How can I reason with an agent about things I don’t understand? What parts do I need to understand, and at what layer?
I’ve been thinking about this a lot. Michael C. Feathers wrote this 22 years ago and it still applies:
Understanding is the key thing that we need to make changes safely.
The question is how much, and which parts. I can drive a car safely without knowing how the engine works. But I should probably keep my eyes open and both hands on the wheel. Or should I say, how can I steer my agent/car with my eyes closed?
Now someone please tell me which part of my code is the street?