GitHub Copilot Attempt at Sprint Planning. AI vs Human Agile Estimates

Описание: I test whether GitHub Copilot in Visual Studio 2026 can generate a realistic Agile sprint plan for a real-world .NET project.

I give Copilot an existing legacy codebase and ask it to:
Perform a code review
Create Scrum sprint plans
Estimate effort and timelines

The results are surprising.

While the AI understands Agile concepts like Definition of Done, backlogs, and iterations, the actual sprint plans reveal major issues:

Waterfall-style sequencing
Poor effort estimation
Lack of working software early in the process
Weak understanding of real-world development constraints

This experiment highlights an important truth:
AI can generate structured plans, but real Agile planning still requires human experience, context, and collaboration.

I also compare different models, including ChatGPT Codex Mini vs full Codex, and break down where each one succeeds — and fails.

If you're using AI tools like Copilot for development, this is something you need to understand.