Seventy-two percent of enterprises now run agentic AI in production, a figure that would have sounded premature two years ago, when most organizations were still treating autonomous AI as experimental technology confined to sandboxed trials.
Those agents are now embedded in customer service queues, software development pipelines, financial workflows, and supply chains spanning multiple departments. The problem is that 60% of those organizations have no formal governance structure covering their deployed agents, according to Agentic AI Institute research. Adoption ran well ahead of the infrastructure designed to keep it accountable, and enterprise leaders are only now reckoning with what that gap costs.
What Changes When Pilots Become Production
The difference between a pilot and a production deployment looks technical on paper. The consequences are not. A pilot runs in controlled conditions with human review at each decision point. A production agent operates autonomously, handles real data, takes real actions, and in some cases moves real money. The accountability calculus shifts entirely.
At theCUBE’s AI Agent Conference held at the New York Stock Exchange in May 2026, practitioners from finance, data infrastructure, and enterprise software described what separates agents that succeed from those that quietly break down. The consistent answer was context—specifically, the proprietary organizational knowledge that frontier AI models don’t carry by default and that most companies haven’t systematically captured. Vanessa Liu, chair at Appen Ltd., put the problem plainly: “You need to train an employee when they come into an organization—even if they’re rock stars, you need to make sure you onboard them well. Same thing when it comes to AI agents: You have to give them the business context so that they are going to be able to run well.”
The Perception Gap That’s Undermining Enterprise AI
Tai Carmi, chief information officer at WalkMe, shared a data point from the conference that the industry hasn’t fully absorbed: 80% of executives believe they are providing excellent AI tools to their workforce. Only a fraction of employees agree. The disconnect doesn’t indicate bad intentions. It reflects a fundamental difference in vantage point—executives see investment levels and vendor contracts; employees see whether the tool shows up at the right moment in the right workflow.
Hassan Taher, founder of Taher AI Solutions and a consultant who has worked with organizations on AI adoption across healthcare, finance, and manufacturing, has observed similar friction in practice. Deploying AI and getting people to use it productively are distinct challenges, and organizations that conflate them tend to measure inputs (licenses purchased, tools provisioned) rather than outcomes, while actual usage and measurable impact remain low.
When Agents Touch Money, Identity Becomes a Legal Question
Financial sector deployments are surfacing specific problems that lower-stakes applications don’t encounter. If an AI agent has the authority to move funds, execute transactions, or enter contracts, the question of who bears legal accountability for its decisions is no longer abstract.
Sean Neville, co-founder and CEO of Catena Labs, described at the conference a concept his company is building toward: a “know your agent” model, analogous to KYC requirements for individuals in financial services. The goal is a framework that lets financial institutions verify which person or company an agent represents, what it is authorized to do, and why it took a given action after the fact. Barr Moses, co-founder and CEO of Monte Carlo Data, added a harder edge to the same theme: courts have already ruled that the company behind an agent—not the user who triggered it—bears full legal accountability for what that agent does in the world.
That accountability exposure exists whether organizations have thought about it or not. Most haven’t.
Token Lock: The Cost Structure Problem Building Quietly
A separate risk is accumulating in enterprise AI deployments that receives less attention than governance. Woodson Martin, CEO of OutSystems, called it “token lock” at the conference: enterprises that commit deeply to a single frontier AI model surrender leverage over their own cost structure as inference costs compound. Rebuilding systems to accommodate a different model is expensive. Organizations that didn’t design for flexibility are discovering their AI costs are growing in ways that are difficult to reverse.
The practical remedy, according to practitioners, is a platform layer that lets organizations swap models at runtime without reconstructing underlying systems—treating model selection as an operational decision rather than an architectural commitment locked in at the start.
The Governance Gap Won’t Close on Its Own
Hassan Taher has consistently argued, in both his writing and his consulting work, that AI governance works best as a foundation rather than a retrofit. The data suggests most enterprises have done the opposite: deployed first, considered accountability structures later. Sixty percent of organizations running agentic AI in production have no formal governance framework in place.
That gap shows up in concrete failures: agents drawing on stale data, skipping reasoning steps, blowing past token budgets, or producing outputs that looked reasonable during testing and failed in production. The agents don’t announce when they’ve gone wrong. Catching errors requires observability infrastructure—monitoring, logging, anomaly detection—that most enterprise AI teams haven’t built.
Steve Hasker, president and CEO of Thomson Reuters, framed the competitive reality at the conference in terms that cut through the hype: the companies that will build durable AI advantages are those with a clear customer problem to solve and defensible organizational knowledge that no frontier model provides on its own. That’s not a technology argument. It’s a governance and data argument dressed as a strategy question.
The first wave of agentic AI adoption cleared the proof-of-concept threshold. The infrastructure that makes it sustainable is the part most organizations skipped.
