The Organizations Winning With AI Are Not the Careful Ones. They're the Governed Ones.

Last week I wrote about Agent 7. The AI agent that followed policy perfectly while exfiltrating thousands of records before breakfast. The response surprised me. Not the agreement on the risk side. I expected that. What I did not expect was how many CIOs followed up with a version of the same question.

"I get the risk. But if I slow down to build governance, am I just handing the advantage to someone else?"

It is a fair question. And the answer is more interesting than most people expect.

IBM published a study this week that should settle this.

The IBM Institute for Business Value surveyed 2,000 C-level technology executives. The results came out four days ago.

The headline finding is this.

Organizations that engineer control into their AI systems deploy 16 times more agents than those relying on manual governance. While delivering 18 percent higher operating margins. And spending 4 times less of their AI budget in the process. Read that again. The organizations with the strongest governance are deploying the most agents. Not despite the governance. Because of it.

This is the thing I keep trying to explain to boards and leadership teams. The control plane is not a brake. It is the accelerator. You cannot move fast without it. You can only move recklessly without it, and reckless is not the same as fast.

The same study found that only 11 percent of CIOs and CTOs say they are fully prepared for the scale of AI agent deployment expected in the next twelve months. And 77 percent say AI adoption is already outpacing their current governance capabilities.

So the majority of technology leaders know they are behind on governance. They know agents are proliferating anyway. And they are watching a minority of their peers pull away because that minority built the infrastructure to say yes faster. That is the actual competitive dynamic right now.

The velocity gap is already measurable.

BCG published research last fall that quantifies what this looks like financially. Companies they classify as future-built — the 5 percent that have fully committed to AI as an operating model — achieve 1.7 times revenue growth, 3.6 times three-year total shareholder return, and 1.6 times EBIT margin compared to laggards.

They plan to spend more than twice as much on AI as laggards in 2025. The gap is not closing. It is widening.

Agentic AI specifically is driving the separation. Agents already account for 17 percent of total AI value in 2025. That share is expected to reach 29 percent by 2028. A third of future-built companies use agents. Among laggards: almost none.

HFS Research calls this the AI Velocity Gap. The widening divide between how fast individuals are adopting AI to get work done and the speed at which enterprises are building the infrastructure to govern it. Their finding: individuals will adopt agentic AI 10 times faster than enterprises can build the governance layer underneath it. That is not a technology problem. That is a leadership problem.

The 65 percent number you probably haven't heard.

Cloud Security Alliance and Token Security published research in April. The headline: 65 percent of organizations experienced at least one cybersecurity incident caused by an AI agent in the past year.

Among those incidents, 61 percent involved sensitive data exposure. 43 percent caused operational disruption. 41 percent resulted in unintended actions across business processes. The agents were not malfunctioning. They were doing exactly what their permissions allowed.

And here is the part that matters for the velocity conversation. 60 percent of those organizations cannot terminate a misbehaving agent. They can watch it. They cannot stop it. If you cannot stop an agent, you cannot confidently deploy one. And if you cannot confidently deploy one, you are not in the 16 times camp. You are in the we are being careful camp, which is actually just the we are losing slowly camp.

The organizations that are winning are not being careful. They are being precise. There is a meaningful difference.

What precision looks like in practice.

I gave a keynote on this last week in San Francisco. The talk was built around a five-layer control plane. Identity. Scope. Approval gates. Behavioral baselines. Kill switch.

Not because five is a magic number. Because those five things are the minimum infrastructure for confident agent deployment. You cannot skip one and call the others governance. They work as a system or they do not work.

Every digital worker needs a passport before it enters production. Not a spreadsheet that gets stale in three weeks. A real operating record. Who owns it. What it does. What systems it can touch. What it is forbidden to do. What approval gates apply? How it get shut down. If that sounds basic, good. Basic controls are exactly what most organizations are missing.

What we are building at Vasion.

I want to be transparent about what this looks like from the inside, because I think practitioners learn more from each other than from frameworks.

At Vasion, we are deploying agents internally across GTM, operations, and internal workflows as part of our 2026 AI-first strategy. We are not watching from the sidelines. We are in it. Which means we are also living the governance problem in real time, not theorizing about it.

What that has taught us is that the endpoint question is more complicated than most people realize.

When most technology leaders think about AI agents and endpoints, they think about the obvious ones. Email. CRM. Slack. Cloud storage. APIs. Those are real surfaces and they matter. But there is a category of endpoint that almost nobody is governing yet.

The physical document layer.

Intelligent print is an endpoint. A multifunction printer connected to a document workflow is an endpoint. An agent that can route, classify, capture, redact, and release physical documents is an agent with reach into regulated data. Often data that is more sensitive than what lives in the CRM. Signed contracts. Patient records. Financial statements. Compliance documents that were never designed to be touched by an autonomous system.

As Vasion builds toward an agentic platform, we are treating document workflows and print endpoints with the same governance posture we apply to digital agents. Every agent that touches a document workflow gets a passport. Every print endpoint that an agent can reach gets scoped permissions. Every action that moves a document gets logged at the action layer, not the application layer.

Most organizations have a blind spot here. They are governing the digital agents they can see and ignoring the physical workflows those agents can reach. An agent that can trigger a print job, release a document to a physical output tray, or route a scanned document to an external destination has the same blast radius as an agent with email send permissions.

The agentic journey and the document platform are not separate conversations. They are the same conversation.

The endpoint is wherever the agent can act. That includes the printer down the hall.

The question I am sitting with.

The IBM study found that 92 percent of enterprise security leaders lack full visibility into their AI agent identities.

That is not a minority problem. That is the default state of enterprise AI in 2026.

The organizations in the other 8 percent are not smarter. They are not better resourced. They made a decision earlier. That the control plane was not optional infrastructure to build someday. It was the prerequisite for everything they wanted to do with agents.

That decision is still available. It is just getting more expensive to make the longer you wait.

The question I left the room with in San Francisco is the same one I asked in the last piece.

Can you name every digital worker running in your environment right now?

If the answer is no, you already know what the first thirty days look like.