Chapter 5
Purpose
The Utopia of the Agentic Enterprise, Part I. The Jobs the Agents Are Here to Take
Even when the economics are well understood, the transition can be harder than the economics alone would predict. The business case has a line for the licences and one for the savings, and none for identity.
At one point I was hiring. As a solution architect at one of the largest manufacturers in the world, I had to staff a team of about 20 engineers to take over a business application from an offshore team that had been struggling to keep the lights on. On paper the job was matching people to the right list of languages and frameworks. My usual question was: tell me something cool you did that wouldn’t be worth a line on your CV. For the right people that conversation took most of the interview, because what we needed was aptitude and attitude more than the name of a particular function.
Most of the developers I met had spent years in large teams with more constraints than opportunities, and had got used to a narrowly defined box on the org chart. You write the code. You don’t ask why, and you don’t need to know who needs it. Someone tells you in templated language what to build, and you build it. Even in a 45-minute interview you could tell that most of them weren’t ready for that arrangement to change. For many years, companies treated software engineers as somewhat mythical creatures, and let them get away with much more than they should have. They were expensive, hard to find and complicated to work with. But out of nowhere, the scarcity that was assumed would protect the careers began to evaporate.
Identity here means why the work matters to the person doing it. It also means whether the organisation still trusts their judgment on the ambiguous case, and turns to them when a decision has to be owned. AI moves all of it. So what people feel as loss may be the trust and the ownership quietly moving elsewhere, and the announcement email doesn’t mention that part.
Authority Before Identity
In 1948 Lester Coch and John French published “Overcoming Resistance to Change”, a study at the Harwood pyjama factory in Virginia.1 Workers who were simply told about a change to their jobs produced less afterwards, and more of them quit. Workers who helped plan the change got back up to speed quickly. What the workers resisted was the way management brought the change in, but the management textbooks kept the title and dropped that part. In 1999 Eric Dent and Susan Goldberg traced the idea back to Kurt Lewin, for whom resistance belonged to the whole system, managers included, and found that nearly every textbook had turned it into something employees do and managers overcome.2 People, they argued, don’t resist change as such. They resist losing pay, status or comfort.
An AI rollout can take the status first. When a model takes over a decision someone used to make, the standing that came with it goes too: the people above stop asking for their view, and the people around them stop bringing them problems. The accountability for how the decision turns out stays where it was. Change-management training still teaches the textbook version, so whoever runs the rollout is likely to treat the objection as resistance and book a workshop for it. The workshop won’t help, because the person has read the situation correctly, and no amount of reassurance changes who makes the decision.
Meaning in the Method
Knowledge workers derive meaning from the way they work, beyond the tasks themselves. In most knowledge professions, the method is part of who people are.
This is separate from the economic argument. If AI takes over most of an analyst’s hours and the pay stays the same, or even goes up, the economics say they should welcome it. But they may be asking something else: whether the work will still be work they recognise as theirs, which is not something the salary review covers.
Professional identity grows through years of practice and the satisfaction of doing difficult work well. When AI produces comparable output in seconds, the economic value may survive, and the value of having built it yourself mostly doesn’t.
That explains why the strongest resistance can come from the people with the deepest investment in the craft, whatever their economic position. A junior two years in may be glad of the chance to skip the drafting. A senior built a reputation over decades on the quality of that drafting, and for them it is closer to loss.
Plenty of people feel nothing of the sort. Nobody ever found meaning in retyping the same method statement into every tender portal, and for the people who had to, AI takes away work they only ever put up with. For them the constraint was always time. After the drafting and the reporting, there wasn’t enough left to do the work that drew them to the profession. Relief is easy to hear. When someone describes what AI changes about their work and sounds lighter doing it, it’s worth following that thread and asking what the work has been keeping them from. Of course, the same person can feel both, so the conversation has to start open enough for either.
A single change-management approach across the workforce misses this. Someone relieved of the paperwork needs the workload redesigned before the freed hours turn into throughput, because hours nobody has a plan for get filled with more of the same, by whoever controls the calendar.
Software engineers may meet the identity problem first. Programming is both the profession AI has changed most and one with a deep craft tradition.
Engineers spend years mastering clean architecture and idiomatic code, until they think in it. Now they work beside a system that turns that craft into working software without any of the intuition behind it. And the software works anyway. The systems were trained on code written by people who cared about readability and structure, so the craft tradition is, quite literally, the training data. Which is a compliment to the people who wrote it, if not a paid one.
In software, where the tools arrived first, some engineers use them, acknowledge the gains, and feel a loss they are not sure how to name.
Craft identity erodes even when the economic outcome is neutral or positive. Moving from “I produce this” to “I check that this was produced correctly” changes how a professional sees their own work.
Manufacturing went through this with CNC. Machinists who adjusted feeds by ear and diagnosed tool wear through vibration became machine operators once the machine took over the cutting. Harry Braverman, writing in 1974 while it was happening, called the mechanism the separation of conception from execution: the knowledge of how the work should be done moves out of the worker and into the office that programs the machine, and the worker is left with the loading.3 And his point was that this wasn’t a side effect of the technology. It was what management bought the technology for, because once the method belongs to the office, the worker is cheaper to replace and easier to direct. Which makes it a rare technology purchase that delivered what the business case promised. The larger loss, arguably, was authority: the machinist used to make the call, and now took it from the machine.
Skipping the Conversation
When a rollout goes straight from “here’s your new AI tool” to “why aren’t adoption numbers higher?”, people use the tool where it costs them nothing: the reports nobody reads. Those uses pad the adoption metrics nicely, and the metrics go into a report of their own, quite possibly drafted with the same tool.
Meanwhile, the hard cases, where AI could create the most value, get routed around the tool, because the change of identity required to trust it with work that matters hasn’t been made.
The behaviour is self-protective, and arguably sensible. Nobody has given the person a reason to believe the organisation values what they are becoming.
Knowledge work is also team work. “We are the copywriting team” or “we are the analytics group” is a shared identity, something larger than a few people happening to sit in the same room. The junior learns by watching the senior handle a difficult brief. The senior stays sharp because the junior asks questions that make them explain what they’d never had to put into words.
When AI replaces most of the team, the organisation records a headcount reduction, which looks good in the next budget round. It has also dismantled the way the team worked together. The people who remain aren’t a smaller version of the same team. The mentoring is gone, and the knowledge that used to pass over lunch and in corridors has fewer places to pass.
Individual concerns respond to coaching and career conversations. A broken team needs a redesign of how the remaining people work with each other. That part gets skipped, because the coaching is booked and the new role descriptions are out, so the identity problem looks dealt with. And it has been, or at least the individual half of it.
Organisations that replace teams with sets of specialised agents meet a version of the same problem. The agents need shared context and handoff protocols, and somebody has to design and maintain those. So the coordination problem moves somewhere else and doesn’t go away. An organisation that struggled to coordinate copywriters will struggle just as much with specialised models, only now with usage fees.
A redesign isn’t finished when the tool goes live, whatever the project plan says. Whether it is finished for the people inside it comes down to a few questions.
Has pay followed responsibility, and authority followed accountability? Has HR updated the role architecture, or only the training catalogue?
Where the answers are no, the organisation has changed the work and left the job around it to take the consequences. The person still has a job, and it is less coherent and less trusted than the one it replaced.
Checklist
- Is what you’re calling identity resistance a correct reading of lost authority?
- Are you listening for relief? Relief needs the workload redesigned before the freed hours turn into throughput.
- If a whole team has lost what held it together, are you redesigning how the remaining people relate, or booking coaching?