Agentic Development Lifecycle: What the Hype Gets Wrong

“The SDLC is dead, long live the ADLC.” Really? Agents can collapse the build phase to hours, but the lifecycle around it still needs human feedback loops, a cost budget, and a process engineered for your organization.

A reaction to Rakesh Gohel’s LinkedIn post on the agentic development lifecycle:
https://www.linkedin.com/posts/rakeshgohel01_aiagents-agenticai-generativeai-share-7505965828287397890-b60e/

What we cover:

Why intent.md is not the same as clear requirements, and why goals and strategy still come first
Why institutional knowledge won’t “magically” end up in CLAUDE.md files
Continuous evals are useful, but an agent can happily build the wrong thing
What “critical code” should really mean: classify changes by complexity and blast radius, not just by code
The cost of running agents continuously, and what happens with timeouts, rate limits and incomplete responses
Two directions to explore: asynchronous handoffs via Git worktrees, and event-driven agents with emergent behavior

Engineering your AI process is becoming a discipline in itself, and no vendor will hand you the answer.