Watch a junior developer learn in 2019, then watch one in 2026.
2019: they guess. They break staging. They feel it in their stomach. They learn and never break it that way again.
2026: they ask. It answers. Staging is fine. And nothing was learned, but data was gone. TO a third party. For free.
Same job title and pay band. A completely different amount of human being at the end of it.
The name was the tell
We call them decision models now. Not writing models. Not search models. Decision.
The output is not really text. The output is your next choice, made before you make it.
The ladder lost its bottom rung
Every senior you respect was assembled out of a thousand small bad calls they survived. The small calls were never the job. They were the training set for the big calls.
A decision model takes the small calls first, because those are the easy ones to automate. Take the small calls away and you do not remove work. You remove the factory that produces people who can decide anything.
In ten years the shortage will not be jobs. It will be anyone who has ever been wrong enough to be right later.
We asked for a tutor and installed an oracle
A tutor asks what you think, then waits. An oracle tells you what is true, then leaves.
One builds a mind. The other rents you one.
Sterile rooms make fragile people
Wrong calls work like germs. Small exposures build the antibodies.
A working life where every call arrives already cleared by a model is a sterile room. It feels safe for years. Then something real walks in, and there is nothing in your system that knows how to fight.
I live in the sterile room
I direct AI agents for a living. Very few people stand in this room all day more than me.
So I now take small risky calls on purpose, like a vaccine. A little dead germ, on my terms, so the immune system remembers its job.
The angry version came first
I wrote about this in September. That one was angrier and ended in doom. It is here... I mean, here.
This is the same alarm with the panic burned off, plus the idea I left out then: the ladder.
Leave the bottom rung in place
Give juniors the small calls and let them feel the results. That is not hazing. That is the curriculum.
And keep a short private list of decisions the model is not allowed to touch. Add to it. Never shrink it.
One question
What is one decision at your job that no model is allowed to make?
I will go first in the comments. Mine is... In the comment section. BTW, I read AND REPLY to every comment.
![Physical AI Foundations and Definitions — Ch.00: Introduction [KR]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fzz2y7pn0xi5qj511e402.png)
![Physical AI Foundations and Definitions — Ch.09: Standards and Definitions [EN]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fw4mzb4ntt5lhj7vgf333.png)
![Physical AI Foundations and Definitions — Ch.06: Relation to Embodied AI and Robotics [EN]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fshqcmgwjgqqejta2saya.png)
![Physical AI Foundations and Definitions — Ch.03: History and Origins [EN]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Frrm68fq83d8whnovwor0.png)
![Physical AI Foundations and Definitions — Ch.02: Physical AI vs Digital AI [EN]](https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fsgutmwrbqzyn5a22lilm.png)