Overall yes. I haven't figured out multiple coding agents. The context switching is devastating. But my favorite workflow is claude code plan for 1 task, iterate plan with claude, tell claude to implement, code review, commit, repeat as needed. AI needs super strict guardrails.

Yep, it's like a competent junior developer

Hmm, record a podcast with me? Would be fun to get a measured take on both sides

Let's do it!

Same

I think I travelled from one end of the spectrum to the other, but not sure in which direction

I can’t stand tab completion

Why not? If it's clear what you're doing (refactoring, making repetitive changes, etc.) then the tab completion model can be more helpful than not

overall agree but if you aren't using some agent requests, you're not at peak throughput

cope

Are all these devs actually bottlenecked by their capacity to *write* code? Like the only time I actually use cursor anymore is to do something tedious (“find all occurrences of … and replace”) while I’m in a meeting. And honestly sometimes grep+vim+sed still feels faster

The time wasted to set up the ai, all it's context, check after every output and reroll the dice when failures. Wow

At some point, you realize how much time you are wasting reading garbage code.