Developers struggle to follow the reasoning behind AI-generated code. A visualization tool could show the evolution of prompts and responses.
The product would record Claude/Codex/Copilot sessions and display them as navigable decision trees with code diffs.
Engineering teams would pay for better insight into their AI tools' behavior during debugging.
MVP could work with a single AI provider's API and basic terminal output.
Risk: AI providers may change their APIs or offer similar features natively.