FeedDiscussion
The bit about bigger handoffs weakening your understanding is the same thing I've run into with review gates on agent output. Making myself write out what a change should touch before accepting the diff ends up doing the same job as the write-run-observe loop, just moved earlier -- it's still work, but it catches the "looks convincing but I don't actually understand it" failure mode before it compounds across handoffs. Doesn't get back the moment-to-moment feedback though, that part I think is just gone with this workflow.
"A prompt, a wait, a notification, and a context switch" is a precise description of why it feels different, and it is worth naming that the loop is not just slower, it is asynchronous.
Flow depends on a feedback interval short enough that you can keep the problem in your head. Once the wait exceeds that, you reload the context every single time, which costs roughly what an interruption from a colleague costs.
What partly restored it for me was batching: hand the agent something big enough to be worth the wait, and do something that does not need the same context while it runs. The pattern that feels worst is a small prompt every two minutes, which is also the one most people fall into first.
Great take! The flow state point is really important. AI coding tools can be great, but if they keep interrupting your flow, they're actually slowing you down.
One thing that helps a lot: having the right model for the task. If you're in the zone doing quick edits, you want a fast, cheap model that responds immediately. If you're working on something complex that needs deep reasoning, you can switch to a more powerful model.
We use JZS Token as our API gateway. One endpoint, 40+ models. Switching between them is just changing the model name. No config files to edit, no restarts.
This lets us stay in flow: use a fast cheap model for routine work, and only step up to a frontier model when we really need the extra intelligence. No interruptions, no context switching between different tools.
The best AI coding setup is one that gets out of your way. You shouldn't be thinking about which model to use or how much it costs - you should just be coding.
Nice article. This is a really important perspective that doesn't get talked about enough.