Anthropic's January 2026 RCT found that developers using AI tools scored 17 points lower on post-task skills tests than control groups. The effect has a name: skill atrophy. Developers who accept AI-generated code without deeply reading it lose the ability to explain, debug, and extend it — creating exactly the kind of comprehension debt that compounds into production incidents.
An AI code explainer exists to reverse that process. Given a function, module, or complex code block, it produces a structured explanation: what the code does, what assumptions it makes, what edge cases it handles (and which it doesn't), and what would break it. Not a summary — an explanation detailed enough to let you make a change safely.
GitIntel's explainer is attribution-aware. When explaining AI-generated code, it notes which AI tool likely generated it, since different tools have different known failure patterns. A Claude Code-generated async function with complex error handling needs different scrutiny than a Copilot-generated SQL query. Knowing the origin informs where to look for problems.
Run `gitintel explain <file>:<function>` or pass a code block via stdin. The output is structured: purpose, inputs/outputs, invariants, known risks, and suggested tests.