Split development work into phases before using Copilot

A developer plans features in stages first, then feeds the approved steps to GitHub Copilot instead of asking it to make all architecture decisions at once.

Done with GitHub Copilot· Intermediate

What they did
They start with a short planning note that divides a feature into separate stages, like API work, middleware setup, and rate limiting. They also run a planning pass in Traycer at the same time to get file-level edits, then inspect and rewrite that plan line by line. Once the plan is approved, they give it to GitHub Copilot with Sonnet 4 to implement the code.
How it went
They say this approach usually gets around 90% of the job done and keeps the code cleaner, while also making reviews simpler and reducing surprise changes across the codebase. They found that fully autonomous runs often burn time on bad design choices, and that the planning step matters less for tiny tasks.
Tips
Use this when a task is large enough to benefit from staged implementation. They also flag cost as a real factor, saying Copilot with Sonnet 4 runs about $10 for roughly 300 requests.

Try it yourself with GitHub Copilot

I’m working on [project or feature]. First, break it into [number] phases with concrete tasks for each phase and show me the plan before coding. Pause after the plan so I can review and approve it. After approval, complete only Phase 1, then stop and report exactly what changed so I can decide whether to continue.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on reddit.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.