OpenAI is making internet search available to all ChatGPT users, allowing people to engage conversationally with the chatbot while seeking answers or information from the internet © AFP Kirill KUDRYAVTSEV
There has been plentiful debate lately about whether GenAI is losing steam or failing altogether – especially after an MIT report. The report states: “Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return. The outcomes are so starkly divided across both buyers (enterprises, midmarket, SMBs) and builders (startups, vendors, consultancies) that we call it the GenAI Divide.”
As to what the current situation is, Matt Carroll, CEO at Immuta provides an expert insight from a conversation with .
Carroll makes the case that the potential for GenAI remains massive, and what is really slowing adoption is governance.
: Why do you think GenAI has slipped so far on the hype scale? What’s keeping it from achieving its potential?
Matt Carroll: I wouldn’t really say it’s slipped. “GenAI” is a loaded term, and what Gartner’s pointing to is that we don’t yet see OpenAI or Anthropicstyle models embedded across every enterprise. That’s not because the potential isn’t real — it’s because adoption takes time. With platforms like Databricks or Snowflake, it took years to absorb them into enterprise workflows, and those were “just data.” GenAI is bigger: it touches unstructured and semistructured data, entirely new workflows, and a massive number of potential use cases. Enterprises don’t turn that corner overnight. The ISV ecosystem has to adapt — tools, middleware, access controls, audit, lineage — and then enterprises can adopt at scale.
What’s really slowing things down is governance. GenAI makes everyone a data consumer, which means anyone can potentially access and act on sensitive or poorquality data. But in a typical enterprise, data governance teams are just 10–20 people. These are experts with years of experience making access and usage decisions, but they can’t scale to thousands of employees on their own. For GenAI to realize its promise, it has to penetrate the governance layer — the middleware for access controls, audit, and security — so those 10–20 people feel like 1,000 or 2,000. Once that happens, broad adoption will accelerate very quickly, because GenAI will empower non–power users to operate like experts. That’s when we’ll see true decentralization and scale.
DJ: Are failed pilots or lack of ROI creating the perception that GenAI isn’t delivering?
Carroll: I don’t think it’s about failed projects. What we’re really seeing is that enterprises are still preparing their data for largescale GenAI use. Inside Immuta, for example, we’re already deploying GPTs for revenue, messaging, finance — but not all data is equal. A huge amount of the work is making sure poorquality or sensitive data never goes into the model in the first place. That’s where governance becomes critical.
The challenge isn’t that the models underperform; it’s that AI is outpacing the way enterprises implement data controls. Most organizations have governance teams of fewer than 20 people managing data access for tens of thousands of employees. You can’t solve that by hiring more — it’s simply not scalable. The only way forward is to use GenAI itself to amplify those teams, so they can govern and enable access at the scale of thousands. That’s when ROI becomes obvious.
DJ: Was GenAI overhyped in terms of its potential?
Carroll: No — the potential is very real. What was misunderstood was the timing. Until enterprises align their data governance and infrastructure with GenAI, you won’t see adoption at scale. But once they do, we’ll hit an inflection point where ROI is clear, and adoption will accelerate very quickly.
DJ: What upcoming AI technologies are you most excited about?
Carroll: What excites me most isn’t a single new technology, it’s how conversational AI is being embedded into everyday enterprise tools. Think about the leap from Photoshop to Canva: you used to need an expert; now anyone can type in what they want and get a goodenough result. That same shift is happening in enterprise tech. Instead of tickets, forms, and clicks, we’ll just describe what we want — access to data, infrastructure, or analysis — and the system will execute. That will fundamentally change how we work by removing bottlenecks and giving advanced capabilities to everyone.
