Most software rituals were never about quality. They were patches for human limits: forgetting, fatigue, handoffs to people who missed the original meeting. AI removes the limits those rituals compensated for, which means the rituals themselves need to be reexamined rather than just sped up.
The Details:
Code review, staged QA, and detailed documentation exist because the person who wrote something is the worst judge of its flaws, and because context decays between people and over time. AI does not forget context and does not tire, so parts of that scaffolding lose their original purpose. That does not make them useless, but it changes what they are protecting against.
The real bottleneck moves upstream, from writing code to specifying what should be built. An AI can produce a clean, tested, working implementation of the wrong idea just as fast as the right one. The cost of building something nobody needed drops close to zero in effort but stays identical in consequence, which makes requirements and architecture decisions carry more weight, not less.
AI tends to write tests that confirm its own code rather than probe it for failure, so passing tests stop being reliable evidence of correctness. The habit of adversarial, human skepticism toward polished output becomes one of the few things automation does not replicate well.
Speed should fund exploration, not shortcut it. When a prototype takes minutes instead of days, teams can test three approaches against the real problem instead of debating tradeoffs in a meeting. Writing a short statement of intent before generating anything forces that thinking to happen instead of being skipped in favor of a clean-looking diff.
Bottom Line: The typing got cheap. The judgment about what deserves to be typed did not, and that is where the real engineering work now lives.
Enjoy this article?
Listen to the Claude Code Conversations radio show or join the community.
