Thoughts, stories — and what I'm working on.
One of my family members described a firm that has its data in several different tools, and its employees still join it by hand, in spreadsheets. I realised that this is the ordinary state of firms and that the people in them do not understand the language of the AI echo chamber. Most firms do not even have the basic connectors. Such firms need help with AI transformation.
Employees have the best AI models and the connectors ready at work, and still do not use them. AI adoption fails not for a lack of tools, but because of the inertia of old thinking. Until people stop and ask whether AI could help, they keep doing by hand what a machine would do in minutes.
I gave the same audit to two AI models. The stronger and pricier Fable 5 needed nine rounds of checking and burnt 2.78 million tokens. Cheaper Opus 4.8 did it in a single round for 721,000, with around ninety per cent agreement. The model did not decide. The prompt did. Fact-checking belongs at the input, not the output, and anyone can put that rule into the brief.
The internet is not one place, it has layers. The public web is now flooded with content from machines, the so-called dead internet and dark forest, and real people have moved on: some behind a gate into the cozy web, some to quiet personal blogs you follow through RSS. What those layers are and how to reach them through an RSS feed, a reader and one OPML file, shown on my own page, the Human…
Walking through the dead internet, I sometimes come across something alive. Most of what reaches us today was written, ranked or summarised by a machine. Maggie Appleton saw it coming years ago: an open web turning into a dark forest, so flooded with generated content that real people retreat into smaller, trusted spaces to find one another. This led me to create a new simple page on my site,…
I installed a small local AI model on my laptop. In near total darkness it recognised what was in front of the camera. That such things exist, I knew. What surprised me was the accessibility. Text is only the first layer. The barrier to automation is not whether something runs on a computer, but whether it has a computer inside it.
Bots and AI agents now form the majority of web traffic, not the minority. My first bot detection relied on a list of known robots and a missing browser language. But scrolling and clicking no longer prove a human. So I moved to five groups of visits, including the unverified ones I count on neither side. The 80% human figure that most analytics tools show probably no longer holds.
On the question of how to blog in the age of AI, I answer for myself. Being a secondary source, retelling what has already been published, commenting on it superficially, does not interest me. What interests me is being a primary source: to do something myself, to find something out, to have a conversation, and to write about it. A unique experience, combined with expertise in the subject, is…
On the evening of 19 June 2026 I sat on the terrace with two things. Heidegger's Being and Time, and my notebook. I wrote questions, not answers. I went back to Hegel's master and slave, which I last read while studying philosophy at Charles University. The machine works for us, but it cannot recognise us. It only imitates recognition. We hand over the work that forms us, and gain a comfort that…
Two large consulting firms handed over reports with fabricated sources and quotes from AI that no one read back. The problem is not AI, but the size of corporations, at which poor work cannot easily be policed. A person who works for themselves and signs with their own name has it easier.
Seventy per cent. That is where the first AI output begins, even with full company context and examples. Getting higher is hard: either the AI lacks good enough context, or it lacks the tools, or it cannot check its own output. The owner of the company is not satisfied with it. Someone has to sit at that computer and supply the company context to the AI. For now, until the AI can do it on its own.
Most of what we see on the internet is, or soon will be, AI-generated: text, images, profiles and comments. The internet is turning into an online game full of bots, where the certainty that you are talking to a human disappears. At the same time, automation is taking away tens of percent of routine work at the computer. So one thing stays valuable: real human experience, a primary source that AI…
One senior marketer rejected AI after a single bad experience. Other professionals automate everything they can and never go back to finished work. They work in the same field, yet their working days look completely different. What decides it is not skill, but whether the work becomes a foundation for the next day or is done again from scratch. That is how the gap between professionals, and…
Europe does not have the capacity to face a full-scale, mass drone war of the kind we see in Ukraine. Three dependencies weaken it: China supplies the material for defence systems, the United States supplies the military capabilities, and the member states themselves cannot agree how fast, or who pays. Plans such as ReArm Europe and EDDI exist, but they move slowly.
AI now produces the graphic, the newsletter and the product page. What remains for the person is judgement over the result, knowing whether the output is good in the context of the company. But most people have worse judgement than AI, and whoever cannot judge quality cannot delegate either. AI can be taught a company's judgement, but only under specific conditions, and even then that judgement…
I decompose the price of an index or sector fund into six components: book value, earnings premium, macro contribution, event contribution, media attention, and the arithmetic remainder. I teach it to my AI prediction system: first it has to build, by working on the past, a view on which a forecast can rest. Looking back at my own best entries, the chart showed they sat in a phase that today reads…
If most routine work on a computer disappears in the next ten years, and a large share of repetitive manual work disappears with it, what happens to the flow of money? Who pays whom for what? A six-layer map of the coming economy (capital, tech service, local service, UBI, discarded, outside) and the hybrid strategy that gives a single person the best odds in any of them.
Four days in Catalonia. No computer, no AI. I bought a notebook and wrote everything by hand. Those pages convinced me the middle path has stopped working. Now I live in two extremes: everything through AI agents, or entirely offline. This is built from the notebook.
I am building an AI stock market prediction system that runs on my own machine, uses free public data, and grades its own forecasts against reality. After three years of tracking market moods in a spreadsheet, I am engineering the calibrated version. This series documents the build itself. What I will eventually share beyond the writing is a decision I will make as the build matures.
I am teaching an AI agent to forecast incoming orders and revenue. In one evening I fought through nine rounds: profile multipliers were lying, ad attribution overlooked the consent gap, on-site visibility motivated all visitors, keyword matching credited the wrong campaign, subtasks held different deadlines, switching off a campaign in scenario mode did not subtract the full size, new markets…