People Power And Politics

People, power and politics

The technology is ready. The organisation usually is not.

---

The part nobody puts on the roadmap

Agents do not only speed up work. They change what the work is, and that lands on people who were not consulted, through managers whose position it quietly threatens.

Everything else in this repo is an operating-model argument. This page is the political one, and it decides whether any of the rest gets used.

The invisible glue work

The email follow-up. The data copied between systems. The manual reconciliation. The thing somebody checks on a Friday because it was wrong once.

Glue work is undervalued, invisible in performance reviews, and it is exactly what an agent absorbs first, because it is repetitive and well-bounded.

When it goes, it can feel like relief or it can feel like erasure, and which one it feels like has almost nothing to do with the agent.

It depends on whether the person was ever credited for it. Someone whose contribution was visible experiences the removal as being freed. Someone whose only visible contribution was the glue work experiences it as being deleted, because the organisation is about to discover it never knew what they did.

Find out who is holding the glue before you automate it. That person also knows every exception you have not documented, which makes them the most valuable person in your discovery and the easiest one to lose.

Middle managers, the hidden kingmakers

The uncomfortable truth: this threatens the middle management status quo more than it threatens anyone below it.

A middle manager's power often comes from being the person who knows how the system works, who can navigate the complexity, who translates between the executives and the floor. Agents threaten to make exactly that expertise less valuable.

They rarely say no. Direct refusal is visible and costly. Instead:

"We need better data quality first."

"The team isn't ready."

"Let's pilot it next quarter."

They slow-roll. They avoid bringing good candidates to intake. They frame it as extra work. They let one early mistake become proof that the thing cannot be trusted.

Every one of those sentences is individually reasonable, which is what makes the pattern hard to name and impossible to escalate.

Flipping a blocker into a champion

Three things, and they are not communications exercises:

Personal relevance. Show how it helps them hit their goals: fewer escalations, better service levels. Corporate return on investment is not an argument to someone whose bonus depends on something else.

Psychological safety, stated explicitly. "You are measured on how effectively your team uses this, not on how many people you remove." If that is not true, do not say it, and understand that everyone will behave as though the opposite is true.

Real participation. Bring them into pilot design. Let them decide which tasks stay human. That is a genuine transfer of authority and it is the only one of the three that cannot be faked.

The mandate trap

Some executives take the aggressive route: AI use becomes a stated expectation, teams must demonstrate why AI cannot do a task before requesting headcount, and usage is tied to performance review.

It is defensible in principle. Left entirely to the middle, the incentive to change is too weak, and this at least creates one.

And there is a documented dark side. One widely-reported case replaced a large part of a customer service function with agents and had to reverse course when quality dropped and customers complained about generic, unhelpful answers. In automating aggressively, the organisation removed people whose institutional knowledge the system could not replace.

The lesson is not "do not automate." It is that over-pivoting to cost containment, without accounting for long-term customer experience and the knowledge walking out, is a predictable failure with a predictable reversal.

Mandates without support systems produce anxiety, not adoption. What makes the difference:

Clarity before speed. People need to understand not only that they must use it, but why and how it helps them.

Support systems. Training, resources, and a safe place to experiment and fail.

Honest timelines. Adaptation takes time and not everyone moves at the same pace. Pretending otherwise does not accelerate anyone.

From user to supervisor

In most technology projects we say "end users" as though they were passive. In an agentic system that metaphor breaks. The human is the supervisor.

Role	The old job	The new job
Front-line service	Writing the responses	Catching tone problems; overriding policy violations
Domain professional	Producing the record	Verifying accuracy; spotting confident fabrication
Engineer	Writing the routine code	Reviewing logic and edge cases; teaching the system through evaluations

Treat these people as "just users" and you will undertrain them. They need supervisory training: how to read an output critically, and how to use the stop when it drifts.

Reading an output critically is a distinct skill from producing one, and being excellent at the second does not confer the first. This is the single most under-funded line in most programmes.

The surveillance trap

Here is the story that destroys trust fastest: using this to monitor employees.

Most organisations now monitor remote and hybrid work in some form, and a large share use automated analytics to evaluate performance. The tooling can log keystrokes, track time, and score activity.

The research direction is consistent and it is not the direction people expect. When work is monitored this way, complaints rise, performance falls, and intention to leave increases. Organisations deploying it can expect reduced productivity and higher turnover rather than the efficiency they were sold.

The mechanism is not mysterious. Without psychological safety, people hide mistakes rather than surfacing them, avoid experimenting, and quietly disengage, and every one of those is fatal to a programme whose improvement loop depends on humans reporting what went wrong.

What builds the opposite

Ask-me-anything sessions where nothing is off limits. "Will this replace my job?" needs an honest answer, not a corporate one. If the honest answer is "some of this work goes away and here is the path", say that.

Agent diaries. A short, honest weekly note on what the agent did, where it worked and where it stumbled. The stumbles are the point: a diary that only reports wins is marketing and everyone can tell.

Ask the frontline first. "If we put an agent here, what is the worst thing that could happen?" They know. They are usually right. And asking converts the most credible sceptics into the people who designed the guardrail.

What leaders are actually signalling

People read behaviour, not memos.

Leaders using agents as status symbols while conducting quiet layoffs tell a story: this is a cost-cutting instrument.

Leaders sharing their own agent failures and inviting critique tell a different one: this is a capability we are building together.

Both stories spread at the same speed and only one of them survives contact with the first incident.

The anxiety is real and it is not irrational

Work provides purpose, structure and belonging, not only income. When it is threatened, identity is threatened, and a substantial share of people in knowledge-work functions report genuine anxiety about it.

Even people who believe they will not be replaced report stress about keeping up. That is a different problem with a different fix, and it is the more common one.

Most of the anxiety comes from uncertainty and from not being involved in the decision, which are both things a programme can actually address. Neither is fixed by reassurance.

_{From the practice canon, developed in collaboration; the frameworks stated here are the author's own. Organisational cases are described from public reporting without naming the organisations.}

_{Back to Home · NORTH and the trust ladder · The talent stack}