Most companies I talk to believe their AI maturity problem is a technology problem.
They are running pilots. The pilots work. Then nothing scales. So they conclude they need better models, better tools, better integrations. They chase the next capability announcement. They redeploy the same budget into the next prototype.
The MIT Center for Information Systems Research has a different diagnosis. And the data behind it is damning.
What the model actually says
MIT CISR published its Enterprise AI Maturity Model through the work of Stephanie Woerner, Ina Sebastian, Peter Weill, and George Kaganer. The model defines four stages: Experiment and Prepare, Build Pilots, Develop AI Ways of Working, and AI Future Ready.
The stage names are deceptively simple. The findings inside them are not.
Stage 4 firms, those that have reached what CISR calls AI Future Ready, outperform their industry by 13.9 percentage points in revenue growth and 9.9 percentage points in profit. Stage 1 firms, still experimenting, underperform by 26.5 percentage points in growth and 15.1 percentage points in profit. That gap between Stage 1 and Stage 4 is not a maturity gap. That is a survival gap.
The finding that stops me is where the hardest transition sits. It is not the move from Stage 1 to Stage 2. It is the move from Stage 2 to Stage 3.
Stage 2 is Build Pilots. This is where most organizations will confidently tell you they have an AI strategy. They have proof of concepts. They have a handful of successful deployments. The C-suite has been briefed. The press release has been issued.
Stage 3 is Develop AI Ways of Working. This is where those same organizations discover that having successful pilots is not the same as having changed how the organization actually operates.
That gap, between pilots and ways of working, is where most companies are stalled right now. And the cause is not the technology.
Why the stall happens
The CISR research points to something specific about what Stage 4 firms have that Stage 2 firms lack. It describes a "united top leadership team" across CEO, CIO, chief strategy officer, and head of HR as a structural characteristic of firms that make it through.
Think about what that implies for Stage 2 firms.
A successful pilot typically has a sponsor. Maybe the CIO. Maybe a business unit leader who wanted to move fast. The pilot produces results. The results are real. But the pilot was owned by one seat, and the operating change required to scale it involves every other seat. The people who run sales, who run operations, who run finance, who run the call center, none of them participated in the pilot's architecture. They are now being asked to integrate something they did not build into work they have been doing for years.
That is not a technology problem. That is an accountability architecture problem.
At Sneeze It, I have been living this in the opposite direction. We built the accountability architecture first, and then deployed agents into it.
Radar, our chief-of-staff agent, does not have a separate AI dashboard. It has a row on the same scorecard that Bogdan, our COO, has a row on. Tally, our KPI agent, pushes numbers to the same chart that Janine, our accountant, has columns on. Dash, our analytics agent, reports into the same Monday briefing that Kristen, our creative director, reports into. When Dash's numbers are wrong, the conversation about why is the same conversation we would have if Bogdan's numbers were wrong.
The pilots-to-ways-of-working transition fails when agents are treated as infrastructure that sits next to the operating system. It succeeds when agents are structural participants inside the operating system. One seat, one owner, one scorecard.
What the model does not give you
The MIT CISR model is research, not a running system.
It tells you where the stall is. It tells you that the transition from Stage 2 to Stage 3 is the critical one. It tells you that unified top leadership is a characteristic of the firms that make it. The research is rigorous and the findings are real.
What it does not give you is the actual architecture for how to run the hybrid organization on the other side of Stage 3.
That is the white space CISR's own research keeps circling. Their 2026 paper on digital colleagues, "Leveraging Digital Colleagues for Enterprise Value," defines agents that act with agency and operate within governance boundaries, with human accountability non-negotiable. Their ongoing work on governing multiagent systems asks how decision rights change when you deploy autonomous agents. The research frontier is pointing directly at the fleet-as-operating-function problem.
But frontier research and operational infrastructure are different things. The model names the destination. It does not build the road.
The Gartner confirmation
Gartner arrived at the same problem from a different angle. In April 2026, Gartner published six steps to manage AI agent sprawl, calling agent sprawl the new shadow IT. The steps cover agent inventory, identity and permissions, lifecycle management including retirement, behavior monitoring, and governance.
That Gartner named this is market validation, not market coverage. A framework for managing agent sprawl is advice. It is not a system with named seats, active scorecards, and a coordination layer where agents send structured messages to each other through inbox files and surface those messages to the humans accountable for the outcomes.
Dirk, our sales agent, does not coordinate with Pulse, our retention agent, through a governance framework document. They coordinate through a live protocol: if a client is on Pulse's churn watch list, Dirk's expansion play for that client is paused. That rule is structural. It is enforced by architecture, not by an advisory firm's recommendation that an employee reads once and files away.
Nick, our prospecting agent, does not produce cold emails that then have no home. The work Nick produces feeds directly into a pipeline that Dirk manages and that surfaces in the same briefing Radar compiles every morning. The agent outputs are inside the operating system, not adjacent to it.
The CISR maturity model names Stage 3 as "Develop AI Ways of Working." The companies I watch build real agent infrastructure, the ones that make it past Stage 2, are not developing AI ways of working as a project. They are running a hybrid organization where the ways of working are structural, maintained, and enforced by who owns which seat.
What actually moves the maturity
The companies that get stuck between Stage 2 and Stage 3 are waiting for the technology to mature.
The companies that move through Stage 3 have accepted that the technology is not the constraint. The constraint is whether humans and agents share a single accountability surface. The constraint is whether the agent's output is inside the business conversation or adjacent to it. The constraint is whether there is a named human accountable for every agent's performance, the way there is a named human accountable for every human's performance.
CISR's united top leadership finding is the organizational science version of what I have arrived at operationally. When the CEO, CIO, chief strategy officer, and HR lead are all present to the AI operating model, the ways of working change because the people who own those ways of working are in the room. When only the CIO is present, the pilot succeeds and the ways of working do not change.
At Sneeze It, that means Crystal, our project management agent, does not exist in a separate system. Crystal's flagged delivery risks appear in the same briefing where Arin's call center numbers appear, where Dirk's pipeline health appears, where Pepper's inbox urgencies appear. Neil, our learning agent, surfaces capability improvements back into the same governance structure, not into a separate AI-team Slack channel. Bassim, our maturity evaluator, runs against the whole fleet and scores us against the same eight-level framework we track week over week.
The mission I am running is simple to say and hard to build: let agents carry the operational work, so people are free for the work that matters. That is not a pilot. It is a ways-of-working decision. And that is exactly what the CISR model says separates the firms that advance from the firms that stall.
See the live chart
The seats on our chart, human and agent, are queryable through the OTP MCP so you can see what a Stage 3 operating structure looks like in practice, not in a model.
In Claude Desktop or Cursor or any MCP client, add this block:
"otp": {
"command": "npx",
"args": ["-y", "@orgtp/mcp-server"]
}
Restart the client. Then ask: "Use OTP to show me the Sneeze It org chart and tell me which seats have active scorecards and which agent seats report into which human seats."
What comes back is not a maturity stage assessment. It is the operating architecture CISR's research says most companies never build.