AI code migration with human-in-the-loop featuring Chris Westerhold from ThoughtWorks
It’s easy to write fresh code but harder to maintain it. Even the most straightforward changes – renaming a function, refactoring a pattern, or updating a deprecated API – can spiral into an endless cycle of debugging and fixing unforeseen issues.
General-purpose coding assistants like GitHub Copilot or Cursor are great for generating snippets or answering questions, but they fall short when applied to large-scale migrations.
They lack:
Deep understanding of your codebase: Off-the-shelf assistants don’t have full visibility into dependencies, tests, or project-wide constraints.
Multi-step reasoning: Code migrations require thoughtful, context-aware modifications, not just single-shot code generation.
Feedback loops for accuracy: To continuously refine their changes based on user feedback, making them better over time.
The solution to code migrations is a human-in-the-loop approach that codifies tribal knowledge, provides valuable context for LLMs with supervised autonomy.
Join Ankit Jain and Chris Westerhold from ThoughtWorks on January 23rd, on building Runbooks and automating code migrations using spec-driven development:
1. Create executable specs - Plan (with AI) to capture intent, constraints, and steps before AI touches code.
2. Version control everything - Specs, AI conversations, and generated changes are all versioned. Fork, improve, roll back.
3. Make it multiplayer - Have multiple engineers collaborate and iterate on the same AI coding session.
4. Build a template library - Create reusable templates for tedious code migrations between frameworks, libraries, or language versions.