Your expertise did not become a commodity

My first job was cleaning old computers and installing expired systems in the machine room of the Engineering Centre at UFRJ, in Rio de Janeiro. I paid the bills tutoring maths and physics.

Racks, pliers, floppy disks, solder, and a kind of failure that has a smell.

Then I went into law. Litigation, tax, platform liability for login security — and that is how I fell into language models. Systems for notary offices, a calculator, an MCP, a plugin. I thought that would be it.

Until a story started going around about professionals like you and like me. That our expertise had aged out. That twenty years of learning had turned into a commodity.

The worst part is that we already suspected it before reading it anywhere.

It is a good story. And it is wrong.

Language models are excellent at producing answers and incapable of deciding which ones matter. They cost billions and cannot tell you which document counts. They do not know what is at stake. They do not know when to stop.

I did not learn that by reading about AI. I learned it by testing, by pushing the models to the edge, by following the people who build them.

You are not being replaced. You are being asked for the part that cannot be delegated, and nobody told you which part that is.

I wrote the book that is that list.

Claude for Boomers, Volume I: what you don't delegate. Nineteen chapters, and each one ends by naming something that stays yours. Checking evidence against the source. Saying which document counts. Deciding what may happen without you in the room.

It is free, complete, no sign-up. It is here.

Volume II is being written. I am publishing the whole path here, including what goes wrong.