You Only Want Seniors. Who Should Train Them?
Automating beginners' work can make sense. Assuming expertise will keep forming on its own makes a lot less.
All essays.
The full archive. Sorted by date, no algorithm, no engagement score: just the things I wanted to say, in the order they came to me.
87 essays · 2019–2026
Tags
Building an AI Project Is No Longer the Problem. Making It Part of Your Company Is.
A survey of 1,050 executives says 74% of organisations already using AI see measurable effects, and only 13% have scaled their projects as planned. I try to explain where projects that worked as pilots get stuck: in the data, the systems, the rules and the organisation that has to take them in. I also declare my interest, because this is what the company I work for does for a living.
A Hundredth of a Second per Target
On Monday the Spanish data protection authority published the first breach notification in which the attack was reportedly carried out by an agent. The interesting part is not that a machine broke into an application. It is that sixteen years ago a researcher worked out how much attention a mass attacker can afford to spend on each target, and the number was a hundredth of a second.
When Does a Company Know Something?
On Friday the form went live through which European manufacturers report exploited vulnerabilities. It contains a field asking for the date and time you became aware, and a twenty-four-hour clock starts from it. It looks like admin. It is the point where the architecture through which an organisation comes to know something becomes a legal matter.
The Cost of Changing Your Mind
For years we looked for digital sovereignty in the provider's passport, the location of the data centre, the software licence. Those are fair questions that miss the decisive one: if we want to change our decision tomorrow, can we actually do it, and at what price? Sovereignty as a real option, and a way to count it.
Redacted
A car stopped dead in a freeway lane at three in the morning, a federal report with three blacked-out fields, two deaths twenty-five days apart, and nobody able to say why. Europe's new Product Liability Directive is criticised from every side as a compensation tool. It does one thing, and that thing matters: it makes not knowing illegitimate.
An Outcome Is Not a Price
Why Italy's B2B market wants outcome-based pricing but hasn't yet built the data, the governance and the trust needed to practice it. And why even the most advanced AI vendors don't really price outcomes: they price the smallest verifiable event marketing can still call an outcome.
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.
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.
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.
87 essays total · 5 featured on the home
← Home