alt Hacker NewsA bit late to this party but what the heck.
I too have started using plan mode less, mainly for two reasons: 1. I plan my prompt more carefully and think about architecture up front 2. I found with more recent powerful models, the clarifying questions were generally not useful because it was pointing out issues it obviously knew the answer to and would have resolved in implementation anyway. So essentially they became time-wasting and anxiety inducing for no good reason
The mental shift I made was to not accept poor understanding of a code base on my part when writing the prompt - if I don’t understand it, I can’t predict what assumptions the model will make even on a basic level.
I used to tell myself that plan mode mitigated that, but I usually ended up mentally glossing over the generated plans anyway.
That just resulted in pure anxiety-driven engineering, where I’d often spend extra cycles verifying what was built and worrying about the design.
So invest the time understanding the system, at an appropriate level. That level will change over time as models get better.
+1 to a lot of what you said here, especially this:
> I used to tell myself that plan mode mitigated that, but I usually ended up mentally glossing over the generated plans anyway.
At a certain point I realized I far preferred delegating the first implementation pass to agents and reviewing the _code_ instead of a plan up front. If you had asked me this question ~2 years ago I'd think anyone would be insane to hand off discretion like this, but frontier models are just so _good_ nowadays.
Let's say you launch 10 working sessions in a day, would you rather:
* Review 10 plans, and _then_ review 10 PRs, and _then_ maybe adjust the approach on 1-2 and merge the other 8
* Review 10 PRs, and maybe adjust the approach on maybe 4 and merge the other 6
The first option just feels like needless attention for the sake of feeling in control; but the reality is that these models are becoming just as good if not better as us humans and our reckoning is here. Our codebase has enough linters, hooks, and guards such that generally the agent just follows our blessed patterns already, so what is plan mode actually doing beyond giving me a false sense of security? I'm still going to read the code anyways.