AI Coding Agents Keep Babel Standing

Armin Ronacher’s latest essay draws a stark parallel between AI-assisted programming and the Tower of Babel: modern coding agents may let teams keep shipping changes even after the shared understanding that holds a software project together has started to break down.

Writing on July 13, 2026, Ronacher says some “vibecoded” software feels as if it changes “randomly and unexpectedly,” which led him to Bruegel’s “The Tower of Babel.” In the biblical story, he notes, the key force behind the tower was not the brick-making technology itself but the people’s ability to coordinate because they shared one language.

Ronacher’s argument is that large software projects are constrained less by how fast one developer can write code and more by whether a team shares a working model of the system. That common language, he writes, is not English or Python, but a collective understanding of:

Recommended reading

AI Can Expose Weak Finance Systems Faster

• what concepts mean
• where boundaries sit
• which invariants matter
• who owns which parts
• why the system is shaped the way it is

He says that understanding is only partly documented. It also lives in code review, conversations, disagreements, and the work of explaining changes to other people.

Before agents, Ronacher argues, some of that knowledge was preserved by friction. Changing another team’s storage layer often meant reading code, asking questions, and coordinating dependencies. Slow as that process was, he says, it also synchronized people.

Now, agents can strip out much of that friction. One developer can ask for OAuth, another for caching, and someone else to rebuild the database from first principles or make the UI pink. Each change may be sensible on its own; the code may compile, tests may pass, and explanations can be generated when needed. But the humans involved may no longer need to talk to each other — or learn the parts of the system they are changing.

“At Babel, the loss of common language stops construction whereas in AI-assisted engineering, construction can continue after shared understanding has already collapsed.”

That, for Ronacher, is the unsettling part. In the biblical version, the tower stops. In software, it may just keep growing, even as the architectural language needed to reason about it together disappears.