What the old masters knew about new tools

Every powerful new tool arrives disguised as a new kind of person.

The stethoscope was going to make everyone a diagnostician. It only made the careful ones faster and the careless ones more confident. People mistook the reach of the instrument for the judgment of the hand that held it, and the confusion passed, and the work stayed hard.

Homo faber, the maker. Older than priest or king, and it has outlived every tool we ever picked up. The flint outlived the hand that knapped it, but the hand is what mattered. The maker is the constant; the tool is the variable.

No tool confers mastery. It only amplifies the judgment already there. Hand a perfect chisel to someone with nothing to say and you get faster nonsense. The material answers only to the question you bring it, so the master’s real skill is not the tool at all. It is knowing when not to reach for it. From outside, that restraint looks like slowness. It is the economy of someone who has already made the mistake you are about to make.

I can show you what restraint looks like in my own work. At the energy exchange I run the Polish model Bielik locally, reading documents where they sit and stripping the personal data before it can travel. The large model, PLLuM 70B, waits until a small one has plainly failed. Most of the discipline is in not reaching. The apprentice grabs the biggest tool because grabbing feels like skill. The master leaves it on the bench.

I learned to see tools this way in a strange place. In 2018, at the Copernicus Science Centre, I produced a festival called the Civilization of Algorithms and put the new tool on a stage beside philosophers and dancers. Next to a body that moves, an algorithm has to be described in ordinary words, and a thing you are forced to describe in words stops being magic. It becomes a set of choices, made by someone, for a reason you can name.

So when everyone began reaching for agents, the newest tool, I did not write a manifesto about them. I built Age of Agents and watched them play, rendered as a live pixel kingdom. You watch them do the wrong thing with total confidence, and you finally understand what you are dealing with. You apprentice yourself to a material by watching how it behaves, not by trusting what it promises.

I have made AI decide about blood and route clean power to open land. The tool changed completely between those two jobs. The questions did not. The tool serves the judgment, never the reverse.

The tool in your hand is new. What it asks of you is not: learn the material first, and only then earn the right to trust the machine.