Patrick Debois, founder of Tessl and a historic figure of the DevOps movement, develops in this article the concept of the "Context Flywheel" — a contextual flywheel that creates a cumulative competitive advantage for development teams using AI agents.

The article builds on the Context Development Lifecycle (CDLC) introduced in a previous post, a four-phase cycle: Generate, Evaluate, Distribute, Observe. The CDLC treats context not as static documentation, but as a versioned and tested engineering artifact.

The central thesis is that the value lies not in a single pass through the cycle, but in its repetition. The first iteration captures conventions and identifies gaps. The second corrects these gaps and reveals more subtle problems. By the tenth iteration, a shift occurs: agents do not simply follow instructions better — the entire team codes differently, faster, more consistently, with fewer corrections.

Debois identifies four simultaneous returns on investment for each cycle. First, agent quality improves across the board. Second, the senior engineer who encodes their expertise clarifies and deepens it. Third, junior developers can read the skills and absorb the patterns, constraints, and reasoning. Fourth, the organization converges toward shared terminology that crosses team boundaries.

The strong strategic argument is that the true competitive moat lies neither in tools nor in models — which are commoditizing and converging, respectively — but in accumulated product knowledge. An organization that runs this flywheel for two years develops an understanding that competitors cannot replicate: catalogued edge cases, mapped user needs, domain reasoning encoded and accessible to agents.

Debois concludes by emphasizing the need for explicit context ownership. Without a designated owner — whether a DevEx team, a platform team, or embedded context engineers — context inevitably rots. Ownership rests on three pillars: maintenance (freshness review, conflict resolution, retiring the obsolete), enablement (easy contribution via CLI and evaluations in CI), and governance (quality threshold via evaluations, conflict detection, deprecation policies). The historical analogy is explicit: the industry has already learned this lesson with documentation, infrastructure, and security.