Agentic CI/CD

Agentic CI/CD adapts continuous integration and delivery practices to AI-agent work, where prompts, tools, traces, evals, saved state, review gates, and deployment checks all shape whether agent output can ship.

Key points

• The playlist's CI/CD talk argues that solo agent builders often recreate weaker versions of established delivery controls once their agents start producing real artifacts ^[src-191].
• Coding agents create new review debt: a pipeline must evaluate not only final code, but also the agent's assumptions, tool use, tests, diffs, and evidence of completion ^[src-191].
• Agentic CI/CD includes save/resume state, liveness models, traceability, eval suites, rollback/recovery paths, human review gates, and clear "done" semantics ^[src-191].
• It is adjacent to MLOps and conventional software CI/CD but not identical. Agent trajectories, prompts, tool schemas, memory, and handoffs become first-class pipeline inputs ^[src-191].
• The broader playlist ties agentic CI/CD to software factories, multi-agent fleets, process discipline, hallucination controls, and harness engineering ^[src-191].

Related concepts

• Agentic Engineering
• Harness Engineering
• Software Factory Model
• Continuous Agent Evaluation
• Mlops Coding Discipline
• Git Worktrees For Parallel Agents
• Long Running Agents
• Agent Evidence Receipts

Source references

• ^[src-191] AI Engineer World's Fair Online Track 2026 playlist update (47 new transcript captures, 2026-06-22 to 2026-07-02)