AI Coding Agents Erode the Shared Understanding That Holds Codebases Together

Armin Ronacher reframes the Tower of Babel not as a parable of pride but of coordination. In the story, God halts construction not by removing the builders’ tools or knowledge, but by taking away their common language—their ability to combine individual work into something larger. Ronacher argues that large software projects have always been constrained less by how fast one person can write code than by how well a team keeps a shared mental model of the system: what the concepts mean, where the boundaries lie, which invariants matter, and why the architecture has its shape. That model rarely lives in one document; it’s sustained through code review, arguments, and the friction of having to explain a change to someone else.

AI agents strip away that friction. A developer can now add OAuth, caching, or a rebuilt database in a corner of the codebase they’ve never understood, without ever consulting the people who once served as gatekeepers. Each change can compile, pass tests, and come with a plausible on-demand explanation, yet none of the humans involved is forced to acquire the shared understanding the change used to demand. Agents feel no pain and never revolt, so the synchronizing pressure that friction provided simply disappears.

The twist on the biblical version is what makes the argument unsettling. At Babel, losing the common language stopped the tower cold. In AI-assisted engineering, construction continues after shared understanding has already collapsed—the tower keeps rising with no immediate failure to signal the loss. Codebases become Babel not because nobody can communicate, but because nobody needs to, and the absence of a dramatic crash is precisely why the erosion goes unnoticed.