Charity Majors argues that the rise of AI-generated code doesn't signal the end of engineering discipline — it demands more of it. Drawing parallels to the shift from handcrafted servers to immutable infrastructure, she contends that AI has made code disposable and regenerable, exposing how much knowledge was dangerously locked inside lines of code rather than in explicit specs, tests, and observable systems. She references Chad Fowler's 'Phoenix Architectures' concept, which frames code as a cache of understanding rather than a durable artifact. The real challenge now is encoding intent, behavior, and invariants into evaluable, replayable artifacts — behavioral tests, characterization tests, observability — so that code can be safely regenerated. Human brains are poor validators; the opportunity is to shift rigor upstream into architecture and specs, and downstream into production observability. 2026, she argues, is shaping up to be a return to engineering discipline after 2025's vibe coding wave.
Table of contents
In 2025, the question was whether AI could ever generate “good” codeIt was reasonable to be skeptical the first timeWhat happened in 2025, exactly?“The real product of a software team is shared understanding”Turns out, this is an engineering problem after allThis has all happened before, and this will all happen againDo you remember the sysadmins?Lines of code are not the ideal artifact to reviewOur brains were not built for validationNondeterministic systems will require more engineering discipline, not lessThis is our chance to bring our engineering values to the mainstream46K Impressions2 Comments