control over your agentic wallet.

So what’s the best way to give a wallet to your agent?

This stumped me for weeks as I tried and failed to find the right wallet that worked for me.

There were so many options, yet some:

• weren’t secure enough

• locked into certain ecosystems

• did not have enough network support

But I think I’ve finally found the answer (though it’s still not the perfect solution yet).

Many agentic wallets just didn’t work

These were all of the options that I considered:

• Storing a private key in plaintext

While it’s convenient to store a private key in plaintext, it’s the easiest way to get hacked.

Anyone who accesses your computer or VPS will have full access to your private key and the funds in your agent’s wallet.

A private key is worse than an API key because while those can be rotated at any time, a private key is lost forever once it gets exposed.

@PatrickAlphaC has been tweeting against storing private keys in plaintext and it’s way too risky to leave your private keys exposed in this way.

• Storing a private key in Foundry

Patrick recommended to uss Foundry to generate and encrypt a private key, and I tried doing that on my VPS.

But then, I realised how this is not feasible after chatting with my agent:

There still needs to be a manual step for me to decrypt the transactions, so the agent still needs access to the private key for it to run autonomously.

It suggested me to store the private key in plaintext, which is not what I want.

I tried asking Grok for other alternatives, and it suggested using Account Abstraction which seemed to complex for me.

And I’ll likely burn through tons of tokens just to troubleshoot and get everything right.

But maybe I’m doing it wrong, and there’s a way to use Foundry with agents where they can sign autonomously without my approval.

• Agentic Wallets

While I was searching for an Agentic Wallet, I found a few options:

1. Coinbase

2. Privy

3. OKX

@wallet just launched their Agentic Wallet a few days ago, and I found it most ideal since it supports multiple chains.

While Coinbase launched theirs a while back, I don’t have enough details on what networks they support (though it seems like it’s just Base).

and when I wanted to send funds from my Coinbase Agentic Wallet, it was only possible for USDC.

For other tokens, I had to use the developer portal which defeats the purpose of an agentic wallet.

While the OKX Agentic Wallet supports more chains and functions, I couldn’t get it to work with other platforms like @lifiprotocol.

Doing something as simple as bridging ETH from Base to Ethereum was such a pain because there was no route with the OKX skills, and LiFi couldn’t work.

But this wallet seems to be the best fit for my needs right now:

OWS wallets are the most secure for now

@OpenWallet’s Open Wallet Standard (OWS) appealed to me because it:

• doesn’t leave private keys in plaintext

• acts like a full private key except that it’s encrypted

• supports all EVM chains and other L1s (Solana etc.)

I tried generating a wallet on my Hetzner-Hermes setup, and it didn’t work at first.

So I just asked Hermes how to do it, and it helped to create my wallet from there.

It’s possible to do it manually, you could throw the error into ChatGPT and tell it that the binary isn’t in your path.

The CLI generated all of the wallet addresses for me, while I didn’t have access to the private keys.

But what I realised was how easy it is to export out the seed phrase, once someone has access to my VPS.

We just needed the wallet export command, and the entire 12-word phrase gets exposed immediately.

Don’t worry, I’ve deleted that wallet already.

I did see the docs mention adding a passphrase, but I wasn’t prompted for it during the wallet creation process (and that’s something I’d like to clarify with the @moonpay team as well).

I did try encrypting the wallet and generating our passphrase with Claude,

but that failed and I couldn’t decrypt the wallet even after typing the right passphrase.

For now, the biggest risk is someone accessing my VPS and typing out the export command to access my seed phrase.

So it’s ‘slightly’ more secure than storing my private key in plaintext, as the hacker needs to know the right command to export it out.

Adding policies to my wallet

Instead of giving my wallet full access via the private key, I could give it an allow-all policy with a known expiry date so that I can easily rotate that key if it gets exposed.

Claude (but actually, GLM-4.7) guided me on the commands to create the policy and how to give access to my Hermes agent.

I tried asking my Hermes Discord bot to send a transaction, and it was able to do it via the API key.

I tried swapping and bridging from LiFi, and both work (after installing the SDK, which you can give your agent all the instructions from here).

Though LiFi charges a 0.25% fee for every swap we make on the SDK platform, it’s still the best one right now for cross-chain swaps across different chains.

So, what’s next?

Now that I finally have the flexibility of signing different transactions with a private key (unlike the OKX Agentic Wallet), this means that my agent can perform almost any other transactions just like I can.

I have many plans on how to trade or implement DeFi strategies with my agent:

• A recursive bot for @boros_fi (since I still don’t understand how funding works)

• A bot to find the best (and relatively safe) DeFi yields

• A trading bot on both perps and normal onchain swaps

I’ve made a horrible error by giving my agent too much autonomy at the start, and it made a 400+tradeformewhenIjustwantedittouse20 per trade.

So it’s a painful lesson for me that I should build out a proper skill for my entire trading strategy, and tell the agent to execute it exactly as it is (instead of giving it vague instructions and hoping that it’ll understand).

There are many more agentic MCPs, skills, and CLIs that I can play with using this wallet, and I plan to add some of them via @yq_acc’s CryptoSkills site.

While I’m glad that the wallet works, I’m still concerned how there was no passphrase generation step when creating my wallet, so it’s still exposed somewhat.

Hopefully, I can gain clarity on how to set this passphrase soon (I’ve struggled to find that step in the docs, even with AI).

I’m currently running my OWS wallet with my Hermes agent (deployed on Hetzner), and I shared my setup here.

Start small by building your High-Signal Digital Brain

Maybe all of this still seems too complex for you, with so many foreign terms and especially having to use the CLI to create your wallet.

This is not the only way to use AI effectively.

Before agents and wallets, you can start smaller and see how AI automates the tasks you hate doing every single day because they’re so repetitive and boring.

You just need a system that works for you and your context, and here’s a full guide on how to build one:

I want in

You’re burning through your tokens if you’re using Claude Opus for everything.

If you don’t need high-quality outputs for a task (like scraping sites), then it makes no sense to keep using high-end models.

When you could just outsource them to cheaper models that do a decent enough job (and some are even free)?

So here’s how I’m routing these boring, repetitive tasks that my OpenClaw and Hermes agents do to free models (though I had to spend $10.50 to increase my rate limits).

Here’s how I route my models to these cheap (and free) ones:

OpenRouter’s free models

I found out about these free models through Nvidia’s tweet promoting their free Nemotron 3 Nano 30B version on OpenRouter.

Nano models are the cheaper and faster models that are great for repetitive work that don’t require much thinking.

I didn’t know that Nvidia had their own models until now, and a 30B version seems decent enough to do some of the simple, menial tasks that I plan on doing.

Many developers are using it for their OpenClaw agents, and I wanted to know how to do the same too.

Both agents allow me to add other API keys apart from my main one (I’m using GLM), so I have the flexibility of switching between model providers and APIs.

All you needed was to generate your OpenRouter API key, add it to OpenClaw or Hermes, and you could start getting free inference.

Or that was what I thought, when it was far from the actual case:

But these models are not truly free

Of course, there always has to be a catch with these free models. OpenRouter has 2 different caveats for its free models:

• OpenRouter will log your data

The Nemotron model page has this warning that all of your requests will be logged, which is likely why Nvidia is giving us free inference.

I’m not too worried about it as I plan to use it for scraping and other tasks that won’t involve sensitive data, but that’s something you may want to consider when using this model.

• OpenRouter has a 50-request limit per day for free models

I was trying out the Nemotron API on Claude Code by typing out random messages just to test its capabilities, before I hit this message.

I knew that it was not possible for Nemotron to give me unlimited inference, but it hit the 50-request limit a day very quickly after I asked it to generate a YouTube transcript.

That’s when I explored OpenRouter’s docs and realised there are 2 different limits:

• < 10 credits: 50 :free model requests per day

• ≥ 10 credits: 1,000 :free model requests per day

Instead of using token limits (like most model APIs), OpenRouter uses a request-based limit, which is likely based on the number of calls that you make.

In this case, the number of tokens that you burn doesn’t matter.

One request can burn 1 million tokens or just a few, and it ultimately depends on how you phrase your prompts.

I don’t think that 50 requests would be sufficient for the tasks that I want, so I topped up 10 credits ($10) to my OpenRouter account.

The fees are expensive at 5% per every top-up that you make, and I had to pay $10.50 in total.

It’s even worse for non-crypto payments with a 5.5% fee (minimum of $0.80), so I stuck with crypto payments via Coinbase Commerce instead.

I could only top up a whole number of credits e.g. 10 or 11, and not something like 10.20.

With a higher request limit per day, I’m now using these models in my agents and my Digital Brain setup:

How I’m using these models

Instead of making my default model for both OpenClaw and Hermes as the main one, I’m using it in subagent tasks instead.

My current agents would still use the main models that I pay for (GLM-4.7, MiniMax-M2.7 etc.), while I’ll instruct them to spawn subagents that use the free Nemotron model to carry out specific tasks.

Here’s a chat I had with my agent on GLM-4.7 to crawl a site, and it spawned subagents to perform this task.

You’ll need to specify this as part of the skill that you create to run this certain task, and let me know if you have any questions on how to do this.

I’m also using it as part of my Digital Brain, which is my Obsidian + Claude Code setup that I’m running by myself and not with agents.

While I’ll use Claude Code (with GLM-4.7) for my main work inside of VS Code, I’ll use the Codex CLI on the same Digital Brain repository to carry out the boring tasks with Nemotron and save my token burn.

Since I can’t add both my GLM and OpenRouter keys to my Claude Code configuration, so I have to use another CLI like Codex instead.

Something to note is how OpenRouter may route your requests to other models, like Gemini 3 Flash Preview, in my case.

I’ve burned over $1.37 over the past few days, even though I instructed it specifically to use Nemotron.

Token efficiency is so important now

Most of us are likely just burning through tokens when calling LLMs, mainly because we’re using the same model for every task.

So building a more efficient routing system between models (depending on the task complexity) becomes a key factor in reducing your token burn.

I’ll be sharing more of my tips on being efficient with tokens, so stay tuned for that.

You don’t need autonomous agents to use AI effectively

Maybe you read through this article and are completely overwhelmed about where to start.

There are so many steps to set it up properly, and many people will give up halfway.

I took a week just to get everything working, without building any useful Skills yet.

For most people, you don’t need a complex system right from the start.

Instead, all you need is to find one soul-sucking, repetitive task that you hate doing and outsource it to AI.

Start your journey by building a High-Signal Digital Brain and make AI work for you, instead of against you:

Start building with AI

OpenClaw is great if you’re technical, but the entire experience is not friendly for non-coders like me.

So when everyone mentioned how easy it is to create a Hermes agent, I decided to try it out.

Instead of migrating my entire OpenClaw setup, I decided to build a new agent from scratch.

This meant having to source another VPS to host my agent (since my WiFi is unstable), and I decided to go with Hetzner.

So here’s the full process of how I deployed my Hermes agent to a Hetzner VPS:

Hetzner’s prices will increase in April

Before we start, just something to take note that Hetzner will increase their plans, starting from April.

Most of the cost-optimised plans (the one that I chose) are affected, and it costs ~ $1.50 extra per month.

Hetzner still provides one of the cheapest plans for hosting a VPS, so it’s a good low-commitment way to try out an OpenClaw deployment.

Registering for an account

I followed Digital Benjamins’ video, which provided a great overview on installing OpenClaw through a VPS.

It included some good security tips that I didn’t use for my initial OpenClaw setup on Tencent Cloud.

But to create an account, I had to verify via KYC, so that’s something that may stop you from signing up.

Purchasing a plan

I chose the cost-optimised plan at 4GB RAM that costs me $4.46/month (which will increase in April).

I followed Benjamins’ guide to generate an SSH Key, which allows me to log into my VPS directly from my Mac’s terminal.

I got ChatGPT to help me with this portion to generate an SSH key on my desktop.

I used this command via the Terminal to create my key,

ssh-keygen -t ed25519 -C "your_email@example.com"

before using this guide that I created to connect it to Hetzner.

This created my server, and I had to copy the IP address (this is something you need to secure and not share it with anyone).

I went to my terminal and entered this command:

ssh root@<YOUR_IP>

After entering the passphrase (which I set inside Hetzner), I could access my VPS via Terminal in the root account.

This is not a good security practice since root access is very powerful (if it goes into the wrong hands), so I followed Benjamins’ advice and created a new user to install my Hermes Agent there instead.

You could follow the guide here on how to do something similar too.

Setting up Hermes on Hetzner

With this new user, I started the setup wizard for @NousResearch’s Hermes.

If you used OpenClaw before, the experience is somewhat similar.

I’m using my @Zai_org Coding Lite plan, where I generated an API key,

and added it to Hermes.

I tried to install Docker, but I wasn’t sure what I should be doing here, and I decided to skip this step.

I plan to ask my bot how to add the docker in future.

Since I’m already familiar with the Discord workflow after doing it multiple times with OpenClaw, I created a Discord bot instead of Telegram or Slack.

There are a few configurations, and the full guide can be found here.

I had to reload my bash first, as the hermes command didn’t work before that,

and I finally launched my Hermes agent with glm-4.7 as the LLM.

Troubleshooting my Discord bot

I tried posting a test message on Discord, which failed, but I liked how the hermes command brought up a chatbox that I can instantly talk to the LLM.

Instead of OpenClaw’s openclaw tui command.

I asked GLM to troubleshoot my configuration, and the test showed that it worked.

What I realised was that the Discord bot would only respond if I mentioned it on Discord. Leaving the message blank was not enough, and I had to write something like ‘hi’ before it finally worked.

Going back to Hermes’ docs, I realised it’s because the REQUIRE_MENTION variable is on by default across all channels, so I got my Discord bot to remove it entirely.

Troubleshooting my GLM connection

While I was installing @okx’s Agentic Wallet, I suddenly faced an insufficient balance error, even though I did not hit the rate limits for my Coding Lite plan yet.

This led to a frustrating 3 hours of troubleshooting everything,

and here’s what finally worked for me:

Instead of using the Z.AI option in the models page, I switched over to the custom endpoint.

I used the coding endpoint instead of the normal one, https://api.z.ai/api/coding/paas/v4, and it finally worked.

So here’s a troubleshooting tip if you’re facing the same problem as I did.

Why are you even using Hermes?

So what’s even the point of having a Hermes agent?

For me, I want to take advantage of its persistent memory that seems to build on OpenClaw’s MEMORY.md system, and create my own trading agent.

I was inspired by @RHLSTHRM’s post on building a 24/7 crosschain agent that:

• Gets market data from the CoinGecko API

• Swaps crosschain with @lifiprotocol

• Executes gasless transactions with Pimlico + EIP-7702

I’m still in the process of building this out, as I want to generate a private key securely first using @PatrickAlphaC’s recommendations instead of storing it in plaintext in my .env file.

There’s so much to learn about building secure agents, and I’ll share everything I learn in the next few articles.

Feel free to use my referral link for GLM-4.7 to get a $10/month plan with 10% off.

You don’t need autonomous agents to use AI effectively

Maybe you read through this article and are completely overwhelmed about where to start.

There are so many steps to set it up properly, and many people will give up halfway.

I took a week just to get everything working, without building any useful Skills yet.

For most people, you don’t need a complex system right from the start.

Instead, all you need is to find one soul-sucking, repetitive task that you hate doing and outsource it to AI.

Start your journey by building a High-Signal Digital Brain and make AI work for you, instead of against you:

Start building with AI

I get it.

Because for a long time, that was me too.

I saw how Claude Code exploded from nowhere to become the main vibe coding tool.

I read through the tweets and became more anxious about being left behind.

I watched people brag about their complex agentic workflows that ran their entire company for them, while I struggled to find ways of using it beyond a search engine.

The more I looked, the more overwhelmed I felt.

Everyone was telling me that I should be using AI.

But no one was showing me how to use it in my workflows.

So I did what most people do.

I tried hundreds of ‘steal my prompt’ posts and threw them into a ChatGPT chatbox.

I watched YouTube tutorials on MCPs, CLIs, and other complex terms.

But nothing seemed to click.

I kept getting generic outputs that I’d never use.

I had to explain my context over and over again to AI, which wasted more time.

I was spending more time teaching AI how to perform certain tasks than when I actually did it myself.

I kept thinking I was missing something.

Maybe I needed the right model.

Maybe I needed the right prompt.

Maybe I needed the right setup that some influencer on Twitter had already figured out.

But over time, I realised the real problem wasn’t that I lacked access to the right tools.

After locking in for 2 months (while I was on hospitalisation leave), I realised I was just using AI without a system.

And when you use AI without a system:

• You keep restarting from zero

• You keep re-explaining yourself

• You keep losing useful outputs in chat history

You’re stuck in this hamster wheel of trying to get AI to work for you, without doing any useful or meaningful work.

And this course is meant to help you change that.

You don’t need a complex AI setup

Most people think they need to use OpenClaw or Hermes to become a master of AI.

But many struggle (like I did) to get everything up and running.

As a complete beginner with zero coding experience, I spent one whole week just to get the basic functions to work.

Most people would give up because it’s so overwhelming, because there are too many things that could go wrong.

But what if there’s an easier way to get started with AI?

Most people don’t need a fully autonomous agent that runs 24/7 and carries out tasks for them.

But they have more than 1 painful bottleneck in their workflow that they hate doing.

Signal OS is meant to help anyone:

• Build a system that knows your context

• Route the right information to the right task

• Build out repeatable workflows that are triggered with one command

What is Signal OS?

Executing tasks with AI is exactly the same as how you’d carry out your daily tasks (like typing).

Every useful AI workflow needs:

• Intent: what you want to achieve

• Memory: the context, files, and experience the system can draw from

• Brain: the LLM that interprets the task and reasons through it

• Nervous System: the routing that connects the right context to the right task

• Hands: the skills that carry out the work

Signal OS is a digital replica of who you are, and what you do.

The more you build it up, the more it becomes a system that helps AI work with your context and adapt to your unique workflows.

What you will build inside this course

Inside Build Your Signal OS, you will build a complete system that is unique to your workflows and preferences, and continuously compounds with richer context (instead of having to prompt your LLM again and again).

You will learn how to:

• Find cost-effective LLMs other than Claude that have higher rate limits and lower costs

• Build a personal AI operating system with Claude Code, markdown files, and Obsidian

• Turn painful manual tasks into repeatable Claude skills

• Save useful AI outputs into an organised file system

• Route the right context to the right task to save on token costs

• Build both social and onchain workflows in one system

Even without OpenClaw or a complex setup, you can build a simple system with AI that carries out the tasks that you hate and saves you time.

I created this course for my past self, who was an AI skeptic and struggled to see how AI could work for me.

This was the course I wish I had, so I didn’t waste my time watching the wrong videos or following the wrong advice on AI.

This is the full blueprint on how I would let AI run my life without the complexities of a Terminal, or even an expensive Claude Pro plan.

If you want to stop feeling overwhelmed by AI and start building repeatable workflows that compound your social and onchain reputation, let’s get started.

Navigating the course

This course is a full newsletter with multiple sections. Depending on your device, here’s how you can get to your intended section:

• Desktop: A table of contents can be found on the left-hand side

• Mobile: No navigation, but your scrolling progress will be saved so you can go back to where you left off

Identifying one painful problem in your workflow

Read more

This is not a hack to start using Claude Code for free.

But I found a simple tweak that lets you get the same experience of using the Claude Code extension on VS Code, while using a cheaper model (some cost $10/month).

Most tutorials teach you how to add a cheaper LLM into the Terminal, but for non-coders like me, I still prefer using the VS Code extension.

So after spending days trying to find a good alternative:

Here’s how I’m using a $10/month LLM while getting the same Claude Code extension experience on VS Code:

Why even use the Claude Code extension?

Call me a creature of habit, but I enjoy the Claude Code experience more than any other LLM chatbox, primarily because I can type out my chats in the central pane.

Other chatboxes are only available as a side pane, which I don’t enjoy, while others focus on coding tasks instead of being a general-purpose one.

Ever since building out my High-Signal Digital Brain with Claude Code and Obsidian, I’m too used to the user flow, but I’ve realised how costly the Claude Pro plan can be.

Especially with the low rate limits that I’d frequently hit in the 5-hour window.

I don’t want to be locked into the Claude ecosystem, and I want to have the flexibility of switching between models and subscriptions.

So after finding this tip on using cheaper models with Claude Code, it finally gave me more clarity:

How you can use another LLM for Claude Code

Claude Code is actually compatible with other models.

So while you get the same Claude Code experience on the terminal or VS Code extension:

You’re using a cheaper LLM to run the same tasks with higher rate limits.

It’s all thanks to more models that are now compatible with Anthropic’s API, so you can call another model to run requests within Claude Code.

So we have to find these models that are cheaper and compatible with the Anthropic endpoint:

Finding the right LLM

For this to work fully, we need to find a model that:

• Is compatible with the Anthropic endpoint

• Offers API key generation with the paid plan

It’s also possible to use a main API endpoint, but I find that it burns through tokens very easily (so I prefer using a paid plan).

These are some options that I considered:

• @MiniMax_AI M2.5 Starter: $10/month

• @Kimi_Moonshot 2.5 Moderato: $19/month

• @Zai_org GLM-5/4.7 Lite Coding Plan: $10/month

As shared in my OpenClaw setup guide, I decided to go with Z.AI’s GLM Lite Coding plan because there’s no tax charges (like MiniMax), and I could use PayPal for my payments.

All of them work the same way:

• Sign up for their coding plans

• Generate an API key from the dashboard

• Add that API key to VS Code

In my case, I generated an API key on the Z.AI dashboard,

and used that API key to connect with Claude Code.

Most of the tutorials only share how to add their API key and use Claude Code in the terminal, and not the extension.

So here’s how I did it for my GLM-4.7 plan:

Adding the API to the VS Code Extension

I followed MiniMax’s guide to add GLM-4.7 as my preferred model on Claude Code.

While their guide is only for MiniMax models, we can expand it towards any Anthropic-compatible endpoint:

• Install the VS Code Extension

Go to the Extensions tab, search for Claude Code and install it

• Open Settings in the Extension

• Type out your model under ‘Selected Model’

• Click on ‘Edit as JSON’

• Edit the JSON file to configure your API endpoint

This is the most challenging part for non-coders like me.

You should already see these two variables inside of the JSON:

{
  "claudeCode.preferredLocation": "panel",
  "claudeCode.selectedModel": "glm-4.7"
}

So now, we need to add one more variable inside like this:

{
  "claudeCode.preferredLocation": "panel",
  "claudeCode.selectedModel": "glm-4.7",
  "claudeCode.environmentVariables": []
}

We’ll expand out the variables to include this inside:

"claudeCode.environmentVariables": [
        {
            "name": "ANTHROPIC_BASE_URL",
        "value": "<PROVIDER_BASE_URL>"
        },
        {
            "name": "ANTHROPIC_AUTH_TOKEN",
            "value": "<LLM_API_KEY>"
        },
        {
            "name": "API_TIMEOUT_MS",
            "value": "3000000"
        },
        {
            "name": "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC",
            "value": "1"
        },
        {
            "name": "ANTHROPIC_MODEL",
            "value": "<MAIN_MODEL_NAME>"
        },
        {
            "name": "ANTHROPIC_SMALL_FAST_MODEL",
            "value": "<FAST_MODEL_NAME>"
        },
        {
            "name": "ANTHROPIC_DEFAULT_SONNET_MODEL",
            "value": "<MAIN_MODEL_NAME>"
        },
        {
            "name": "ANTHROPIC_DEFAULT_OPUS_MODEL",
            "value": "<BEST_MODEL_NAME>"
        },
        {
            "name": "ANTHROPIC_DEFAULT_HAIKU_MODEL",
            "value": "<FAST_MODEL_NAME>"
        }
    ],

I’ve left them as environment variables, but you’ll need to make edits to them (and remove the <> brackets too):

• Provider Base URL (for my case, it’s https://api.z.ai/api/anthropic, but it will differ based on the provider you choose)

• LLM API Key: This is the key that you generated from your plan, keep it safe as anyone can access your plan once they get access to it

• Anthropic Model: This depends on what model you’re using (I’ll put it as glm-4.7)

If you’re familiar with the Anthropic model family, there are 3 types:

• Opus: Most powerful, but most expensive

• Sonnet: Great for most tasks, mid-tier in costs

• Haiku: Fast and cheap, but not as capable as the others

So it’s good if you want to customise your model routing so you get to use the cheaper models for lightweight tasks (and reduce your token burn):

• ANTHROPIC_DEFAULT_HAIKU_MODEL: Choose the cheapest model (e.g. GLM-4.7-Flash)

• ANTHROPIC_DEFAULT_SONNET_MODEL: Choose the mid-tier model (e.g. GLM-4.7)

• ANTHROPIC_DEFAULT_OPUS_MODEL: Choose the high-end model (e.g. GLM-5)

This depends on the model provider that you’re using (let me know in the comments if you need help). But if you prefer to just use one model across endpoints, it’s possible to do that too.

By setting everything as your main model (e.g. glm-4.7).

So this is how I’d fill up my JSON for my GLM-4.7 plan:

{
  "claudeCode.preferredLocation": "panel",
  "claudeCode.selectedModel": "glm-4.7",
  "claudeCode.environmentVariables": [
        {
            "name": "ANTHROPIC_BASE_URL",
        "value": "https://api.z.ai/api/anthropic"
        },
        {
            "name": "ANTHROPIC_AUTH_TOKEN",
            "value": "XXXXX1234567"
        },
        {
            "name": "API_TIMEOUT_MS",
            "value": "3000000"
        },
        {
            "name": "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC",
            "value": "1"
        },
        {
            "name": "ANTHROPIC_MODEL",
            "value": "glm-4.7"
        },
        {
            "name": "ANTHROPIC_SMALL_FAST_MODEL",
            "value": "glm-4.7-flash"
        },
        {
            "name": "ANTHROPIC_DEFAULT_SONNET_MODEL",
            "value": "glm-4.7"
        },
        {
            "name": "ANTHROPIC_DEFAULT_OPUS_MODEL",
            "value": "glm-4.7"
        },
        {
            "name": "ANTHROPIC_DEFAULT_HAIKU_MODEL",
            "value": "glm-4.7-flash"
        }
    ],
}

Just an FYI, I’m not revealing my real API key here.

If you’re unsure if your JSON is configured correctly, you could always plug it into any LLM and ask it to review and make changes (if necessary)

• Test the model

After editing the JSON, you can close the window and save the changes.

Now, we can start testing out if the connection works.

On any window, we can click on the Claude icon (you can go back to the Extensions tab to find this icon if all of your windows are closed),

and the chat should load.

I did a test and asked what model they’re using, and it gave this output (which shows that it works).

If you want to switch between models, use the /model slash command.

While they show the defaults like Opus/Sonnet/Haiku, they’re routed to the model that you configured in the JSON above.

And this is exactly how you can use any cheaper Anthropic-compatible LLM with Claude Code on the VS Code extension.

The setup is tedious, but worth it

The most painful part of this entire setup is editing the JSON.

While it takes some time to understand how JSON works for non-coders (like me), I can just throw it into ChatGPT to help me configure the file correctly.

I feel it’s worth the hassle because now you’re able to:

• Switch between any model for which you have a subscription

• Avoid being locked into the Claude ecosystem

• Have the same Claude Code UI experience while at a lower cost

This is the foundation of my High-Signal Digital Brain, where I’ve built out a personal OS with Obsidian + Claude Code (while using a cheaper model) that never forgets any of your context and can does the repetitive onchain and social work for me.

If you have any questions on the steps above, feel free to let me know in the comments and I’ll help you out.

Feel free to use my referral link for Z.AI’s GLM plans here (get 10% off your first order).

AI doesn’t have to be overwhelming.

I was an AI skeptic and thought that it would never work for me. But after playing with it for 2 months straight, I’ve built out repeatable workflows that save me from doing soul-sucking tasks that I hate.

If you want to learn how to add AI to your workflows in a way that works for you (instead of yet another influencer template), fill up the form below.

I’ll reach out to you for more information if you’re a fit:

I want in

I’ve completely lost my edge in airdrops because my entire strategy involved low-value actions that can be copied at scale.

No matter how much effort and time you put into grinding, the outcomes you get will be far worse than what you expect.

I’ve been too focused on grinding harder when I should have gone all in on learning how to work smarter.

So I’m completely revamping my airdrop strategy and adapting to this new normal with AI:

Airdrops are so boring now

2025 was the worst year I had for airdrops. Throughout the entire 12 months, I burnt out at least 3 times because I was aimlessly grinding on both my social and onchain footprint.

I spent the day looking for tasks to complete, and airdrops turned into another job for me, while not giving me a good ROI on my time.

And not forgetting InfoFi, which became the worst plague that could possibly happen to a space that was already filled with the farming mindset.

The old projects threw money at us for:

• Grinding out simple social-based tasks or testnet campaigns

• Completing Galxe or other campaigns that had zero barrier to entry so anyone could participate

The Arbitrum airdrop was the turning point because everyone expected every other L2 to use the exact same criteria.

Which is impossible since anyone can spin up bot farms so hundreds of wallets can meet the criteria, or what I call the Sybil war.

Projects continue farming us because they know there won’t be retained users if there’s no hint of a token.

So they spin up endless campaigns and point programs to give us dopamine hits so we continue interacting.

There are too many teams with shady practices that come up with ‘Sybil detection measures’, only to reward all the tokens to themselves.

And of course, there are KOLs who hype up the airdrop raise expectations to unrealistic levels that eventually are not met.

That model of airdrops is dying. It no longer works in this age where low-value actions can be easily replicated by AI.

I’ve completely lost the motivation and fun in hunting airdrops. I’ve realised that 90% of the projects that I farmed will be ones that I’d never touch again.

The only reason why I used it was for the token, and not because it was genuinely useful in helping me achieve a goal.

Now, when I look at the timeline, I struggle to find new projects I should be interacting with.

While airdrops have taught me a lot of DeFi skills that I would never have touched, there’s nothing exciting to me now.

Maybe because I’m a horrible trader that loses money all the time, so all of the new perp DEXes never appealed to me.

I’ve made the mistake of pumping up volume while burning fees, and eventually get a low ROI based on my time and capital.

So, will this be the death of airdrops? I don’t think so, it’ll just evolve into a new model:

Reputation, judgment, and curiosity still matter

Most of us are stuck in the hamster wheel of airdrop farming.

We wake up, put in the work, and hope that we get a good return. We wait for someone to spoon-feed us alpha and we follow it diligently, but it gets diluted the moment it’s shared.

If the only skill we have is knowing how to do things, we will be made irrelevant by AI.

Everything with a workflow can be done autonomously, so what edge do we have anymore?

This was something that cost me my job, and there are 3 main moats that I see to stop us from getting replaced:

• Reputation

The biggest revelation I had last year was how reputation is the moat for Sybils. What you do onchain and what you say online build out your brand.

That brand is what makes you stand out from Sybils. It proves that you’re a high-value individual who deserves rewards, and not just another Sybil with spammy transactions and AI-generated posts.

We’ve seen how @ethos_network is focusing on peer-to-peer human verification as the next defensible layer against Sybils, or [ ] as a new social network that is based on social capital.

The trust recession is the greatest opportunity we have to stand out by actually doing things and sharing about what we did.

Once both our social and onchain profiles match, it gives a greater reason for someone to trust us.

Because too many accounts are making outrageous claims without the proofs to back them up, we start to trust them less (even if they get tons of engagement).

Your reputation gives you access to information and opportunities, so guard it with your life.

Focus on giving value instead of extracting whatever you can through quick cashgrabs and paid shills.

If you’re here for the long run, being patient gives you outsized returns.

• Judgment

Oversensationalised news is everywhere.

Everyone is telling you:

• AI is stealing your job

• The middle class is disappearing because of automation

• You have a few years left to escape the permanent underclass

But most of it is just noise that should be filtered. And how you let that affect your emotions will determine your success.

We are in the age of doing, where anyone can tell AI what they want to build, and it spits out something for us.

Right now, the limiting factor is time. I have so many ideas on what I want to do, but I have no time to do them all.

Judgment helps us determine the highest-ROI actions we should complete with the limited time we have.

Judgment lets us find signal in the sea of noise, so we follow the right people that actually leads the life that we want to have (and not just any plain old larper).

So focus on honing your judgment and building your signal-to-noise filter through these 2 steps:

• Be extremely deliberate with the content that you consume (and curate your feed aggressively)

• Constantly reflect on your results (both wins and failures) as the biggest signal you get is from the actions you do

• Curiosity

With all the information in the world available at our fingertips now, there is absolutely no reason why you should say that ‘you don’t know’ how to do something.

Being open-minded and trying new things was the principle I adapted to my airdrop strategy, and it’s one of the most important skills today.

Though it needs to be paired with judgment so you know what to explore further and what to skip.

The AI boom has taught me that staying with the status quo is the fastest way to lose out.

As a crypto user, I had access to all the agentic tools long before the current crowd, but I didn’t explore further because I thought I was ‘bad at coding’.

But even right now, it’s not too late.

Too many are stuck in the loop of consuming instead of creating. They prefer to explore through others’ experiences instead of creating their own.

And once you break out of that cycle, you’re already in the top 1%.

This chart shows just how far ahead you are of others, once you take that first step and pay for an LLM subscription.

And it’s not too expensive either, I’m using GLM-4.7 and it costs me $10/month. You don’t need to use the frontier models like ChatGPT or Claude, other cheaper models can provide considerable outputs too.

And right now, you can build literally anything. You just need to tell AI what you want to build, and it’ll do exactly that for you.

The more specific the instructions we give to AI, the better the output. It’s not something where anyone can one-shot an amazing product, which is only possible for power users.

While for the rest of us who have zero coding experience, getting better with AI comes through constant iteration, instead of just giving up.

Staying curious and wanting to learn more gives us an edge over the average passive consumer.

What I’m doing now

My account has been fully focused on airdrops, but that will change.

Over the past 2 months, I’ve been extensively playing with AI when I was on hospitalisation leave, and this has helped me understand its capabilities better.

These were the most difficult barriers I faced:

• Getting everything set up the way I want (I took 5 days for my OpenClaw stack)

• Connecting to different APIs (some like Google are so frustrating to generate the API keys)

• Managing token burn (spamming questions is the fastest way to reach your rate limits)

And right now, here’s how I’ve set up my OpenClaw (as an Arsenal fan):

• Arteta: Manager and overseeing the operations of the other 3 agents

• Saka: Running my personal brand with content ideation and social listening

• Dowman: My job hunt assistant to find opportunities for me to apply to

• Gabriel: My wallet tracker and alpha hunter, will probably give him a wallet soon to optimise for profits

These are the functions that I want to automate in my life, and I’ll be sharing how I set them up along with other guides on efficient token usage and more.

So if you want to learn how to integrate AI into your social/onchain workflows, stay tuned for more posts in future.

Get out of the high-noise, low-agency loop

Those who get left behind are the ones who consume instead of creating something meaningful.

Airdrops have been a painful lesson, where I was stuck in the grinding loop that started giving poorer returns because of overdilution.

But with AI, I now see how I can compress time to explore new strategies (both onchain and social), get results, and then iterate to improve my strategy.

And I strongly encourage you to take that first step and use AI to create anything.

Find one task that you hate doing and outsource it to AI to get it done.

And once you become confident with that workflow, start to scale up and continuously solve problems you face in your life.

Once you see AI in this light, you’ll notice that it is just a tool, just like the Internet. It compresses your time by removing all the boring automation from your life.

If there’s a best time to adopt AI into your workflows, it’s now.

So start taking action, and start changing your life by becoming AI fluent.

You don’t have to be overwhelmed by AI.

If you want to work with me to build your AI fluency and stop your fear of falling behind:

Fill out the questionnaire below and I’ll reach out to you:

I want in

Too many influencers shill their cheap OpenClaw setups, but they require you to commit to a 2-year period, or the cheap prices are only available for the first year.

Over the past 4 days, I’ve been searching through YouTube and Reddit to find the best way to set up OpenClaw without paying a high fee.

https://x.com/fipcrypto/status/2029391150919106568

I finally settled on this setup that costs me $10.92/month for the first year, but will increase to $14.50 afterwards (provided that the pricing still remains the same).

This includes a combination of:

• Tencent Lighthouse: $10.99 for the first year, $50.40 (+tax) afterwards

• GLM Coding Lite Plan: $10/month

So here are my considerations that brought me to this setup (and to help you see if it’s a fit for you too):

Why I’m using a VPS

I chose to run my OpenClaw on a VPS because of these reasons:

• My WiFi is unstable at home, so it could lose connection at any time

• I don’t want to use my main device to run OpenClaw because of the security risks

If you have an extra device that’s lying around, then it’s possible for you to skip this immediately.

@0xdgw shared with me how he’s using an old Apple device that he hasn’t touch in a while and it’s been working well for him.

So that’s how I started on the hunt for a good VPS that meets my needs:

Finding the best VPS for my OpenClaw setup

But when I tried searching for a good VPS setup, I was bombarded with videos telling me how amazing Hostinger was with their cheap prices.

The best deal only activates when you choose a 24-month plan. If you’re like me and you don’t want to be locked up to just one VPS, the monthly cost for the basic plan will skyrocket to $19.49/month.

There were others that I considered too but they were lacking in certain areas:

• Hetzner is extremely cheap, but I’m not sure if the VPS will be laggy since I’m in Singapore (and they’re in Europe)

• Railway’s UI is great, but I had issues with integrating to Discord and Slack through the OpenClaw web UI

• DigitalOcean has one-click OpenClaw installation, but my account was locked and it took a while before it got reactivated

I wanted to start OpenClaw immediately, and saw a Reddit comment on how Tencent’s Lighthouse promo made a 1-year plan very affordable.

So I decided to check out their plans, and they gave one of the cheapest promotional rates (for up to 1 year).

And even if their specs are not the best (2GB RAM, 40GB SSD), I found it was good enough for me to explore how to use it effectively first (since I’m a complete beginner).

So I decided to go with Tencent Cloud and bought the $10.99 USD plan for 1 year (including taxes).

The original cost of this plan is $50.40 USD (excluding taxes).

After using it for 4 days, here are my quick thoughts on the plan:

• UI is not user-friendly at all

While you get a one-click installation of OpenClaw, there are many items that you need to set up yourself too.

Once you launch your instance, you’ll be greeted with a command line (which was intimidating for me).

I had a tough time looking through the documentation, which skips certain steps that someone like me desperately needs.

And it doesn’t help when their formatting is not the best.

So this is my greatest gripe with Tencent, and I spent many hours troubleshooting simple problems with my LLM (and burning lots of tokens along the way).

It does take a while to get used to the command line and the interface, but I’m more familiar with it now (so feel free to ask me anything if you face any questions).

• File explorer and CLI

Right now, I’ve only used Railway and Tencent as my VPS, so this could be biased.

While Railway’s UI is great, I had problems finding the command line interface to install software, or the file explorer to modify my configurations.

Up till now, I have no idea how to access both of them.

While for Tencent, the CLI and file explorer are right in your face, and I could easily use both (with the guidance of ChatGPT).

Though one silly mistake I made was not showing the hidden files. After restarting my server, I had a mini panic attack when I couldn’t find all of my files (that were actually still there but hidden).

This is just my personal preference, but I find it much easier to navigate across Tencent’s UI, especially during troubleshooting when everything goes wrong.

I made a mistake with my Railway config and I can’t even launch OpenClaw now, so I gave up on it completely.

Choosing a model to run OpenClaw

The easiest way I could have used was just choosing the $30 SGD/month plan for Codex / Claude Code (it could be cheaper for you depending on your country).

But the rate limits are not the best for Claude. Thanks to the @ethos_network vibeathon, I had a first-hand feel of what the Claude Pro plan was like, and it’s not the best.

https://x.com/fipcrypto/status/2016169170501829119?s=20

I kept hitting the 5-hour rate limits when I was making simple modifications to the code for my project, and I’ll likely face the same issue when building my OpenClaw agent.

Since I’m completely new and would likely burn tokens on simple tasks.

So these frontier models aren’t the best fit for me, and I decided to look at the Chinese models instead:

• Kimi

• GLM

• MiniMax

The differences in coding capabilities don’t matter to me, I just want something that’s cost-effective so I won’t keep hitting the rate limits.Most of them claim that we’ll get 3x usage of Claude Pro for a lower fee, so I wanted to try it out.

In the end, I went with Z.ai’s GLM Coding Lite at $10 USD/month for these reasons:

• I could pay for the subscription with PayPal (I have some USD balance there and MiniMax charged taxes for their API and coding plans)

• I could generate an API key for my plan to use with OpenClaw

The quarterly and annual plans are cheaper, but I don’t want to be locked up with one model provider and I’m going for the monthly plan instead.

In just 4 days, I’ve burned through 69 million tokens and used up 86% of my weekly quota.

I’m sure this can be reduced after the initial setup, which likely burned a lot of unnecessary tokens based on the simple questions that I asked.

Right now, I’m still building an agentic orchestration workflow with 1 manager and 5 bots based on @jumperz’s configuration, and most of them are using GLM-4.7 Flash instead of the main model.

Subscribe now

But I’m still monitoring to see if the token burn can be reduced:

So, what’s next?

The GLM coding plan may not be sufficient for all of my agents, so I’m still looking for the best model setup to optimise my costs.

Some that I’ve been considering:

• Using @bc1beat’s Clawrouter to optimise model costs with x402 calls

https://x.com/bc1beat/status/2030322072329548219

• Buying a ChatGPT Pro plan that includes Codex, so I can do the main planning and delegation of tasks with higher rate limits

At just $10.92 a month (for the first year), this Tencent Cloud-GLM setup is the most affordable that gives me a low-cost option to experiment (and fail) with OpenClaw before I decide to kill off the project completely, or upgrade to a better VPS.

Most of this could be extremely confusing to you, so feel free to DM me if you have any questions about setting up OpenClaw to work for you.

I plan to do a livestream that will walk you through the complete steps to install OpenClaw from scratch, so let me know if you’re interested in that too.

If you enjoyed this article, here are the links to sign up for the 2 software:

• Tencent Cloud Lighthouse Instance

• Z.AI GLM (referral link, get 10% off your first order)

You don’t have to be overwhelmed by AI.

If you want to work with me to build your AI fluency and stop your fear of falling behind:

Fill out the questionnaire below and I'll reach out to you:

I want in

Many anxious thoughts are running in my head right now with the looming uncertainty come end March, and it couldn’t come at a worse time:

• The algo has completely wiped out my reach on this account

• I’m questioning the type of audience I’m attracting

• Airdrops have become such depressing events

The account that I’ve worked on since 2023 on the side was meant to give me a tool for freedom, but it’s far from that now.

Since the end of 2025, I’ve started to seriously question why I’m still here, when airdrops have become toxic, and crypto is completely falling off a cliff thanks to AI.

AI has become the major disruptor of everything. The era of hard work is dead. Anything that can be easily replicated is now a commodity.

Where then do you and I fit inside this new normal?

The anxiety only gets worse now because I have no job left after March, but it could be my biggest opportunity yet:

You have every right to fear AI (but you don’t need to)

I’m writing this down for everyone who is scared of AI replacing their jobs, because I’m exactly where you are right now.

AI will come for you, but only if you are replaceable.

I was exactly that at my job. I was excellent at following orders and executing tasks, but had zero agency. I struggled to come up with ideas that could lead to exponential results, preferring the safe routes because of the fear of failure.

But it’s exactly because of this inaction that cost me my job.

Anyone who’s just a follower and a doer (like I was) would not be of much value to a company, when they could just use AI to complete those tasks (with faster and better results).

This was something I saw in airdrops too.

Back during the Arbitrum era, there were tons of farming services that helped you automate transactions for LayerZero and ZKsync, which were ultimately filtered out or were just not eligible.

https://x.com/cedehub/status/2026333763400024529

Any low-value task is one that is easily replicated, and will not be eligible for rewards.

Which is why testnets no longer make any sense since there is zero risk or upfront costs.

But even with all this doom and gloom, AI is not your enemy.

I spent too long fearing AI or belittling it, but I finally saw what role it should play in my life:

AI is your amplifier

One of my key goals during my post-surgery recovery was to learn about AI. And I’m so relieved that I put in all that time to become more familiar with it.

In the past, I thought that AI wasn’t that big of a risk. But over the past few months, the cycle of new model releases has become shorter with the tech accelerating at a crazy pace.

And as usual, the fear-mongering has become so intense online, with everyone pressuring you to escape the permanent underclass before it’s too late.

But anyone can say anything without any repercussions here. There are too many accounts that make over-sensationalised claims about the returns that AI gave them, when most are fake or are blatant lies.

The only signals that you should trust are the actions that you did yourself. You know the real results because you experienced it yourself, and that guides you on what works and what doesn’t.

https://x.com/fipcrypto/status/2026642850604282212

With this principle, I focused on doing rather than watching yet another vibe coding lesson that likely won’t motivate me enough to start. Over the past 7 weeks, I tried and failed many times before building these out:

• A full content system to expand 1 idea into multiple pieces of content across different platforms

• A vibe-coded site for @ethos_network’s Vibeathon that lists both onchain and social metrics for a protocol

• A full-fledged documentation on @mintlify with nice formatting that I’d never have done myself (because of all the coding that I know nothing about)

https://x.com/fipcrypto/status/2027549069393244220

After spending days organising and consolidating my life, I now have a full-fledged Digital Brain that runs (almost) my entire digital life, doing all the boring tasks that I hate.

I was inspired by the content libraries that @Zeneca and @yannisDeFi were building out, and I built mine out with Obsidian and Claude Code.

https://x.com/fipcrypto/status/2027549084836630625

These experiences taught me that functionality matters more than the aesthetic, the product needs to work and solve the problem.

In my case, I decided not to build out a UI for my library and chose to use Obsidian in the plain Markdown format instead.

And after seeing @AzFlin’s wallet tracker, it finally clicked for me that I could expand my use for AI and go onchain.

https://x.com/fipcrypto/status/2026461906542932386

There are so many possibilities that I can build right now:

• A recursive learning trading bot that helps me profit from perp DEXes (instead of blindly farming them)

• Integrating both onchain (do) and social (post) in one unified system

• A ‘judgment’ protocol that learns from my past results in airdrops to suggest high-ROI ones to target

Through this entire period, I’ve learnt that AI is just a tool for you. Those who can master it are the ones who understand this:

You need a 3R System to be successful with AI

I’ve been reading @thedankoe’s content even before he blew up on CT.

My Signal Creator Challenge was heavily inspired by him too.

And one of the key things that I learned from him:

AI works best if you create a system and get it to execute tasks for you (which was exactly what I was doing for my job).

You become its manager, giving it the outcome you want to achieve and the steps to achieve that goal. It knows your direction, and will do the brute work for you.

But if you don’t give it enough context or are unclear of the goal that you want to achieve, the results will be average.

Once you start thinking of everything in systems and atomic tasks, you can find ways to insert AI inside that workflow.

Something as simple as mass unscheduling tweets from @typefully can be done with just one prompt.

https://x.com/fipcrypto/status/2028003312264749107

You don’t even need to know coding (like me), all I did was give the model the API documentation and my keys, and it executed that task right away.

When it would have taken me at least 20 minutes to unschedule all of them manually.

So I’ve learnt that we need to build 3R systems to maximise our returns with AI:

• Repeatable: The system works across different contexts and situations with a clear workflow that can be adapted to new circumstances

• Recursive: AI consistently learns from your results (both wins and failures) and improves the system over time

• Reputable: The system that works for you becomes your IP when it gets used by others to achieve the same outcomes as you

Once you build out this 3R system for a task, it continuously compounds with AI amplifying your actions.

How I plan on getting my job back with AI

The pressure is real for me because I now have additional financial commitments to manage. Everything was fine with my full-time job, but it’s now changed.

There are 3 routes that I see myself going in:

• Another full-time job

• Freelancing for clients

• Going full-time with my personal brand

With my newfound experience in building with AI, I have many ideas on saving time on all the repetitive tasks that I hate doing for these 3 routes(especially for job applications and resume updates).

I may even explore a hybrid by choosing to go for both freelancing and full-time just to tide things through.

I’m both scared and excited at this uncertainty because the stakes are high, and it’ll be a big challenge. But I’m taking a big long on myself (like what @icobeast said some time back) and it will eventually pull off (or I’ll make it happen).

https://x.com/icobeast/status/1916518862767395121

So throughout the next month, I’ll share more on how I integrate AI to build out solutions to the problems that I face.

And hopefully, that inspires you to take action along with me:

You don’t have to fear AI

Yes, I’ve become yet another account that talks about AI. Just like many other CT accounts that have pivoted since it gets the most clicks and eyeballs right now.

But I’m not forgetting my roots, and want to explore how I can use AI to compound my onchain and social reputation.

I’m sure many of you are in the same boat as I was in before. A recent poll in my Telegram channel showed that many are keen on learning how to implement AI into their workflows.

I’m grateful for this 7-week break that gave me the time to explore and fail to understand how to use AI effectively, even as a non-coder.

It has been an enlightening experience for me, and I want to help you achieve the same too.

So that’s why I’m building out Signal Compounder, a system that helps you build repeatable workflows that compound your social and onchain reputation over time.

Let me know how I can help you by filling out this form.