Seventy Weeks, One Epoch

It was my son who alerted me. He noticed the whole thing early, the way young people do, and because he noticed it I started exploring early, and thinking about it early, which turned out to matter more than the exploring.

But my actual first touch came before that, and it came sideways, the way important things usually arrive.

A room that did not want me

At one point in my life I worked with the Dota 2 team at Natus Vincere. Not on drafts or strategy – I would have been useless there. My part was medical: recuperation, recovery, the unglamorous machinery that keeps a body working when its owner is young and quite sure that bodies are infinite.

Their superiors wanted the team to progress on all fronts. The team, reasonably, wanted to play Dota.

So one hasty morning I stood in front of all of them with a presentation. I can still see the faces. Polite, patient, entirely elsewhere. What on earth is this oldish boring guy talking about. I did not blame them then and I do not blame them now. From where they were sitting, that is exactly what I was.

One of them was famous. Properly famous. And among the things he was famous for is this: Dendi vs. OpenAI at The International 2017. Main stage, one against one in mid lane, a human legend against a bot from a research lab most people had never heard of.

He lost.

So OpenAI entered my life well before it became a chat window that everybody knows. And the idea behind that bot was already the idea we are all living inside now. You do not teach it Dota. You put bots in an isolated environment, let them play against each other, and you let them run for epochs – pass after pass after pass, more repetitions than a human life has room for. And one day, by nothing more mysterious than the sheer number of repetitions, they know how to play.

I remember thinking that this was strange, and a little sad, and almost certainly correct.

A monster GPU, for its time

The journey with my son properly started with Stable Diffusion. He was among the first people I knew to run it himself, on his monster GPU machine.

I smile writing “monster.” What counted as a monster GPU then would not be let near the models we run today. The word survived. The hardware did not.

Still, I remember images arriving out of noise, slowly, in front of us, and the particular quiet that fell in the room while we waited for them.

Journey, journey – and then Midjourney

Then came Midjourney, and Midjourney lived in Discord. To make a picture you typed your request into a chat channel like everybody else and waited your turn in public.

Which is how an oldish guy who had gone through life perfectly content never to touch Discord ended up opening a Discord account.

I still do not know why it worked that way. But it worked. And it was the first time I ever paid a subscription for the use of a model, and what surprised me was not the pictures. It was the popularity. A user journey this complicated should have buried the product, and instead the whole world stood in that queue with me. Which meant there was something about it I was not seeing. I am not sure I have found it yet.

Confidently wrong

Then, quite a bit later, ChatGPT, and I was amazed. After that I went wandering through open models, running this and trying that, mostly to see what the floor looked like underneath the ceiling.

Two discoveries from that period stayed with me.

The first was how confidently it can be wrong. Not hedging, not vague. Wrong the way a very charming and very certain person is wrong, which is the hardest kind of wrong to catch.

The second was prompting – the techniques, the incantations, the whole small folklore that grew up around getting a good answer. And here is the thing about that folklore: most of it stopped being relevant. It mattered maybe two years ago. I still see people posting it today as if nothing had moved. The tools changed underneath the advice, and the advice kept walking.

---

The first of April, 2025

On 1 April 2025 I subscribed to Claude Pro. Then Claude Code. Then I never stopped.

Building, learning, exploring, reading, researching. Not a project with an end date. Just a way of spending the day, and then the next one.

Seventy weeks

That was seventy weeks ago, and it went by like an epoch ending.

I mean that word the way the bots meant it. I was not in an isolated environment and I was certainly not fast, but the mechanism was not so different. Pass after pass. The same problems approached from a slightly different angle each time. And one day you notice you know things you cannot remember learning.

I am on Claude Max 20x now, running seven or eight projects at once and finishing most of them – at least the ones where the thinking was ready. The rest taught me that thinking is the slow part. And that finished projects spawn new thoughts, and new thoughts spawn more projects, which is why the list never actually gets shorter and why I would be disappointed if it did.

Quite early on, after only a few weeks, I hit the wall everybody hits: memory. Every conversation began from nothing. I explained the same architecture, the same preferences, the same history, over and over. That is fine once. By the fiftieth time it is intolerable.

So I looked around properly – real research, real exploration, serious solutions built by serious people. And then I dropped all of them and wrote my own, in the early autumn of 2025. I released it on GitHub that December. It has taken more than a hundred commits since, and it is still changing every month. I would not trade it for anything now. It sits in the middle of my toolbox and most of what I do passes through it.

Four stars

It has four stars.

Seven months in the open, the best memory solution I know of, and four stars. I say that with a smile, because there is a real lesson inside it and I would rather have the lesson than the stars.

Part of it is simple. People who polish their own instrument and their own toolbox are unlikely to adopt anybody else’s – open source or not. And I am one of those people, so I understand them completely. The person who would benefit most from what I built is precisely the person who has already built their own.

The larger part is this. There is a serious barrier between a good product and an adopted product, and it stands well before anyone has even noticed that you exist. Getting across it is slow, unglamorous work, and it will take eighty percent of your effort. The product gets what is left.

Coding is solved. I mean that plainly. Writing the code is no longer the hard part of making software. Everything around it is the hard part – what to make, for whom, why they would care, how they would ever hear about it, and what it feels like in their hands on the first day.

That reordering took me a while to accept. It is probably the most useful thing seventy weeks gave me.

---

So why a website, and why stories

Because the moral of all this is small, and it is not about me.

Any one of us can learn. And once learning actually starts, and there is some inner drive behind it, you can go remarkably far – because the tools are all there now, in the open, waiting, cheap, patient with beginners in a way no teacher has time to be.

Anyone can improve their own professional sphere with them. Anyone. The personal sphere too, to a degree – though not in the way people assume. Not psychological consulting. Diet, exercise, scheduling, the ordinary architecture of a week. That is a topic for another article.

And we are in this together. Whatever value each of us manages to produce raises the whole floor – the economy, and, however strange it may sound to say it, our free time and the quality of our days.

So this site is about learning. Learning to use the tools. It will not be only about programming – I use these things for marketing, sales, design, research, productivity, and education in every sphere I touch, and I would like to write about all of it as it comes. The first post here told you how I got to the starting line. This one is what happened after the gun.

An invitation

There is one thing I would like to offer directly, and then I will get out of the way.

I am starting a small learning-by-doing curriculum in coding and IT. A limited group. For children, and for grown-ups who have been meaning to start for years. AI tools throughout, and an individual approach to each person, because at this size there is no reason to teach any other way. Learning by building things that are actually yours and that actually run.

Rather than describe it, here is a sample lesson – a real one, open in your browser, no signup. Edit the page, watch it change, let it check your work.

If that sounds like you, or like your child, write to me. A DM on LinkedIn reaches me, and I answer.

The bots had to do it alone, in the dark, for epochs. We get to do it together, in daylight, and we get to talk to each other while it happens.

That seems like the better arrangement.