Why do your best people resist AI? Usually because they have standards and the tools are not meeting them. Treat AI resistance in agencies as a people problem and you get mandates, training days and burnt-out seniors redoing machine work at 7pm. Treat it as a standards problem and your resisters become the people who make AI good enough to ship.
The quality stakes have rarely been more public. In Marketing Week last month, BBH co-founder Sir John Hegarty called much of today's marketing "repetitive, cliché, boring". He argued the industry is now staffed by "left brain thinkers" who understand spreadsheets but not ideas, and noted that "something like 95% of new product launches fail", a rate he says has not changed in 20 years despite far more data. Pour mediocre AI output into that system and you get more of the same, faster.
Why do experienced people resist AI tools?
An agency founder speaking at an AI adoption session in Singapore this week was refreshingly candid about this. He spent 18 months trying to make his roughly 20-person agency AI-native and pushed hard for adoption. When he looked closely at who was resisting, he found they were not technophobes. They were resisting bad output: the system was producing poor work and they were being made to use it.
These were people five or six years into the job. They knew what good looked like. When the AI draft fell short, they had to redo the work themselves, so they went back to the old way. In his telling, AI resistance often simply meant the tools did not meet the team's standards. Behaviour change fails when quality fails.
That is a useful reframe for any agency MD. The account director who will not send an AI-written status report, or the strategist who rebuilds the AI's persona deck from scratch, is not being difficult. They are doing your quality control, unpaid and unthanked.
Is forced AI adoption making agency burnout worse?
It can. Done badly, AI adoption doubles the work: produce the draft with the tool because you have been told to, then fix it by hand because the client cannot see it like that. Add the guilt of "not adopting fast enough" and you have a recipe for exhausted senior staff.
Done well, it does the opposite. In a Forbes Agency Council piece on how agencies stay essential, Multiply's Matt Jayson singled out campaign setup and QA as the most automatable work: "For one high-growth client, that's easily 20 hours a week of invisible work." That is the drudgery AI should take away, alongside timesheets, status reports, resizing, budget tracking and competitor audits.
The difference is whether the output meets the standard without a second shift of rework. If your team is working longer since you rolled out AI, the tools are not saving time. They are moving it to the evening.
Why does trust in AI break down so quickly?
Trust Theory offers a plain explanation. We trust a person or a system when we believe it is competent, reliable and acting in our interest. Each of those is earned through experience and conversation; none can be announced. A tool that produces three poor drafts in a row loses its claim to competence. A tool imposed without asking the team loses the sense that it is on their side.
That is why mandates backfire. A usage dashboard tells leaders who logged in. It cannot tell them whether people trust the output. Only conversation does that: sitting with the strategist and asking what she changed, and why.
Clients run the same calculation. The Singapore founder's point was that you cannot send half-baked work to customers, because that is how you lose their trust. Your resisters are protecting the client relationship.
Who should decide what "good" looks like?
The people who already know. The founder's fix was to give subject-matter experts an explicit role as approvers: they define what good looks like, and nothing ships without their sign-off. Everyone benefits from AI, but not everyone needs to build it. The experts set the bar.
It also helps to read the familiar signs of resistance as signals about standards rather than attitude.
| What resistance looks like | What it usually means | What to do |
|---|---|---|
| People quietly return to the old template | The output fails their quality bar | Ask to see their edits; fix the system, not the person |
| Drafts take longer to fix than to write | AI is adding work, not removing it | Measure editing time; pause the workflow until acceptance rises |
| Nobody will put AI output in front of a client | They cannot see where the conclusions came from | Require a traceable source for every AI claim |
| Seniors opt out while juniors over-rely | The standard was never written down | Make experts approvers who define what good looks like |
| Jokes about "the robot" in every all-hands | The tool was mandated without conversation | Involve the team in choosing and testing tools |
The Forbes panel's recurring themes point the same way: accountability, judgement and owning the work AI cannot be held liable for are what keep agencies essential.
How do you turn AI resisters into approvers?
- Find your loudest sceptics and ask to see their edits. The difference between the AI draft and what they actually sent is your specification for "good".
- Write the standard down. Tone, structure, sources, what must never appear. If it lives only in a senior's head, neither AI nor a new hire can meet it.
- Make the experts approvers, with real authority. They sign off before anything AI-assisted reaches a client, and they can pull a workflow that is not meeting the bar.
- Measure editing time, not log-ins. If fixing takes longer than doing, stop and repair the workflow before rolling it out further.
- Feed corrections back. Turn every approver edit into better prompts, templates and examples, so the next draft needs less fixing.
- Start with the invisible work. Setup, QA, reporting and monitoring, where a miss is cheap and the time saved is felt at once.
- Talk about it openly. A fortnightly half-hour on what worked and what did not builds more trust than any rollout email.
What makes AI output trustworthy enough to ship?
Evidence your experts can check. A senior strategist will accept an AI conclusion when she can see where it came from and judge it herself. She will reject one that arrives as a confident paragraph with no trail.
That is why on the Defy platform every AI conclusion traces back to the real published posts behind it. Strategy is reverse-engineered from what audiences and competitors actually publish, not from surveys, so the approver can open the evidence and decide for herself.
Humans, the founder said, are not the best at change. They are the single biggest blocker and the single biggest enabler of AI transformation. Which one your people become depends on whether the tools respect their standards.
