March 29, 2026 AI, AI Agent No comments Vibe Coding
Conway’s Game of Life, Coded From a Mall
When I was in university I coded “Conway’s Game of Life”, which is a primitive simulation of cellular life, if there are a certain number of live cells around they produce more cells and if conditions are not favorable cells die off. I don’t know what the meaning of life is. Would at some point the argument come to say that AI is alive? This is very philosophical, this post is instead a very practical showcase of some of the most recent tools and advances in AI as I play with them.
Just have a look at this photo:
As you can see I’m somewhere in a mall, asking my Claude Cowork running on my mac at home to create a folder ~/Projects/gameoflifecowork and generate a single HTML page with Conway’s game of life implementation. I came home and there it was an html page right in that folder with the implementation. It is perfectly working. I’m adding it here. If you are reading from e-mail you would need to open the blog to see it in action.
[if you are reading from e-mail you might need to open in browser to see The Game Of Life]
Cowork is an extremely powerful (and dangerous tool). It is not yet a very smooth experience. Sometimes I need it to prompt multiple times, most of the time there is no proper feedback in the Mobile app so I don’t know what is happening. For instance, I also asked it to go to my “Downloads” folder and locate any concert tickets and tell me how much I paid for them. Unfortunately there was no way to see on the mobile app the answer to the tickets (not in any chat), but when I came home I saw a dedicated chat open on my computer that had the answer.
Under the hood Cowork runs Claude Code to implement the game of life, so I was wondering if I can compare different models and how good of a job they do, so I installed Open Code and connected Gemini API, Anthropic API, and also local LLM!
With Gemini API I generated a really great fast full screen Game Of Life, which cost me about 0.20$. Local llama3.1 unfortunately is not suitable for this task, I had to give it many additional instructions and it messed up every time until in the end I got an empty html file with some broken functionality, which I fixed with Copilot just to get it render:
Gemini’s Game of Life was full screen and rendered perfectly:
I then switched to Claude Haiku 4.5 to generate the Game of Life you can see above.
This implementation cost me about 0.33$.
What was a small mini-assignment at university to code the Game Of Life, which I did with C++ and probably took me few days, now turns out to be just 0.20$-0.30$ throwaway code just to test different API integrations. Right now I understand why those cells live and die as I wrote this algorithm myself in C++ but I’m wondering if at one point we will not know what is happening inside implementations generated by AI.
Life remains for the most part a mystery to science, though we are getting closer and closer to understanding how it works. At the same time a reverse is happening with AI we are slowly getting further and further from understanding what goes into the beautiful implementation of Game of Life.
Completely Sold On AI
March 22, 2026 AI, AI Agent, Personal No comments
Today I’ve done multiple personal things that fully sold me on AI. No way back. Sorry.
Use Case 1: Complex Tax Filing
No question using AI for Tax purposes is useful, but today I got it to the next level. To understand my tax situation: I moved from Canada to the US in 2025, I have rental property, stocks bought and sold in both Canada and US, retirement accounts in both countries, stocks/ETFs transferred from different institutions back and forth, etc. With AI handling all of this is basically avoiding nightmare and frustration taking many days on.
For specific example, I had VFV.TO which is ETF traded on Toronto stock exchange, not only I got some purchases and sales starting 2021 they tracked through multiple brokerages and income from dividends was reinvested back into shares, not only it is over many years, banks, currencies, countries, but also dozens of statements and documents, it is also treated in special way in USA where they treat it as Passive Foreign Investment Company (PFIC). Other than that the day I moved countries I “deemed disposition” of my assets. If I had to do all these things manually I would just get petrified, but instead I can just throw about 30 documents at AI and ask it to generate a final spreadsheet with all of the transactions, and also have source documents referenced so I can double-check. It all worked really-really well.
Use Case 2: Organizing My Files since 2000
Yes, I crossed the line and allowed Claude Cowork access to my local files.
I keep files dating from the early 2000s, keeping most of the things archived in Dropbox. I would keep such things as any of my university course works, archives of code snippets I have written in 2005, archives of my blog, some old presentations, tax documents, all kinds of agreements, all kinds of things that basically track my life since it became digital. It grew over many years. I tried to organize things in some iterations in the past, but then mostly gave up on just naming things more nicely and using search. But with AI I was just able to say “Create a plan to organize files in this folder X”, then iterate over the plan and let it execute things. The best part is that this is just natural language.
Use Case 3: Organizing Life
I previously posted that I do track things in life and that I use AI running chats for each area of life, like “Health”, “Finance”, “Career”, etc. This is all good, but then I wanted to build my personal AI agent to help me with tracking, but tools like OpenClaw and Claude Cowork made that unnecessary. Now, I’m playing with migrating my life planning into simple .md files. My life is essentially one repository that I can query and manage with natural language. Nuts.
Use Case 4: Posting on Instagram
Haven’t posted my climbing videos on instagram for a while, but I thought maybe I would post one latest video. I thought that analyzing video is not great with LLMs, but no. I uploaded my climb, it recognized that I was wearing martial arts t-shirt, it recognized at what points major moves where happening and I asked it to suggest music that would work well for the video matching my style but also the pace of events in the video. This is nuts. As next step I was thinking of fully automating video edit and posting on my behalf. Nuts.
Use Case 5: Work
While I cannot talk too much about work, I would just say that there was a step-function improvement in AI use and productivity. Producing code is faster, iterating over ideas is faster, getting things out just gets faster. I need to constantly adapt so that I don’t become a dinosaur that dies with a fallen AI asteroid.
Conclusion
It appears I grew from an AI skeptic, to AI learner, to fully embracing it by now. No way back.
edited with Claude Cowork Computer Use from mobile phone
Surviving the AI Asteroid: Hype Cycles, Overreactions, and What Comes Next
March 15, 2026 AI, AI Agent, Opinion No comments
Previously I compared AI to an asteroid, with the simple premise that if your job is boilerplate CRUD, your job is a dinosaur, and if your job involves high leverage of AI, including building AI-integrated systems, your job is the mammal that survives.
AI Hype is real, it is everywhere, and honestly is somewhat annoying. My LinkedIn feed is oversaturated with all of the AI noise, I keep overhearing “AI” when walking past random people on the street. Everyone has their say on AI, including me. AI washing is a real marketing tactic used by many companies. Many people would get into AI just purely because of FOMO. And, honestly, that fear is justified, because what if you are really missing something and will be left behind. No one wants to be left behind. As a simple example, it is even hard to get a Mac mini because everyone is buying them to run their personal OpenClaw. I’m playing this game as well. I did set up OpenClaw on Docker just to see what I could do, but until I have a sustained workflow, I won’t be buying a dedicated machine. It’s easy to confuse playing with new tools for actual productivity. But, maybe, I’m missing out.
There is no smoke if there is no fire. Last week Anthropic released a labor market impact study claiming that hiring has slowed in highly AI-exposed roles since ChatGPT launch. For us, software engineers, the study claims that AI can theoretically automate about 90% of our jobs and it appears current automation is only at about 30%. If this is true and if this is happening soon, some kind of a combination of the two will happen: 1) our jobs will transform by a lot so that we are building ever new and more complex things that AI cannot and/or 2) there will be significant reductions in software engineering jobs. I don’t know if 1 or 2 would be a larger component of transformation but we should be preparing for both!
Image credit: Anthropic https://www.anthropic.com/research/labor-market-impacts
As I use Claude Code, it becomes apparent how it becomes more and more capable over time. It is no longer a question of whether the threat is real. It is absolutely real. The asteroid is here! It has hit the ground already. The transformation is already happening and if you don’t see it, you might be in trouble (that is unless you are a plumber or someone with a low exposure job). I still see value in myself by figuring out what problems to solve and then directing the work to my AI agents, sometimes finding myself directing 4 of them at once, which is really cool to be able to make progress on 4 things at once, but it is also terrifying. Does it mean I’m now 4x productive? Does it mean we need ¼ of engineers to do the same job? Or we will simply see another instance of the Jevons Paradox. Historically, making software development cheaper/faster didn’t mean we hired fewer engineers. What we have seen is that demand for more software has increased thus increasing the number of software engineers. But still, there are so many open questions that come with this transformation: like
- where do we get senior engineers if there is no need for juniors?
- would we ever reach “lights out” codebases not requiring human code reviews?
- would we hit the energy limits when it is more cost effective to pay a biological human rather than energy consuming AI? … and so many questions.
- what’s your question?
What goes up will go down. Things happen in cycles. What is born will die. I made the mistake of buying real estate in Vancouver at the peak of the market in early 2022, my property price has not yet recovered. Do you remember the COVID tech hiring frenzy followed by layoffs? Our industry over-hired only to lay off people after that. Do you remember the 2008 financial crisis and what happened to real estate prices only to recover some years after that? I am not old enough to remember the dot com bubble but it is all the same all across. My argument here is that we will see some kind of cycle of overreaction with AI as well. Some companies will over-invest in AI and not get anything out of it. Some companies will lay off too many people only to hire back. What we are looking at are micro movements, but what is more interesting is what will this bring us long term, what is the macro movement here?
Image credit: Gemini on my prompt. The image is too colorful for my taste but it is kind of fun.
Dear reader, prepare for multiple outcomes. They say to have peace, you must prepare for war. Build a strong financial safety net, constantly stay on the lookout for what is changing, and adapt relentlessly. Every technological shift creates massive, unseen upsides. The rules of the game are changing and instead of panicking, the goal is to understand the new rules so you can keep playing.
My 1-Hour Open Claw Setup: Docker, Llama 3.1, and Telegram
February 22, 2026 AI, AI Agent, HowTo No comments
I saw all the fuss about Open Claw online and then spoke to a colleague and she was saying she is buying a Mac mini to run Open Claw locally. I could not resist the temptation to give it a try and see how far I can get. This post is just a quick documenting what I was able to do in like one hour of setup.
If you’re like me and find it difficult to follow all the latest AI hype and missed it, Open Claw is an open-source AI agent framework that connects large language models directly to your local machine, allowing them to execute commands and automate workflows right from your terminal or your phone.
A quick preview below. This is just nuts. In one hour I was able to run OpenClaw on Docker talking to llama3.1 running locally and communicating with this via Telegram bot from my phone 🤯.
Back in the old days I would open some kind of documentation and follow steps one by one and unquestionably get stuck somewhere. This time I started with Gemini chat prompting it to guide me through the installation and configuration process. This proved to be the best and quickest way.
Decision 1: Running locally or in some isolation.
I think this one is an obvious choice. Giving hallucinating LLMs permission to modify files on my primary laptop sounds like a recipe for disaster. Decided to go with Docker container but if I find the right workflows I might buy Mac mini as well.
Commands were fairly simple, something along these lines:
git clone https://github.com/openclaw/openclaw.git
cd openclaw
./docker-setup.sh
docker compose up -d openclaw-gateway
Decision 2: Local LLM or Connect to something (Gemini, OpenAI)
This is a more difficult decision to make. Even though I’m running an M4 with 32GB, I cannot run too large of a model. From reading online it is obvious that connecting to large LLMs has an advantage of not hallucinating and giving best results but at the same time you’ve got to share your info with it and run the risk of running into huge bills on token usage. Since this was purely for my self learning and I don’t yet have good workflows to run, I just decided to connect using a small model llama.3.1 running via Ollama. Since it was running on my local machine and not docker, I had to play a bit with configuration files but it worked just fine. And yeah, the answers I would get are really silly.
Later I found that ClawRouter is the best path forward. Basically you use a combination of locally run LLM and large LLMs you connect to. I might do this in the next iteration.
Decision 3: Skills, Tools, Workflows
This is just insane how many things are available. Because this can run any bash (yeah, in your telegram you can say “/bash rm x.files” – scary as hell) on the local the capabilities for automation with LLMs are almost limitless.
Conclusion
I can barely keep up with all of the innovations that are happening in the AI space but they are awesome and I’m inspired by the people who build them and feel like I want to vibe code so much more instead of spending my time filling-in my complex cross-border tax forms over the weekend.
Building Personal AI Agent
January 11, 2026 AI, AI Agent No comments
Rich people always had access to assistants (+chief of staff) that would help them with all kinds of chores, would advise them on things, or just do things behind the scenes. We live in an interesting time of AI, where anyone has access to these fabulous LLMs that can do some of those things for us. Like, I’m sure most of us are doing our travel plans with some help from LLMs, or we make buying decision, etc. It is just mind boggling how good these things are becoming!
In parallel, as software engineers, we keep hearing about AI agents all the time. We use AI agents at work. The most useful and prominent example of LLM agents are coding agents. You are likely using Claude Code at work. I already heavily rely on LLMs to track many of my personal goals, to critique me, to give suggestions, etc.
But what if I build a personal AI Agent to avoid repeating things and to make it watch me more proactively?
Vibe Coding AI Agent
There we go! Let’s build something simple first. My use cases for LLMs are fairly simple, nothing too crazy and very closely tied to my Life Goals and areas of life. For example, I have chats with Gemini labeled like “Nutrition”, “Finance”, “Career”, etc. When I eat my breakfast I snap a picture of it and estimate my nutrition intake. When I’m considering stock buying I do research with Gemini. When I plan a trip I build an itinerary with LLM, etc. I track my weekly progress in google docs. I track my finances in spreadsheets. The more I think about this the more I realize there is a room for a personal AI agent that is highly tuned to my personal needs and would orchestrate all of this. Additionally there won’t be any ready solution online, because this is so personal, so I’ve got to build one agent for myself!
Email Digest Agent after ~3 hours of Vibe Coding
So what I’ve built in 3 hours is a “personal AI agent that automatically generates daily digest emails by fetching data from multiple sources in parallel: it retrieves stock market insights for ~X tickers, extracts the current week’s goals (with checkbox tracking) from a tabbed Google Doc, pulls my monthly focus items, pulls net worth data from a Google Sheets dashboard, then uses Gemini to generate a professionally formatted HTML email summary and sends it via SendGrid. The system is built with LangGraph for workflow orchestration, uses OAuth2 for secure Google API access, and preserves formatting details like checkboxes (✅/⬜) by converting them to email-compatible emoji before direct insertion into the final email.”
Tech Stack
- LangGraph 0.2+ – Workflow orchestration with parallel execution
- LangChain 0.3+ & Gemini API (gemini-2.0-flash-exp) – LLM integration
- Google Docs/Sheets API – Data sources with OAuth2
- Alpha Vantage API
- SendGrid – Email delivery
- Python 3.9+ – Runtime
Conclusion
This project is just scratching the surface of what is possible to be built very quickly for personal needs. The most exciting realization for me wasn’t the technical implementation but how accessible it is to connect things together. Setting up all of the API keys and then vibe coding all together is so straight forward that it is just unbelievable.
I will continue this project next week to make it actually properly work for my needs and then will host it on some server to send me those digests. Next steps would be to supplement it with prompts to LLMs to give me quick ideas for what I should focus on, better tracking, etc so I can achieve my goals quicker. So exciting!
Go ahead and build something for yourself!
