Author argues companies must store AI agent memory in owned Git repositories

The author argues that as AI agents increasingly perform core company work, a business's true moat and intellectual property is its memory, which must be stored in a version-controlled repository owned by the company rather than in proprietary vendor features. Storing operational knowledge, documentation, rules, and history in plain files within a git repository ensures organizational sovereignty and prevents catastrophic disruption if an AI provider changes its terms, prices, or policies. The author outlines a practical framework using an AGENTS.md entry point, role-based skills, and git-tracked documentation to maintain multi-harness compatibility across different AI models. Open article →