--- title: "Your AI Context Layer Is Being Built on Stale Metadata" url: "https://atlan.com/context-and-chaos/issue/your-ai-context-layer-is-being-built-on-stale-metadata/" description: "Why governance programs decay after go-live, and why AI agents make that decay existential" keywords: "Data Governance, AI Agents, Context Engineering, Stale Metadata" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ A Context & Chaos issue (Atlan's practitioner newsletter) by **Amanda Darcangelo, Sr. Lead Data Consultant at OneSix**, who helps organizations build data governance programs and has audited dozens of post-go-live catalogs. Published May 21, 2026, 8 min read. Why governance programs decay after go-live, and why AI agents make that decay existential. Key points: - Governance programs built as one-time projects decay right after go-live: definitions go stale, data owners leave, and analysts route around the catalog with Slack channels and tribal knowledge. - AI agents inherit the stale context humans were silently compensating for. Analysts feel the wrongness and ask someone; agents read the catalog and act, treating outdated definitions as ground truth. - The fix is a shift from governance-as-project to governance-as-operating-discipline: event-driven discovery, dedicated stewardship roles with freshness in their performance reviews, and operational funding instead of one-time capital budgets. ## The call An organization celebrated a governance go-live about 18 months earlier: swag, a ribbon-cutting for the new catalog, a steering committee every Tuesday. Now adoption has flatlined, metadata is stale, the committee has disbanded, and the catalog is dead links and VARCHAR(255) definitions. The same organization now wants to build an "AI Context Layer", using the same project mindset that failed the first time. ## What the autopsy looks like - Thousands of definitions, most untouched for over a year; the glossary was populated by a team that has rolled off; half the listed owners have changed roles; lineage reflects an architecture refactored six months ago. - Practitioners route around the catalog with their own Slack channels, tribal docs, and "ask Sarah, she knows how that table works." - Governance failed because it was treated as a deliverable, not a discipline. - **Financial services client audit:** 11,400 business term definitions; 64% not updated since the implementation sprint 14 months earlier; 38% of listed data owners had changed roles or left; lineage accurate only for the go-live architecture, which had since been refactored twice. The catalog was wrong because nobody's job depended on keeping it right. ## The project trap Projects have a start, a finish line and a CapEx budget: select a tool, configure it, train, declare victory. Governance is a capability, not a tech implementation. Treating it as a project is building a house and firing the maintenance crew on move-in day; decay starts when the project team disbands. ## Why AI makes this existential - A human analyst can spot a weird number and ask a teammate. AI agents only know what they are told; tribal knowledge does not transfer. If "Active Customer" changed last week but governance ended last quarter, the agent makes autonomous decisions on a lie. - **Regional insurer example:** an AI agent in a customer retention workflow pulled "At-Risk Customer" definitions written during the governance project, before the product team revised activation criteria after a policy change. It ran for six weeks before anyone noticed, flagging customers the product team had decided not to pursue. No catastrophic outcome, but six weeks of an autonomous workflow on an outdated definition. ## From project to operating discipline An AI context layer built right is the continuously maintained fabric of governance, metadata and feedback loops between the data stack and the AI applications consuming it. In practice: - **Event-driven, automated discovery:** a production schema change should reach the context layer immediately, not at the next quarterly review. - **Dedicated stewardship:** a real role (or heavily incentivized function) with a seat at the business table; governance survives where someone's performance review depends on it. - **Feedback loops instead of annual audits:** monitor how AI uses definitions in production and catch hallucinations or PII leaks before users do; when an agent misapplies a policy, route that signal back. - **Operational funding:** governance is a utility, like electricity; you don't finish paying for the lights. **Case:** after its second failed governance implementation, one organization dissolved the project team, converted two data steward positions into permanent operational roles with coverage rate and definition freshness in performance reviews, and moved the governance budget from a capital project line to an operational utility line alongside data infrastructure. Twelve months later catalog freshness rose from **31% to 74%**, and stewards were included in AI scoping conversations from the start. ## The bottom line Ask: when your governance project ends, does the governance end with it? If yes, you aren't ready for AI. Agents scale successes at machine speed and failures faster. The context your AI operates on is only as current as the last time someone maintained it, and if nobody's job depends on that, nobody will. ## About Context & Chaos A community newsletter where practitioners, builders and thinkers share stories and lessons on context engineering, governance, architecture, discovery, and the human side of data and AI work. [Browse all issues](https://atlan.com/context-and-chaos/). Related reads: - [Data Governance vs AI Governance: Why It's the Wrong Battle](https://atlan.com/context-and-chaos/issue/data-governance-vs-ai-governance-why-its-the-wrong-battle/) (March 2026) - [BI-Ready Is Not AI-Ready](https://atlan.com/context-and-chaos/issue/bi-ready-is-not-ai-ready/) (March 2026) - [Context Graphs as AI Evaluation Infrastructure](https://atlan.com/context-and-chaos/issue/context-graphs-as-ai-evaluation-infrastructure/) (April 2026) - [Conceptual Modeling Is the Context Engineering Nobody Is Doing](https://atlan.com/context-and-chaos/issue/conceptual-modeling-is-the-context-engineering-nobody-is-doing/) (April 2026)