The workflow that works: explore, plan, implement, commit

If you adopt one workflow from this course, make it this one, adapted from Anthropic's Claude Code best practices (whose own phrase is "explore, plan, code, commit") and applicable in any tool: Explore → Plan → Implement → Commit. The most common failure with AI coding isn't bad code — it's the AI confidently solving the wrong problem because it jumped straight to implementation. Separating the phases prevents that.

Explore. Before any code, have the AI (or yourself) understand the relevant code and constraints. Point it at the files involved, ask it to explain the current behavior, identify the patterns to follow. This grounds the work in reality instead of assumptions.

Plan. Have the model produce a plan before it writes code — the approach, the files it'll touch, the edge cases. Read the plan and correct it. This is where you catch "it's about to solve the wrong problem" cheaply, before any code exists. Most tools have a plan mode for exactly this. The one exception, per Anthropic: if you could describe the diff in one sentence, skip the plan — trivial changes don't need ceremony.

Implement. Now let it write the code, ideally in small, reviewable increments rather than one giant dump. Watch it work and course-correct early.

Commit. Get it to working, verified, committed state — with a clear message — before moving on. Small committed increments are easy to review and easy to roll back.

The reason this beats "just ask it to build the thing" is that it inserts human checkpoints at the two moments that matter: after exploration (is it understanding the right things?) and after planning (is it about to do the right thing?). Correcting a plan costs a sentence; correcting a finished 500-line implementation costs an hour.

One more habit that pairs with it: course-correct early. Tight feedback loops beat long sessions. If the AI is off after a couple of corrections, don't keep piling on instructions — reset the context and rewrite your prompt from scratch. A fresh, well-framed start almost always beats wrestling a derailed session.

This workflow is the backbone. The rest of Module 2 — rules files, spec-driven development, verification — are refinements of it. But even on its own, "explore, plan, implement, commit" turns AI coding from a slot machine into a controllable process. Build the habit now.