Martin Fowler & Kent Beck: Frameworks for reinventing software, again and again

When asked what ideas resonated most over a quarter-century, Fowler cited refactoring as a steady anchor, while Beck offered a blunter picture: «I get thank you so much for test-driven development. I also get this: test-driven development ruined my life. My dog left me, my house burned down, and it's all your fault.» The divisiveness of TDD has always been part of its design—practices meant to provoke and push boundaries rarely win universal love.

Yet an AI researcher recently told Beck that 20 years of TDD advocacy now pays dividends: with powerful, unpredictable agents in the loop, the discipline of verification becomes non-negotiable. Fowler agrees: «When we've got a big powerful genie, you really have to learn how to verify that it's doing the right thing for you, which we've been practicing for 25 years.» The irony is that the craft many dismissed as over-engineering has become the guardrail for an industry handing more control to machines.

Fowler and Beck have weathered transformative waves before—object-oriented languages in the '80s and '90s, the rise of the internet, and the microprocessor revolution that Beck witnessed as a Silicon Valley kid. When the Intel 4004 hit, suddenly a computer wasn't a mortgage-sized box; it was a chip anyone could experiment with. That explosion of possibility mirrors today's AI moment. Beck recalls, «If you can figure out how to write software, if you can figure out how to design hardware around this thing, you can suddenly do things we can't even imagine.»

But every shift brought hype, snake oil, and casualties. Object-orientation scared many; the internet had skeptics (including some who doubted its importance). Fowler learned to be «skeptical about my skepticism,» requiring curiosity to probe whether claims hold up. His advice: run the smallest experiment that satisfies your own bar for evidence, and recognize that your first interaction may not signal truth. When Fowler tried GitHub Copilot in Emacs a year and a half ago, he dismissed it as garbage—until colleagues like Simon Willison demonstrated sustained, skillful use. That taught him not to flip the «bozo switch» prematurely.

Beck worries about the «resoloing of programming.» Extreme Programming created safe social environments for fundamentally antisocial people, fostering hours of daily conversation. Now, the fantasy is one programmer managing six agents—«really I'm managing a team.» Beck pushes back: «No, you're not. You're using six tools at once.» Agents don't bring differing beliefs, energy levels, or the friction that produces insight.

Fowler and Beck both see promise in pairing two humans with multiple agents. Fowler notes, «If it's two of us, we can control the genies perhaps a little bit better. And we also have that same interaction.» Beck has found pairing with genies productive precisely because models are slow—«every time the models come out and they're faster, I'm like, 'Oh, there's less time to talk.'» While a prompt runs for three minutes, pairs discuss naming philosophy, conditional structure, and strategy. If the response returns in 15 seconds, that conversation vanishes. The social, conversational core of programming may be what separates effective teams from isolated operators with fancy autocomplete.