How AI is Shrinking the SDLC: in greenfield, brownfield, regulated industries

How AI is Shrinking the SDLC I work with experimental AI-first teams, exploring how agentic engineering impacts Lead Time. Here's what I'm seeing. And unlike some people say, I think that SDLC is not killed by agents. I think it compresses into something more lightweight. One person with AI can generate what used to require a team. The bottleneck shifts from writing code to validating it. But this isn't uniform across all contexts. Greenfield, brownfield, and regulated environments each compress differently. Scenario 1: Greenfield / MVP / Internal Tools Context: New project, no users, low error cost, speed is critical. What changes: Tiered Code Review: security-critical code (auth, crypto) — 100% human review; everything else — automated checks + spot-check Observability as primary safety net (canary releases, auto-rollback) Iterations are now significantly faster, to the point where customer gets updates during the demo Important: even in greenfield, AI code contains 1.7x more issues