With the current generation of model, it really isn't necessary to restart every time you don't like something. Certainly this depends on the model. Most of my recent experience is with Claude Sonnet/Opus and Gpt-5.x.

I very often, when reviewing code, think of better abstractions or enhancements and just continue asking for refactors inline. Very very rarely does the model fall off the rails.

I suppose if your unit of work was very large you might have more issues perhaps? Generally though, large units of work have other issues as well.

▲

rdrd 11 hours ago | parent [-]

Yes I too have found newer models (mostly Opus) to be much better at iterative development. With that being said if I have very strong architectural/developmental steer on what I believe the output should be [mostly for production code where I thoroughly review absolute everything] it’s better to have a documented spec with everything covered rather than trying to clean up via an agent conversation. In the team I’m in we keep all plan.mds for a feature, previously before AI tooling we created/revised these plans in Confluence, so to some degree reworking the plan is more an artefact of the previous process and not necessarily a best practice I don’t think.

	▲	XenophileJKO 9 hours ago \| parent [-]
		Understandable. Certainly my style is not applicable to everyone. I tend to "grow" my software more organically. Usually because the more optimal structure isn't evident until you are actually looking at how all the contracts fit together or what dependencies are needed. So adding a lot of plan/documentation just slows me down. I tend to create a very high level plan, then code systems, then document the resulting structure if I need documentation. This works well for very iterative development where I'm changing contracts as I realize the weak point of the current setup. For example, I was using inheritence for specialized payloads in a pipeline, then realized if I wanted to attach policies/behaviours to them as they flow through the pipeline, I was better off just changing the whole thing to a payload with bag of attached aspects. Often those designs are not obvious when making the initial architectural plan. So I approach development using AI in much the same way: Generate code, review, think, request revision, repeat. This really only applies when establishing architecturs though, which is generally the hardest part. Once you have an example, then you can mostly one-shot new instances or minor enhancements.