Half the companies racing to put autonomous AI agents into production have quietly admitted something awkward: the data they already own is the part they trust least.
That’s the buried headline of Informatica’s CDO Insights 2026 survey of 600 data leaders. Among those already running or planning to use agentic AI, 50% cite data quality and retrieval as their single biggest barrier to production. Another 57% still point to data reliability as the wall between a pilot and a real deployment.
The pilots keep multiplying. What sits underneath them hasn’t moved in a year.
Steven Seah, who runs South Asia for Informatica from Salesforce, has seen this play out often enough to call the ending early. Building the agent, he says, is “relatively straightforward.” The hard part is the data underneath it — keeping it “current, trustworthy, and explainable.” When it isn’t, you get “inconsistent or inexplicable outputs”: the agent doing the wrong thing with total confidence, and no one can say why.
The trap CDOs keep walking into
Most data leaders think they’ve already handled this. A decade of effort went into it — connecting systems, piping everything into a lake, and wiring up the warehouse or data lakehouse.
None of that is the same thing, Seah argues. “Trusted context is not the same as connected data,” he says. “You can have data flowing freely across systems and still have agents that can’t act on it reliably — because there’s no metadata explaining what it means, no lineage showing where it came from, and no quality signal indicating whether it’s fit for the decision at hand.” The plumbing can be immaculate, and the agent is still guessing at meaning.
His fix is to stop treating governance as a box you tick once the data has already moved. Metadata, lineage, and quality should ride inside the data itself — “not sitting in a separate layer someone has to consult after the fact.” Do that, and “agents can act with precision.”
The number that should concern every CDO
The cost of getting it wrong, in Seah’s telling, runs to “a USD3.1 trillion annual cost from fragmented, ungoverned data — not from a lack of data, but a lack of shared understanding.”
He leans on that last distinction, and CDOs should too. Scarcity was never the issue. “Most enterprises have more data than they’ve ever had,” he says — the trouble is that “meaning is scattered across systems, each with its own version of truth.”
Steven Seah @ Informatica: “When the foundation isn’t in place, you’re not scaling intelligence — you’re scaling confusion.”
The symptoms are mundane and costly. In a merger, two teams can’t agree on which org hierarchy survives. In sales, rival reps swear they own the same account. In the supply chain, you can’t trim spend on suppliers you can’t clearly see. “These aren’t edge cases,” Seah says. “They’re daily operating costs that compound into compliance risk and missed opportunity.”
AI doesn’t soften any of that; it speeds it up. “An agent can’t act on context that is stale, inconsistent, or locked in a silo,” Seah says. “When the foundation isn’t in place, you’re not scaling intelligence — you’re scaling confusion.” What gets spent down, he adds, is the inventory no company keeps in reserve: “time, money, and trust.”
Headless data management is here
At Informatica World 2026 recently, the company announced the first enterprise data management platform to deliver fully headless data management — enabling metadata, data quality, and master data management tools available as MCP (Model Context Protocol) servers and CLAIRE Agent skills, allowing agents to invoke these capabilities directly from any LLM or IDE. “Every Informatica capability — metadata, quality, MDM — becomes invokable inside any agent workflow, on any platform,” Seah says. “No rearchitecting. No custom integration.”
“This headless data management innovation shifts complexity into a unified platform, enabling agents to handle data tasks and ultimately build a trusted, always-ready data foundation across all surfaces, agents and workflows,” he adds.
The honest closer
How would a skeptical data leader know any of this is real, and not just a fresh coat of paint on the word “integration”?
His test is a hard one. Seah says the evidence won’t be a new logo or another acronym — it’ll be “clear accountability for decisions made by AI agents,” with governance and quality “traveling with the data itself rather than being managed separately.”
Which sets up the question every CDO will eventually field, probably from legal, probably at the worst possible moment. When the agent makes the call: “What data did it use? Was it accurate, and can we stand behind that outcome?”
If you can’t answer that, you don’t have an AI strategy. You have a very expensive guess.
Image credit: iStockphoto/charles taylor