Introducing Bridgekeeper

In “Programming as Theory Building”, Peter Naur argues that the purpose of programming is to inculcate a shared mental model of a real-world problem; programming is not merely needed to produce code.

AI coding assistants have greatly decreased the cost of producing code. They have not decreased the cost of understanding it. “Cognitive debt” describes the productivity costs that accumulate when teams defer understanding. Margaret Storey describes the consequences of cognitive debt: velocity without understanding is unsustainable.

Bridgekeeper mitigates cognitive debt. It identifies the most salient changes in a diff, asks the author and reviewers to predict how they work, and judges whether their explanations demonstrate a mental model of the code.

We were inspired by research in pedagogy and psychology.

Retrieval Practice

Asking a reviewer to predict what a function does is more effective than showing them the answer. Practicing recall strengthens memory more than re-reading.

The Generation Effect

Information you produce yourself is retained better than information you receive passively. Explaining a change in your own words builds a stronger mental model than approving it.

Metacognitive Calibration

People are bad at judging what they know. A Socratic prompt forces calibration: you cannot bluff your way past a follow-up question that targets the exact ambiguity you skipped.

Private Beta

We are running a private beta with a handful of teams. If you are shipping AI-assisted code and feel the cognitive debt accumulating, request access from the contact form.