The pioneer of Extreme Programming and coauthor of the Agile Manifesto argues AI has moved the bottleneck from writing code to directing and reviewing it — the part of the work the profession spent decades selecting against — and engineers were never formally trained for it.
Kent Beck, the engineer who helped coin Extreme Programming and coauthor the Agile Manifesto, has spent four decades defining how software gets built. In a recent appearance on The Pragmatic Engineer podcast, he argued the profession is now facing its strangest correction yet: AI has started to do the writing, and what is left is the part engineers were never formally trained to do.
The mechanism is simple to state and hard to absorb. As AI coding assistants like Claude Code and GitHub Copilot generate a larger share of working code, engineers spend more of their day reviewing, redirecting, and stitching AI output. The practice, increasingly called vibecoding, is collapsing the old boundary between writing software and deciding what should be built. The bottleneck is no longer typing. What is left is judgment.
Beck, who has been writing about AI coding agents since at least mid-2025, framed the resulting pressure on the profession in unusually direct terms. Engineers, he said, "are kind of assholes, sometimes". He meant it as a self-critical diagnosis, listing weak emotional regulation, lower natural empathy, and a tendency to be more direct than peers can absorb. These are, in his words, "hideous qualities" the industry needs to confront.
The diagnosis lands harder because Beck is not an outsider taking shots. He is a founder of the discipline. He helped write the Agile Manifesto in 2001 and created JUnit, the testing framework that became the default unit-testing library for Java and shaped how an entire generation learned to verify their code. When he says the profession built its identity around the wrong skill, the claim comes from inside the discipline.
The macro picture supports the bottleneck shift, even if it does not validate Beck's specific list of "hideous qualities." PwC's 2026 AI Jobs Barometer tracks how AI exposure is reshaping demand across knowledge-work roles, including software engineering. Stack Overflow's December 2025 developer survey analysis found that AI has measurably changed the entry-level pathway for junior developers, the rung of the career where pure code fluency once served as the main filter. Both pieces of evidence point in the same direction Beck is describing: the rate-limiting skill at the keyboard is changing.
There is a second, related argument Beck has been making in interviews that cuts even harder against easy reassurance. As paraphrased by Biggo, he has said that "nobody knows" how to build software in the AI era, which is not a complaint about any individual engineer but a practitioner-side acknowledgment that the playbook has not been written yet. The same observation shows up in his Pragmatic Engineer newsletter profile, where his career is read as a long series of bets on practices the rest of the industry would only adopt later.
What Beck is really asking the profession to confront is what he calls a "cosmic practical joke": the industry selected for and celebrated technical brilliance, then quietly depended on a set of human skills like empathy, patience, and the ability to absorb disagreement that the same selection process filtered out. AI did not introduce that contradiction. It just removed the code-writing work that used to hide it.
For engineers who have already invested in product judgment and clear communication, the reframe is an opening, not a threat. For hiring managers, the practical question is whether interviews designed to test code fluency are still measuring the right thing, and whether the junior pipeline that once filtered on typing speed can be redesigned around the parts AI cannot do. For the profession at large, the watch item is whether the industry will update its identity to match the work AI has left behind, or keep selecting for the wrong skill and call the gap a talent shortage.