Enterprise AI agents get access to just 45% of company data on average, according to a survey of 300 data and technology executives released August 12 by MIT Technology Review Insights in partnership with Google Cloud. The gap between the best and worst performers is stark: organizations classified as data laggards feed their agents 30% or less of enterprise data, while data leaders clear 70%. That disparity is now the single biggest predictor of whether agentic AI actually works inside a company.
The report frames the problem as a mismatch between what agents need and what legacy systems can deliver. Agents don't just answer questions — they take actions across supply chain, point-of-sale, and HR systems, which requires structured and unstructured data pulled in real time with the right business context. Data warehouses updated as recently as a few years ago were not built for that workload.
“The shift from answering questions to taking actions means AI agents need data from across the enterprise, in all its structured and unstructured forms, and with the right business context.”— MIT Technology Review Insights, Report authors, in partnership with Google Cloud
Trust tracks directly with data access. Only around 50% of surveyed organizations say they trust the decisions their AI agents make. Among data leaders, that figure hits 100%. The report treats trust not as a cultural variable but as a downstream effect of data readiness — agents that see more of the business make better calls, and executives notice.
Key facts
- 01A Google Cloud-sponsored survey of 300 data and technology executives found AI agents access just 45% of enterprise data on average.
- 02Data laggards give agents access to 30% or less of company data; data leaders exceed 70%.
- 03100% of data leaders trust their agents' decisions, versus roughly 50% across the full sample.
- 0466% of laggards say legacy data systems limit AI agent scaling; only 8% of leaders report the same constraint.
- 05100% of respondents plan to use agentic AI within two years, with 69% expecting wide deployment.
The scaling numbers are equally lopsided. Two-thirds of data laggards — 66% — say legacy data systems limit their ability to scale AI agents, and 68% say those same systems prevent agents from making decisions at speed. Among data leaders, just 8% report either constraint. The leaders have not solved agentic AI; they have solved the plumbing underneath it.
The timing pressure is real. Gartner projects that AI agents will augment or automate 50% of business decisions by 2027. All 300 respondents plan to be using agentic AI within two years, and 69% expect wide deployment. Companies that cannot expand data access in that window will be running agents on a starvation diet while competitors run them on a full plate.
“If Gartner's prediction that AI agents will augment or automate 50% of business decisions by 2027 proves correct, organizations must eliminate bottlenecks or risk depriving agents of the data they need to make the right decisions at speed.”— MIT Technology Review Insights, Report authors, in partnership with Google Cloud
The playbook the leaders are following is not exotic. The top-cited initiative across all respondents is improving agent access to structured and unstructured data. Enhancing data and AI governance with business context ranks close behind. Data leaders add a third priority: automating data management itself, so the pipeline that feeds agents does not require an army of humans to maintain.
The Google Cloud framing is deliberate. Google has spent the past two years positioning Gemini and its cloud data stack — BigQuery, Vertex AI, and the agent tooling layered on top — as an integrated answer to exactly this problem. A report showing that fragmented legacy data is the blocker is, functionally, a case for buying more of a single vendor's stack. That does not make the underlying numbers wrong, but readers should note the sponsor.
The counterweight worth flagging: an average of 45% data access is a snapshot, not a destiny. Retrieval-augmented generation, MCP-style connectors, and lightweight agent frameworks are all lowering the cost of exposing more data to agents without full platform migrations. Some laggards will close the gap through tooling rather than replatforming. The report also does not disclose industry mix or company-size distribution among the 300 respondents, which matters — a regional bank and a hyperscaler face very different data-modernization bills.
The market signal here is that enterprise AI spending is moving one layer down the stack. Model choice — Gemini versus Claude versus GPT — matters less than most vendors want to admit if the agent cannot see the data it needs to act. The next phase of AI budgets will flow to data plumbing, governance, and connectors rather than to model licenses, and the cloud providers with the deepest integration between data warehouse and agent runtime are positioned to capture most of it. That is the strategic subtext of this survey, and it is why Google Cloud paid for it.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
