Tactics Journal research. Get ahead of the game.

I dropped everything to research agentic AI for three months because I saw the impact agents will have on individuals. Not chatbots. Agents. Autonomous systems that work for you, day and night, at near-zero cost. I built one for football.

Tactics Journal research does four things.

It monitors more football content than any person could. Articles, press conferences, podcasts, coaching interviews, analysis — across leagues, languages, and cultures. Every hour, it pulls in new content and stores it as searchable embeddings. German tactical blogs, Portuguese analysis, press conferences where managers drop phrases weeks before anyone writes about the concept behind them.

You gain experience from the entire world of football, not just the English-speaking portion.

It finds ideas that haven’t crossed over yet. Not by reading headlines. By classifying sources as frontier or mainstream, then looking for ideas circulating in smaller, independent sources before they show up in major coverage. Ideas backed by multiple sources across different football cultures score higher. Single-source weak signals get penalized.

It writes daily deep research reports. Only candidates that pass a quality gate get written up. Each report is researched against a stored corpus and the live web, citation-checked, and revised. Every report requires counterevidence. If the evidence is thin, the report says so.

It tunes itself. Built on the same idea as Karpathy’s autoresearch, the system experiments on its own detection and reporting settings. Keeps what works. Discards what doesn’t.

Everything will be open source. There is no point in hiding how your product works when it is this easy to take something, replicate it, and customize it to fit exactly what you want.

A report will be published daily. One report per week will be available to anyone for free. Subscribe for $15 per month to access every report. More subscribers means better models, which means better reports. The quality scales with interest.

Subscribing is cheaper than hosting it yourself. A subscription costs a fraction of the ~$40/month you’d spend running it yourself on hosting and LLM calls.

I’m still testing and fine tuning things. Join the waitlist to get early access.

Most people interact with AI through chatbots. ChatGPT, Claude, Gemini. You ask a question, get an answer, then go apply it yourself. The work still falls on you.

Agents are different. You don’t ask them a question. You give them a job. Research, execution, testing, delivery. The tedious work runs in the background while you focus on the things that actually matter.

Two breakthroughs stopped me in my tracks.

OpenClaw by Peter Steingberger showed what autonomous agents look like when they act as a personal assistant, working unprovoked. Not waiting for a prompt. Just doing the work.

When you try OpenClaw, Hermes Agent, Zo Computer, Manus AI, Perplexity Computer, or Claude Cowork you see where the world is going. When you see things like Mission Control or Polsia you witness the power of building agents that work for you.

autoresearch by Andrej Karpathy showed something else: a small system for training LLMs where an AI could run experiments by itself. It made a change, tested it, checked whether the result improved, kept what worked, and discarded what didn’t. By morning, you had a trail of experiments and a better result. It was built for LLM training, but the loop itself can be adapted to other kinds of optimization.

The cost to build and run things like this is collapsing toward zero. The differentiator is no longer knowing how to build. It’s having the imagination and courage to create something.

Mark Cuban said it recently, “Software is dead because everything’s gonna be customized to your unique utilization.”

The generic tool era is ending. What replaces it is individuals building exactly what they need.

Even if you have no experience in coding, with tools like Codex and Claude Code, anyone can ship something. You are not behind if you start today. But there is no excuse anymore.

And if you don’t want to build anything, there will be someone ready to build it for free or at a lower cost.

This will bring changes for the Tactics Journal. There will be two arms: Opinion and Research.

Opinion will always be authored by me. I watch matches. I think. I write.

There are two types of LLM users: those who use it to learn everything, and those who use it so they don’t have to learn anything. I want to use it to enhance what I do every day.

When everything can be made self-made, what becomes scarce is the thing that can’t be generated. Human time. Human attention. The proof that a real person chose to spend their limited hours on something rather than everything else they could have done. A handwritten letter means more in a world where AI can write a perfect one instantly. The effort is the message. The opinion side of this site will always be me. My observations, my voice, my time.

The research side is an embodiment of what I think we should be paying attention to, tuned by me. I watch matches and write about what I see. The system reads everything I can’t and surfaces what I’d miss.

There are new breakthroughs, solutions, models, and ideas coming out every day. You need a full-time job just to keep up with it. But the trajectory is clear. The cost of intelligence is approaching zero. The cost of human attention and judgment is not.

I want to build platforms where agents can act unprovoked. Not waiting for a question. Just doing the work they were built to do.

The people who build these things for football, for journalism, for scouting, for anything will have an unfair advantage. Not because the technology is secret. Because they had the taste to know what to build and the initiative to build it before everyone else.