Descriptions:
Patrick Debois — widely credited as a co-founder of the DevOps movement — delivers a conference talk at AI Engineer proposing a “Context Development Life Cycle” (CDLC) as the natural successor to the software development life cycle. The central thesis: as AI coding agents take over code generation, the highest-leverage engineering work shifts from writing code to designing, testing, distributing, and iterating on the *context* that drives those agents. In his framing, skills, workflows, and reusable prompt packages are the new code artifacts.
Debois walks through each phase of the proposed loop. Generation covers prompts, CLAUDE.md / agent.md instruction files, pulled library documentation via MCP, spec-driven planning that agents break into sub-tasks, and tickets as context inputs. Testing introduces LLM-as-judge with sandboxed tool execution — running a curl against a live endpoint rather than just inspecting code — and addresses the non-determinism problem by recommending multi-run evals with error budgets (e.g., a critical test suite allowed to fail at most once in five runs) rather than binary pass/fail gates. Distribution covers Git-based context versioning and shared skill packages reusable across multiple projects. The observe-and-adapt phase closes the loop with CI/CD integration and optimization prompts that feed eval feedback back into context improvement.
The talk draws an explicit parallel to 2009-era DevOps — “what if ops looked more like dev?” reframed as “what if context is the code?” — and is delivered with candid acknowledgment that the framework is still unpolished. It is most valuable for platform engineers, AI tooling builders, and team leads thinking about how to operationalize prompt and context management at an organizational scale.
📺 Source: AI Engineer · Published May 03, 2026
🏷️ Format: Deep Dive
