The plan has to change when the evidence does
o3 and o4-mini arrived on 16 April. Tool results still need to steer the next action.
Retrospective reconstruction of the week beginning 2025-04-21, written on 16 November 2026. This is not a claim of publication at the time.
OpenAI introduced o3 and o4-mini on 16 April. Stronger reasoning is useful, especially when a task turns out to differ from its description.
A repository may have a different structure. An API may reject a field. A browser may show an unexpected account state. The original plan should survive only as long as its assumptions do.
After an action, read the result. If it contradicts the plan, adjust the next step. Repeating the same failed request with a more confident explanation does not count as persistence.
Keep the feedback loop small enough to inspect. One bounded change, one check, then the next change. If several edits happen before any test, diagnosing the failure becomes harder and rollback becomes messier.
The user does not need every internal step narrated. They need to know when something has finished, when the direction changes materially or when a decision is genuinely theirs. Keep the detailed evidence available without turning the conversation into a scrolling execution log.
The plan is allowed to be wrong. The system needs to notice before it keeps building on top of it.