The Enterprise Change Function Is About to Split in Two
Been meaning to write and post this one a while but life got busy.....
I’ve been working on agentic AI deployments for the past year or so. And a pattern has become hard to ignore.
In some organisations, the Change function is in the room when the agent gets designed. Contributing the organisational intelligence that shapes how it speaks to employees, what it’s allowed to do, what must be true before it goes live. Genuinely influencing decisions.
In others, Change arrives after the design is locked. With a comms plan. For a system whose human experience was decided without them.
Both functions have good people. Both are doing their jobs. But they are not doing the same job. And as agentic AI scales, they will not be equally positioned to do work that matters.
I think this split is structural. And I think it’s accelerating.
Why Agentic AI Specifically Is the Forcing Function
The traditional Change Management model was built for passive technology.
A system goes live. Employees need to understand it, use it, eventually embed it. OCM bridges the gap — communication, training, readiness, adoption. It works because the technology waits. People choose when to engage.
An agent doesn’t wait.
It reaches out. It makes judgements. It initiates conversations on behalf of the organisation. It can contact an employee on a Tuesday morning when they’re heading into a difficult performance review, ask them about their workflow, and have no idea that’s what’s happening.
The human experience of an agent isn’t shaped by how well employees were trained to use it. It’s shaped by design decisions made months earlier — how the agent introduces itself, how it handles being told ‘not now,’ how it responds when someone is clearly distressed, who it escalates to and when.
By the time the traditional OCM function enters with a readiness plan, those decisions are locked.
You can’t communicate and train your way out of an agent that was designed without the human lens. The trust damage is already engineered in.
The organisations that understand this are building a different kind of Change function. Not one that manages deployments. One that shapes them.
Two Futures, Described Plainly
Here’s what each looks like in practice.
The consultancy model
This Change function arrives at the design gate, not the go-live gate. Its primary contribution isn’t a communications plan — it’s organisational intelligence. The kind that takes years to develop and can’t be generated by an engineering team on a deadline.
It knows where the trust deficits are in the employee population the agent will contact. What prior programmes left behind. How decisions are actually made — not how the org chart says they’re made. What the communication culture means for how the agent should speak. Which manager populations are most likely to amplify resistance and which are likely to carry the adoption.
That intelligence gets brought to design sessions. It shapes decisions. It changes what the agent does and how it does it before any code is shipped.
This function also facilitates the governance conversations. Who owns the agent’s behaviour post go-live? What must be true before it contacts employees? What does the oversight model actually look like in practice — not on paper? These conversations are hard and politically charged. They need a facilitator with the skills to run them and the credibility to hold them to conclusions.
That’s what this version of the Change function does.
The delivery model
This function is professionally mature, well-run, and increasingly peripheral to the decisions that determine outcomes.
It delivers well. Communications are clear. Training is thoughtful. Readiness assessments are thorough. The practitioners know their work.
But the function arrives after the agent’s interaction design is locked. After the first contact message has been written. After the escalation paths have been configured. After the governance structure has been agreed — or not agreed, as is more often the case.
The most it can do is manage what results from those upstream decisions.
Prosci’s research across more than 1,100 organisations found that 63% of what’s blocking AI adoption are people issues, not technology issues. Most of those people issues weren’t created at adoption. They were created at design. An agent that triggers job security fears in a population that was never consulted. An agent that contacts employees using a tone that feels wrong for the culture. An agent whose escalation path reflects an org chart nobody actually follows.
The delivery model Change function arrives in time to manage these problems. Not in time to prevent them.
What the Research Is Saying
The case for the consultancy model isn’t just a change management argument. It’s a risk argument backed by consistent evidence.
RAND Corporation’s 2024 study found that 84% of AI implementation failures are leadership and governance failures — not technical ones. Gartner is projecting that more than 40% of agentic AI projects will be discontinued by 2027, driven by weak governance, poor risk controls, and limited organisational readiness.
Capgemini’s 2025 research — surveying 1,500 senior leaders across 14 countries — found that only 27% of organisations trust fully autonomous AI agents. Twelve months earlier that number was 43%. In one year, trust dropped by 16 percentage points. The reasons cited: data privacy concerns, algorithmic opacity, ethical uncertainty. Not technical failure.
That’s a human and organisational design problem. And it is getting worse, not better, as deployments scale.
The Microsoft Work Trend Index adds the employee dimension: more than half of employees worry AI will replace their job, and nearly half are concerned about how their data is being used. These are not concerns that emerge after go-live. They’re present from the moment an employee hears an agent is coming.
If the Change function isn’t shaping how those concerns are addressed in the design — what the agent says in its first message, what data it uses and how that’s explained, what employees can do if they don’t want to engage — nobody is. The engineers are designing a technically correct system. The trust architecture is being left to chance.
What Determines Which Future an Organisation Gets
Three things, in my observation.
The first is how leadership frames the problem. In organisations where agentic AI failure is understood as a human and governance risk — not just a technical one — Change gets resourced and positioned differently. In organisations where it’s still primarily seen as a technology deployment with a people side, Change arrives too late and too narrowly scoped.
The second is whether the Change function has claimed the design space. This isn’t just about getting an earlier invite to meetings. It requires showing up with something the design team needs that they can’t generate themselves — the organisational intelligence, the trust mapping, the real decision anatomy. When a Change practitioner walks into a design session with a completed trust deficit map and specific implications for the agent’s first contact message, the conversation changes immediately. It doesn’t require a political argument about OCM’s relevance.
The third is timing. Organisations that get this right now will build a methodology, a track record, and internal credibility that compounds. The ones that figure it out after the first major agentic incident — a trust breakdown, a governance failure, an escalation that lands badly — will spend months retrofitting what should have been designed in from the start.
Something Most Senior Change Leaders Won’t Say
The delivery model is comfortable. It has predictable scope, clear deliverables, and established expectations. Leadership understands it. Budgets get allocated to it. There’s no ambiguity about what success looks like.
Claiming the design space is messier. The contribution is harder to measure. The skills required extend beyond traditional OCM into territory — AI system literacy, trust architecture, interaction design — that most practitioners haven’t developed yet. And it requires initiating, rather than responding, which is a different professional posture.
But comfortable is a short-term position.
93% of senior leaders in Capgemini’s research believe organisations that successfully scale AI agents in the next twelve months will gain a competitive advantage over peers. That scaling is happening now. The Change functions that are already in the design room are building the credibility and the methodology that will define the profession’s role in AI for the next decade.
The ones that stay in the delivery lane will be doing increasingly efficient work that’s increasingly peripheral to the outcomes that matter.
This isn’t a comfortable observation. It’s an accurate one.
Which side of this divide is your function on? And is it there by design or by default?
I’d genuinely like to know what you’re seeing. Drop it in the comments — I read them all.
Cheers, Andy