Google's Adam Bender on the Future of Software Engineering

Adam Bender, a Principal Software Engineer at Google with nearly fourteen years at the company, argues the AI coding debate is asking the wrong question. Whether a model writes good code and ships fast misses the deeper split between programming and engineering: programming is a solo act, while engineering is what keeps code integrated, maintainable, and economically viable across many people over years. AI has accelerated the former but barely touched the latter.

Drawing on his work leading a Large-Scale Change across Google's entire codebase — refactoring TODOs across millions of lines he'll likely never read — Bender explains why developer ecosystems behave like complex adaptive systems, where a change in one node produces effects nobody predicted. He warns that copying Google's technical processes without absorbing the cultural assumptions behind them, like the trust underlying Large-Scale Changes, is doomed to fail.

For engineering teams, the practical stakes are concrete: as AI multiplies code output, testing, review, and intellectual control over the system don't automatically scale with it. Bender notes agents are strong at unit tests but weak on integration testing, that internal APIs must now be hardened like public ones, and that teaching new engineers a decade of judgment in months remains an unsolved problem.