Understanding debt is a form of technical debt specific to AI-assisted development: the gap between the code a team has shipped and the code its engineers actually understand. Where classic technical debt lives in messy structure, understanding debt lives in missing knowledge — code that works, passes tests, and is owned by no one who can explain it.
It accrues quietly. An agent produces a working diff, it merges under a review that never really engaged with it, and the team's shipped surface area grows faster than its comprehension. Nothing looks wrong until reasoning about that code becomes necessary — an incident at 2am, a security finding, a refactor that has to touch it — exactly when the missing understanding is most expensive to recover.
Like financial debt, some of it is a reasonable trade: prototypes and throwaway scripts do not need to be understood. It turns toxic when it settles into production systems that must be maintained and secured. Paying it down means making comprehension a merge-time gate rather than an afterthought — confirming someone understands a change before the team owns it.