Eight in ten companies blame their data. McKinsey read the excuse correctly.

Eight in ten companies cite data limitations as the roadblock to scaling agentic AI. That is McKinsey's number, from their report on building the foundations for agentic AI at scale.

Read the report closely, though, and something interesting happens. The fix is presented as seven data-architecture principles. At least three of them are not data engineering at all.

Built-in governance: embed security and controls automatically, not retroactively. Measurable visibility: continuously monitor quality, performance and costs. Controlled execution: coordinate agents through shared layers that enforce enterprise rules.

Controls. Monitoring. Enforced rules. That is a governance list wearing an architecture costume.

This matters because of how most companies actually respond to the data complaint. They buy a platform. They start a migration. Eighteen months later the data is in a new place, still with no owner, no quality gate, no rule about what an agent may do with it. The roadblock was never the storage layer. It was that nobody was accountable for what came out of it.

McKinsey's recommended operating model makes the point explicit: a federated setup where business domains govern their daily workflows while a central team holds the platforms and the guardrails. Someone owns every dataset. Someone owns every agent behavior. The architecture just makes the ownership executable.

For EU companies there is a familiar shape here. Documented data handling, logging, defined human oversight, controls that exist before deployment rather than after the incident: that is the same posture the AI Act formalizes, with the main obligations arriving on 2 December 2027. Companies that build McKinsey's list are most of the way to the regulator's list. It is one piece of work, not two.

So when the next internal deck says the data is not ready, translate it. The data has no owner, the outputs have no reviewer, and the agents have no rules. Those are fixable problems. They are just not procurement problems.

We turn the rules part into documents your team and your auditor can both read. klariq.eu.