Resistance to AI looks like agreement that never turns into action. Treat it as a power problem. Give the experts ownership of the rules, let teams switch individual checks off, keep every decision small and reversible, and answer the question about jobs early, in plain words.
In most organisations, resistance to AI doesn’t look like objection. It looks like a slow yes.
The team lead agrees to the pilot, then keeps moving the training date. The analyst uses the new tool, but keeps his old spreadsheet going “just in case”. IT supports the project in principle and then takes three months to answer a question about access. Nobody has said no. Nothing is moving.
If you only plan for open opposition, you’ll miss almost all of it.
Follow the power
Rosabeth Moss Kanter argued in Harvard Business Review back in 1979 that powerlessness can corrupt too. People who feel their influence shrinking tend to protect what they have left. In change projects, a real or perceived loss of power is one of the most reliable signs that resistance is coming.
AI shifts power in ways that aren’t always obvious. The biggest shift usually isn’t between departments. It’s the move from knowledge held in people’s heads to knowledge held by the organisation.
Think about the experienced person who knows which system to check, which data to distrust and which customer is likely to cause trouble. That knowledge makes them valuable. A tool that captures it makes them feel less so, however carefully you explain the benefits.
Researchers call the underlying feeling psychological ownership. When people feel that a piece of work is theirs, changing it feels like changing them. That’s not irrational. It’s worth taking seriously.
Make them owners, not subjects
A better communications plan won’t fix this. Giving people real control over the thing that’s changing will.
Make the experts the owners of the new version of their knowledge. Let them write and maintain the rules. Let the team switch individual checks off if they don’t trust them. A group that can turn something off doesn’t need to resist it. You lose a bit of consistency, and you gain the sense of ownership that actually predicts whether people will use it.
Ask the people doing the work to define what “done” or “ready” means. That’s influence many of them don’t have today, and it tends to turn critics into co-authors.
Keep each decision small
Big requests invite slow yeses. Small ones get answered.
Ask for system access one system at a time, read-only first, with a clear way to reverse it. Release funding in stages, tied to decisions. Ask your sponsor for something specific at each stage rather than general support, so their backing stays active rather than nominal.
Give credit away. Let the sponsor present the early wins. Send the management information to the managers whose teams produced it before anyone else sees it. Routing results to the board over their heads makes an enemy for no gain.
Say the difficult thing early
The question people most want answered is usually the one leaders avoid: what happens to my job?
Answer it early, in plain words. Make sure the version the finance director hears is the same as the version the frontline team hears. People compare notes. And think hard before attaching a productivity target to the first phase. Nothing confirms people’s fears faster.
Know whose commitment you need
You don’t need everyone’s enthusiasm. You need genuine commitment from the few people who can make or break the project, and willing co-operation from everyone else. Map who those people are, what they stand to lose, and what they’d need to feel in control.
And be honest about your own position. If you built the first version, your job is to give it away. If it still belongs to you in twelve months, the change has failed, even if the tool works.
Is your AI project stuck in a polite queue?
Tell me what’s not moving. It’s rarely the technology, and there’s usually a smaller decision someone can say yes to.