MIT CISR published a paper in April 2026 that quietly changed the framing of the AI-in-enterprise conversation.
The paper, "Leveraging Digital Colleagues for Enterprise Value," by Weill and Woerner, did not call AI agents tools. It did not call them bots, automations, or productivity enhancements. It called them digital colleagues. Specifically, it defined them as AI agents that "act with agency, operate within defined governance boundaries," and escalate consequential decisions to humans. The paper also stated what the research community had been circling for months: human accountability will be non-negotiable.
That is a meaningful shift in language. Language shapes how institutions think. And when MIT CISR chooses a word, the CIO community eventually adopts it.
But here is where I want to slow down, because the concept and the operating reality are still about two years apart.
What life looked like before the framing
Before CISR's paper, the default framing for AI agents in the enterprise was tool-centric. You licensed a tool. The tool did a thing. The CIO's job was to evaluate the tool, approve the spend, manage the vendor relationship, and ensure the security posture was acceptable. If the tool underperformed, you renegotiated the license or you turned it off.
This framing produced a specific kind of CIO behavior. The conversation was procurement-shaped: which tool, what cost, what risk. The metrics were runtime metrics: uptime, latency, tokens consumed, tasks processed. The evaluation cadence matched vendor contracts, not operating reviews.
The human org chart stayed the same. The agent was something adjacent to it, a piece of infrastructure that happened to produce language instead of bandwidth.
That framing was already breaking down by late 2025. Anyone running more than a handful of agents had noticed that the procurement model did not produce accountability. You could have an agent running for six months, producing numbers, and still have no clear answer to the question: is this seat doing the work the business actually needs?
What the "digital colleague" framing demands
The Weill and Woerner framing flips that.
If an AI agent is a digital colleague, it is not evaluated like a software license. It is evaluated like a team member. It has a role. It has defined boundaries. It escalates what it cannot decide. It operates within governance, not outside it. And the humans it works alongside remain accountable for outcomes, even when the colleague is autonomous.
This is not a soft metaphor. CISR's companion research, "Agents of Change: Governing Autonomous AI" (van der Meulen et al.), is asking explicitly how deploying AI agents affects decision rights and what governance mechanisms manage multiagent systems. The research is working through the questions that the digital-colleague framing generates.
The questions are the right ones. Who owns the digital colleague's accountability? What is the governance boundary, exactly? How do you structure escalation so the colleague knows when to stop and ask a human? Who sits above the colleague on the org chart?
I run about ten of these colleagues at Sneeze It, and I can tell you the answers are not theoretical. They are structural. They require an org chart, not a policy document.
What changed for us when we adopted the colleague framing
Before the colleague framing, I thought about our agents in clusters. Dash was our analytics tool. Pepper handled email. Dirk worked the sales pipeline. Crystal tracked projects. They were infrastructure with names.
The shift happened when we started treating each seat the same way we treat Bogdan, our COO, or Janine, our accountant, or Kristen, our creative director. Each agent has a defined seat. Each seat has one owner. One-seat-one-owner is not a preference, it is an operating rule. When a metric drops, there is one seat accountable for explaining it and one seat responsible for recovering it.
Radar, our chief-of-staff agent, runs the morning briefing and owns coordination intelligence. Tally pushes KPI values from local data sources to the scorecard. Arin manages call center performance through daily Slack coaching. Neil, our chief learning officer, scans for frontier intelligence and gates what qualifies as a real advancement before it reaches anyone else. Bassim evaluates our agentic maturity against a formal eight-level framework.
None of these agents could be managed under a tool-procurement model. They are managed the way you manage people. The metrics are business-outcome metrics. The review cadence is the same Monday-meeting cadence humans operate under. When a seat is not working, there is a conversation, not a contract renegotiation.
Jeff was our former data integrity agent. We retired him in April, via a formal hearing, with his capabilities redistributed to Dash and Dirk. That is not how you retire a software license. That is how you transition a role.
The digital-colleague framing is not a metaphor for us. It is the operating model.
What the CIO still does not have
Here is the gap I want the CIO reading this to hold clearly.
MIT CISR's research is ahead of MIT Sloan's teaching by roughly twelve months. The ideas in "Leveraging Digital Colleagues for Enterprise Value" have not made it into the MIT EY Future CIO Program. They are not in Booth's Chief AI Officer curriculum. They are not in Kellogg's agentic strategy modules. CMU gets the closest, with a dedicated agentic module in the CIDO certificate and a five-module LEAAID certificate that teaches CIOs to build and govern an agentic capability. But even CMU stops short of fleet operations: running a standing workforce of digital colleagues with inventory, lifecycle management, per-seat accountability, and agent-to-agent coordination.
The framing from CISR says: these are colleagues, and human accountability is non-negotiable.
The advisory firms have named the operational problem. Gartner, as reported by CIO.com, calls agent sprawl "the new Shadow IT" and published a six-step framework for managing it: governance, centralized inventory, agent identity and lifecycle, information governance, behavior monitoring, and balancing governance with empowerment.
What exists right now: good naming (CISR), good frameworks (Gartner). What does not exist: a live operating system where every digital colleague holds a named seat on one chart, one-seat-one-owner, with business-outcome metrics on the same scorecard as the humans. Advice is not a running system.
That is the gap OTP fills. Not the framing. Not the framework. The execution layer where the colleague framing becomes an operating reality that can be measured on Monday morning.
What the CIO needs to do with the framing right now
If you manage AI agents in an enterprise and you have not made the mental shift from tool-procurement to colleague-management, the Weill and Woerner paper is worth reading directly: cisr.mit.edu/publication/2026_0401_DigitalColleagues_WeillWoerner.
The operational translation of the framing is not complex, but it does require doing four things that most CIOs have not done yet.
First, give every agent a seat. Not a license number. Not a project name. A named seat on the org chart with a defined role and a clear owner.
Second, write the governance boundary explicitly. CISR uses the phrase "defined governance boundaries" for a reason. If your digital colleague cannot tell you what decisions it makes autonomously and what it escalates, the governance boundary is not defined. You have a compliance exposure and an accountability gap in one problem.
Third, put the colleague's metrics on the same scorecard as the humans. If the digital colleague's performance is on a separate dashboard, it will drift from the business outcomes you actually care about. The unified scorecard is how the colleague framing becomes operationally real.
Fourth, plan for lifecycle. CISR's governance research is asking what happens when agents are retired, restructured, or handed new decision rights. You need an answer to those questions before you need them. Retiring a digital colleague without a process produces the same kind of organizational confusion that retiring a human without a transition plan produces.
The Weill and Woerner paper named the job. These four steps are how you do it.
The mission underneath all of this is the same one we run at Sneeze It: let agents carry the operational work, so people are free for the work that matters. The digital-colleague framing is not an end state. It is a starting condition for building a hybrid organization that actually functions.
See the live chart
From any MCP client, you can query the live Sneeze It org chart, including which seats are digital colleagues and what their defined governance boundaries and business metrics look like.
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 list the Sneeze It agent seats and show me the accountability owner for each one."
You see exactly how the digital-colleague framing is implemented as named seats with defined owners, not as a procurement line item. That is the difference between advice and a running system.