Flask creator Armin Ronacher warns AI coding agents are eroding the shared understanding that holds large codebases together — the Tower of Babel keeps rising because nobody needs to talk anymore

The Tower Keeps Rising: When AI Agents Make Coordination Optional

Armin Ronacher created Flask, one of Python's most widely used web frameworks. He has spent his career thinking about how large software systems hold together. Now he argues that AI coding agents are quietly pulling that structure apart, and he has a metaphor that makes the problem impossible to unsee.

It starts with the Tower of Babel.

In the biblical story, humanity unites to build a tower to heaven. God stops the project not by taking away their tools or their engineering knowledge, but by taking away their shared language. Construction halts because people can no longer coordinate. Ronacher sees the same dynamic at play in modern software, with one critical twist: AI agents have found a way to keep the tower rising even after the shared language is gone 1.

Here is his argument, stripped to its core.

Large software projects have never been bottlenecked by how fast an individual can write code. They are bottlenecked by how well people can coordinate their understanding of the system they are changing 1. That shared understanding is not English or Python. It is the unwritten consensus about what concepts mean, where boundaries sit, which invariants matter, who owns what, and why the system has the shape it does. Ronacher writes that this language "lives partly in documentation and code, but also in code review, conversations, arguments, and the experience of having to explain a change to somebody else" 1.

Before AI agents, that understanding was maintained by friction. If you wanted to change someone else's storage layer, you had to read their code, ask them questions, and coordinate with teams whose services depended on it. This was slow. Much of that slowness was waste. But not all of it.

"Some of it was the process by which your understanding became mine, and by which both of us discovered whether we still agreed about how the system worked," Ronacher writes 1.

That friction synchronized people. Agents remove it.

One developer asks an agent to add OAuth. Another asks one to add caching. A third asks one to rebuild the database from first principles and, as Ronacher notes, "make the UI pink" 1. Each change can be reasonable in isolation. The code compiles. The tests pass. Explanations can be generated on demand. Nobody needs to talk to anyone else, or even acquire the shared mental model that the change once would have forced them to learn.

This is where the metaphor gets unsettling. At Babel, the loss of common language stopped construction. In AI-assisted engineering, construction continues after shared understanding has already collapsed 1. The absence of immediate failure is precisely what makes the problem invisible. Ronacher puts it in a single line: "The tower does not fall, and so we do not notice what was lost. It just keeps rising." 1.

He adds a line that sharpens the point further: "agents do not feel pain, only humans do" 1. The pain of navigating an unfamiliar codebase, of tracking down the person who wrote a module, of discovering that your mental model is out of date, was not just inconvenience. It was a signal. It told you that the system was drifting, that coordination was breaking down, that something needed attention. Agents absorb that pain on our behalf, and the signal disappears with it.

This is not an "AI is dangerous" argument, and it is not a laziness complaint. Ronacher is clear that agents make individual developers dramatically more capable of changing a codebase 1. The problem is not that agents do their job badly. The problem is that they do their job so well that they eliminate the coordination layer that large-scale software development has always relied on.

The so-what for anyone building software at scale: the feedback loop that once alerted you to architectural drift is being bypassed in real time, right now, across every team that has adopted AI coding agents. You will not feel it happening. The tests will pass. The agents will explain everything. And the first sign that something is wrong will be a production incident that nobody on your team can fully explain, in a codebase where nobody holds the complete picture anymore.

The tower does not fall. It just keeps rising. And the higher it gets, the harder it will be to find anyone who remembers what it was supposed to be.