Focus of Letter 41

It’s through one of its features that I started to test Arena this week: Agent Mode. You give an instruction, it goes through the steps alone, text, search, code… and at no time do you know which model did what. No choice possible, no name revealed at the end.

At the same time, Is it Fake? seems to do the opposite: we give it a URL, it returns a trust score out of 100, with sources and a status for each statement. This score, too, comes from a mix of models (Gemini, GPT, Claude) that finally sorted for us.

A tool that acts without us looking, another that judges without us looking. The movement is the same. We’ve already crossed this movement with Echo, the game I created on agency (letter 30): it’s never a big decision to “delegate everything”, it’s a succession of little “I let it pass”.

This little moment when we let it pass is what I wanted to make visible this week, with a tool that has been taken over and enlarged: the one to help you choose generative AI applications made last year. It’s an interactive decision tree, a bit like a reflexive todo-list to ask questions before and not after the choice. It’s a bit like what I do before each publication on the site.

Four dimensions, partly inspired by the work of André Tricot: conformity, usefulness, usability, acceptability. Three paths depending on the context, personal, professional, or with students. Sentences such as “have I checked that this use complies with the rules of my school?” or “does this tool really bring me something that I didn’t know how to do?”. At the end, a score per dimension and a list of points of vigilance. The results can be exported and two tools can be compared with each other.

It’s not a tool to clear one’s conscience before choosing or to not choose. I think it’s a grid to slow down a notch, the time to ask ourselves the questions that we tend to skip.

And maybe that’s the question: do we still have the reflex to ask ourselves them without a grid? Do we really need a tool to remember that we can and must choose?

Echoing

 → Letter 30 on Echo and Agency

 → letter 33 on Holotab and the choice of the autonomy level