Home / All essays / Tags / AI Agents

AI Agents

Agents produce more than a human can read: ex-ante governance doesn't hold, and the open question is which level cannot be delegated.

I use agents every day to write code, which is exactly why I don’t believe in governance that assumes human time to read everything that ships. That time is gone. The question is not whether to trust them but where to put the control: in the instruction file that grows past the threshold where the agent can hold it, in the floor below that nobody reviews, in the level of judgment that cannot be delegated. The history of automation proves the optimists right only in aggregate. I write about what happens in the detail.

29.08 2026
№ 91

AI AI Agents Digital Sovereignty Software Development Politics of Technology

You Can't Export the Function

On August 28, OpenAI announced it will cut Cursor's access to its models: a contractual clause between third parties, and a function thousands of teams used every day disappears from their environment. No data lost, no API broken. After data and memory, lock-in moves once again: into behavior no export contains.

18′ reading time

4,061 words

Read →

28.08 2026
№ 90

AI AI Agents Compliance Cybersecurity EU Regulation

Authority Arrives Before Intelligence

We describe agents in a technical vocabulary, tool calling, memory, sandboxes. But an agent is not interesting because it does things on its own: it is interesting because someone granted it the right to do things that have consequences. And rights, unlike intelligence, are not acquired. They are delegated. With everything that word carries.

19′ reading time

4,239 words

Read →

27.08 2026
№ 89

AI AI Agents Software Development Work Digital Transformation

Measure What Stays Scarce

Inside Meta, code changes grew 220% while features reaching users grew 36%, and incidents climbed. That is not proof that AI doesn't work: it is the sign that our metrics were counting a scarcity that no longer exists. The first AI-native crisis won't be a technological collapse. It will be a dashboard full of excellent numbers.

18′ reading time

4,023 words

Read →

26.08 2026
№ 88

AI AI Agents Digital Sovereignty Software Development SME Work

The Moat Is Not the Data

Thomson Reuters built its model using less than ten percent of its own content: the hard part was not the corpus, it was deciding what learning correctly from it should mean. Perhaps the advantage of vertical AI is not the data others lack, but the judgment criteria others have not yet had enough experience to build.

16′ reading time

3,599 words

Read →

27.07 2026
№ 81

AI AI Agents Software Development Compliance CRA Open Source

The Floor Below

Around the third month of using an agent, someone says the model has got worse. Almost always it isn't the model: it's the scaffolding, which came apart without throwing an exception. What degrades quietly is exactly what nobody is required to check, and in Europe, from September, that friction costs a non-conformity.

9′ reading time

1,816 words

Read →

24.07 2026
№ 80

AI AI Agents Work AI Act Compliance EU Regulation

The Level You Can't Delegate

ATMs didn't kill the bank teller, and the spreadsheet created more accountants than it destroyed bookkeepers. The fastest typist in the office disappeared anyway. Agents are repeating that move on entire processes, and the one function no workflow can expel is already written into European regulations.

1,893 words

Who Controls What AI Agents Produce?

There's a question that has been turning in my head for weeks—the kind that arrives at eleven at night when you're going back over everything that was produced during the day. The question is simple, almost banal: how do we govern what we can no longer read?

8′ reading time

1,790 words

Read →

← All tags