Share with your CDO
Only 12% of enterprises have AI agents in production, and a mere 3% are using AI to handle most of the work in key processes, according to research published by MIT Technology Review Insights in partnership with Atlan. The diagnosis from Atlan co-founder Prukalpa Sankar is blunt: the models are capable enough, but they’re operating without the organizational context that would make them genuinely useful. The missing architectural piece is a context layer for enterprise AI, a governed translation layer sitting between business systems and AI platforms.
What this means for your business
The 38% of respondents still piloting AI and the 35% stuck at copilot deployments aren’t behind because they chose bad models. They’re behind because raw model capability and business utility are two different things, and the gap between them is filled, or not, by how well the AI understands the organization it’s working inside. If your data governance infrastructure exists primarily to satisfy compliance requirements rather than to feed structured, machine-readable context to AI systems, you’re in the larger half of that distribution, and the ceiling on your AI ROI is architectural, not budgetary.
Sankar’s “blindfolded intern” framing is memorable but also conveniently points toward Atlan’s product category, and the research is sponsored, so the 12% production figure should be read as directionally right rather than precisely authoritative. The core claim still holds on its own logic. An AI agent (a software system that takes autonomous actions, not just answering questions) tasked with, say, flagging anomalous procurement spend needs to know what your company calls a vendor, how your approval hierarchy works, and which data fields are trusted sources of record. Without that, it hallucinates confident but wrong answers against your own data. The context layer is what encodes that institutional knowledge in a form the model can use without a human prompt for every edge case.
The companies that close this gap fastest won’t be the ones with the biggest model budgets. They’ll be the ones whose data catalogs, business glossaries, and metadata standards were already treated as live infrastructure rather than documentation artifacts. If your data governance program has been underfunded or treated as an audit checkbox, that debt is now sitting directly in the path of every AI agent initiative your business wants to run. The decision this reframes isn’t whether to buy more AI, it’s whether the investment in data context infrastructure comes before or after you’ve already committed to an agentic roadmap that will stall without it.
Based on reporting from Creating AI value with enterprise context, originally published 2026-10-28 16:32:00.
