Better AI won’t fix bad company data: Fuel Cycle’s Daryush Laqab

Companies racing to deploy AI agents may find their biggest hurdle in the information those systems use, according to Daryush Laqab, chief product and AI officer at market research firm Fuel Cycle.

“A great AI will not fix bad data quality,” Laqab told ET AI.

McKinsey’s 2026 State of AI survey shows how quickly the stakes are rising. Around 40% of respondents at companies with more than $1 billion in annual revenue said their organisations were scaling AI agents in at least one business function, up from 27% a year earlier. Across companies of all sizes, 44% said they were scaling AI across the enterprise, up from 38%.


But more agents need more reliable data. In an April 2026 analysis, McKinsey said eight in 10 companies cited data limitations as a barrier to scaling agentic AI. The consultancy urged companies to improve the quality of their data and modernise the systems that give agents access to it.

Also Read: AI agents could rewrite the rules of office work: Dell’s Rob Bruckner

Laqab argues that companies need to pay more attention to those foundations as they increase their AI spending. He wants businesses to invest more in AI, but says much of the money now flows to chips, hardware and power. He expects investment in data, testing and validation to grow as companies put agents to work across their operations.

“I think everybody should invest more in AI,” he said. “There is a lot of investment actually going to the upstream part of the AI ecosystem.”

The data behind the agent

Laqab says even increasingly capable models from companies such as OpenAI, Anthropic and Meta lack the knowledge that makes an AI system useful inside a particular business. Companies hold that context in their own data.

“Your enterprise context is your proprietary data. That is the part that will make AI useful, that may even make AI real,” he said.

That creates a problem for companies that connect agents to fragmented or inaccurate records. McKinsey warned in its April analysis that agents can make inconsistent decisions when they draw on data from disconnected systems. Laqab says a stronger model cannot solve that underlying problem.

“Although AI is the cool thing right now and everybody is rushing towards AI, in order to make it useful, you still need the data foundations,” he said.

Also Read: AI’s trillion-dollar question: What happens if the boom slows?

Companies risk focusing on the technology while overlooking the work that supports it, Laqab argued. “Sometimes we just obsess over the shiny object and forget that foundations matter,” he said.

Testing what agents do

Better data alone will not settle every question about whether an agent works reliably, according to Laqab. He expects businesses to add checks that test agents’ actions and catch mistakes as they move beyond trials into regular operations.

“I think we’ll build a lot more quality assurance and quality gates for our overall agentic flows,” he said. He expects those checks to help companies trust AI systems and use them more widely.

Laqab compared this stage of AI adoption with the early years of cloud computing. Companies first moved existing applications to the cloud, then redesigned those applications to take advantage of it, he said. He expects businesses to follow a similar path with AI where they will add agents to existing systems before they rethink the data and workflows those agents rely on.