Values
Intent before implementation
People own purpose and judgment. Agents own the doing.
People steer the system. They decide what is worth doing and judge whether they got it. They work at the level of intent, and they check the result at the same level they specified it. People drive this work, and nobody runs a factory in the dark.
A unit of work can be no bigger than a person can understand, define and assess. So your job is to raise the level of abstraction, until the same human intent covers more work.
The trap
You drop down a level to read the code, review the diff or adjust the method. It feels responsible, and it turns you into the agent’s reviewer.
When implementation is right
- You’re debugging the machine. Read the code when no other mechanism can detect a problem, and treat that as the exception.
Where it shows in the grid
“Hand over goals, not tasks, as big as you can assess,” “Humans spend their time on problems and experiments, not on reviewing output” and “No human reads or writes code.”