Agentic SDLC: AI-generated code clogs pipelines until human gates slow it back down
Pavan_Belagatti · x · 2026-11-13
Pavan Belagatti argues the developers at risk today are those who can't work with agents and workflows. AI code generation became so fast that reviews queued up, making shipping slower — the industry only noticed after assuming AI = faster.
Teams then pushed agents across delivery, adding risk, and responded by placing human gates at critical stages to approve or reject agent output.
The accompanying diagram lays out a full Agentic Software Delivery Pipeline: Coding Agent (generate/open PR) → QA Agent (tests) → Security Agent (SAST/deps/secrets) → Review Agent (first-pass PR review) → Policy Agent (merge gate) → Release Agent (feature flags, rollback) → SRE Agent (SLOs/incidents), with an auto-retry loop back to the agent until tests and scans pass; developers, code owners, release managers and on-call engineers sit at the checkpoints. Takeaway: agentic SDLC works best with agents and humans together in the pipeline.
Related event: Agentic SDLC: Human Gates Needed as AI Floods the Dev Pipeline(2 posts)→
More from coding & agent
- Meta's Muse can now make phone calls for you, rolling out to US paid subs — alexandr_wang · 2026-11-13
- Agent UX for hours-long autonomous work remains unsolved, dev argues — andreisavu · 2026-11-13
- Instrument parser agent: show it a few sample files and it builds your lab data pipeline — nlarusstone · 2026-11-13
- Programmable demo recording via Claude tested, coming to a browser near you — smtabatabaie · 2026-11-13
- Mercor launches OWE-Bench: 26 real engineering projects that stump coding agents — ben_burtenshaw · 2026-11-13
- Source, Sink, Drain: an IBM engineer's blueprint for running an AI software factory — TejasKumar_ · 2026-11-13