Back to Guides
Enterprise Software 15 min read

The skeptic in row three: handling resistance in an AI workshop

The skeptic in row three: handling resistance in an AI workshop

Three AI workshop objections cover almost every room: this will take my job, the output is wrong, and I do not have time for this. They sound like one problem and they are three, each needing its own answer. Treat them as one lump of attitude and the arms stay folded all morning.

The person who matters most is in the third row, said nothing in the introductions, and has not opened a laptop. They are rarely the least able person present. They are often the most senior, with the most context about the work, and the sponsor needs them most.

Resistance usually starts with the invite

Most of the room’s posture is set by the calendar invite. Three lines in a sponsor’s announcement reliably produce a defensive room.

“Mandatory AI training.” Mandatory answers a question nobody asked out loud: what happens if this does not work for me. It also frames the session as compliance, and people arrive accordingly.

“To help us do more with less.” Whatever was meant, this reads as a headcount sentence. If a reduction plan exists, people already suspect it. If one does not, the phrase invents the suspicion.

No named outcome. An invite promising to explore AI tells a busy person their afternoon belongs to someone else’s curiosity.

The fix costs one paragraph: name the tasks the session will work on, say the material is the team’s own recent documents, set attendance as expected rather than compulsory, and state the company’s position on roles. Our guide to designing a session that changes Monday morning treats that announcement as part of the design, because it caps what the room will attempt.

How the invite lands is measurable. Slack’s Fall 2024 Workforce Index, covering more than 17,000 desk workers, found 48% uncomfortable telling their manager they had used AI for a work task, because it felt like cheating (47%) or would make them look less competent or lazy (46% each). A room carrying that will not volunteer its real questions until the sponsor removes the penalty.

The three AI workshop objections need three different answers

The peer-reviewed framing beats the persona charts most vendors publish. Golgeci, Ritala, Arslan, McKenna and Ali, in a 2025 integrative review in Human Resource Management Review (35(2), article 101075), model resistance to AI at work as three separate things: fear, low perceived efficacy, and antipathy toward the technology. Each has its own counter. Fear responds to legitimation, low efficacy to access and practice, antipathy to seeing the tool augment a person rather than replace one.

In a room, that maps onto what people say:

What you hear What it is What works What backfires
“This is going to take my job” Fear The sponsor naming the company’s position on roles, before any exercise A facilitator promising AI will replace nobody, which is not theirs to promise
“I am not technical enough” Low perceived efficacy A four-minute exercise the person finishes on their own document Telling them it is easy, which lands as a judgment
“The output is wrong and I will fix it anyway” Evidence-based antipathy Agreeing, then running the failure live and naming the tell A polished demo chosen because it works
“I do not have time for this” Fear or low confidence in a scheduling costume Asking which task they would hand over if it worked, then building on that Arguing about the calendar

The first row is not the facilitator’s to handle, and the fear behind it is not irrational. Pew Research Center, surveying 5,273 employed US adults and publishing on February 25, 2025, found 32% expecting workplace AI to mean fewer job opportunities for them long run against 6% expecting more. The job question needs whoever authorized the session, in the room, saying one or two sentences they will be held to. A vendor cannot borrow that authority, and a room can tell when someone tries.

The last row matters because the time objection is the one most often taken at face value. Someone with genuinely no time will still name a task they wish they could hand over, and that answer says whether you face fear or a calendar.

The correct objection is the most useful thing anyone says

The best outcome of a session is not a room of enthusiasts. It is a room that knows where the tool fails on their own work. Enthusiasm does not survive the first real deadline; a known failure list does.

The skeptic who says the output is wrong is usually right. Stack Overflow’s 2025 developer survey, with roughly 49,000 respondents, found 46% actively distrust the accuracy of AI tools against 33% who trust it, and only 3.1% in the highly-trust band. The top frustration, named by 66%, was “AI solutions that are almost right, but not quite.” That is a technical sample, and the wider pattern runs the same way: KPMG and the University of Melbourne, surveying more than 48,000 people across 47 countries, put regular intentional use at 66% and willingness to trust at 46%. Your room has met the almost-right answer and lost an hour to it.

Behavioral research explains why reassurance makes this worse. Dietvorst, Simmons and Massey, in the Journal of Experimental Psychology: General (144(1), 114 to 126, 2015), found that people abandon an algorithm faster than a human forecaster after watching both make the same mistake, even when the algorithm is more accurate. Seeing the error does the damage, so a session built on clean demos only moves that moment to a desk, with nobody there to read it. Their 2018 follow-up in Management Science (64(3)) points at the fix: aversion drops when people can modify the thing. In a room, that means handing over control of the failure.

So schedule it. Give the room twenty minutes to break the tool on genuine company material: the paraphrase that changes a contract term, the number it invents when the source is silent, the table it reads backwards. Collect what they find, by name, on a list the sponsor keeps. Row three will have the best entries, because those failures only surface against real domain knowledge, the argument behind running exercises on last week’s real work rather than toy prompts. Our library of exercises worth running treats this verification drill as load-bearing.

Three things come out of that. The skeptic is the expert instead of the holdout. The room leaves with a written list of where not to trust the tool, which is the start of an internal rule. And the sponsor sees their own documents in a way no vendor assessment shows them.

Why the skeptic is often your best performer

Resistance tracks tenure more than ability, which is a commercial fact rather than a people problem. Someone with ten years in the work holds a sharper model of what correct looks like, so they catch the almost-right answer faster. They also have more to lose from a public mistake and less to gain from looking modern. The eager adopter in the front row is often fluent at prompting and weak at spotting a wrong answer, the more expensive gap.

The money follows the senior skeptic. They own the workflow you want changed, their team copies their habits, and if they leave unconvinced nothing downstream moves. Spend the session there, not on the people already nodding.

One design choice follows. With their permission, put that person’s own document on the screen for the verification drill and let them narrate what the tool got wrong. It is the strongest demonstration available and it costs nothing, which matters when the goal of a short session is behavior change rather than coverage.

The mandate trap

The default answer to resistance is to make the training required. The evidence says compulsory AI enablement buys attendance and little else.

A LinkedIn survey reported by Fortune on August 28, 2025 found 51% of professionals calling the intensity and frequency of AI training requirements excessive, and 35% nervous talking about AI at work for fear of sounding uninformed. Volume produces fatigue and shame at once, which is poor ground for a room where people have to admit what they cannot do.

The usage data points the same way. BCG’s AI at Work 2025 study of 10,635 employees across 11 countries, published June 26, 2025, found 72% using AI at least several times a week but only 36% saying their training was enough. Regular use ran at 79% among those with more than five hours of training, against 67% below that line. Dosage changes behavior; compulsion changes the attendance sheet.

What works is narrow and dull: expected attendance with a named business reason, a session short enough that the time objection dies, and a route onward for anyone who wants more. Keep the exit visible, because someone who could leave at the break and stays is participating.

The first ten minutes

The opening block decides most of the rest. The moves that hold up:

  1. Have the sponsor open, then leave or stay by design. Two minutes on why the session exists and what it is not about. If they stay, people will not admit ignorance. If they leave first, the job question goes unanswered. Answer it, hand over, stay out of the exercises.
  2. Name the hard objection yourself. Say that some people here think this will replace parts of their job, that the concern is fair, and that the session will not pretend otherwise. An unnamed fear runs the room from the back.
  3. Make the first exercise unloseable. Four minutes, one real document, one instruction, a visible result. Low confidence is a belief about oneself and it moves on evidence, not encouragement.
  4. Invite the failure out loud. Ask who has already had one of these tools produce something wrong, and spend three minutes on it. Cheapest credibility a facilitator can buy.
  5. Do not take the bait on the wider AI argument. Energy use and training data are real questions a working session will not settle. Acknowledge, park it in writing, return to the document.

Two moves that look helpful and are not. Opening with a slide on how the technology works answers a question nobody asked. And pairing the skeptic with an enthusiast produces one person typing and one watching, which confirms the skeptic’s view of their own place in this.

What a session cannot fix

Naming the limits is how a sponsor avoids paying for the wrong thing. Four kinds of resistance are not an enablement problem.

An unannounced reduction plan. If roles are being cut and the room half-knows it, facilitation will not produce trust, and trying damages the sponsor. Announce first, or move the session.

A tool policy that does not exist. If nobody can say which assistant is sanctioned and what may be pasted into it, the careful people are right to hold back. A written rule is a thirty-minute decision that retires the whole category.

Personal accounts instead of company ones. Asking people to run work exercises on a login they pay for is a fair thing to refuse, and no facilitator can argue them out of it.

Antipathy that is a considered position. Some people have thought about this and would rather not. Team capability does not need unanimity, which is the wider point in our case for building capability rather than buying licenses.

What to ask a vendor before you sign

Four questions, and the answers that should worry you.

Ask A good answer Walk away if
How do you handle the person who thinks this threatens their job? You answer it in the first five minutes, and we tell you what to say The facilitator offers to reassure the room
What happens when someone shows the tool getting something wrong? There is a scheduled block for it, and we keep the findings It is framed as a disruption to manage
What do we get if half the room stays skeptical? Your failure cases in writing, plus a shorter list of tasks worth handing over A satisfaction score
Who is the hardest person you have had in a session? A specific, unflattering story A story where everyone was won over

The last one sorts practitioners from presenters. Anyone who has run enough rooms has lost one, and the useful answer names what changed after.

Frequently asked questions

How do you handle an employee who refuses to use AI tools?

Separate refusal from objection. An objection is information and belongs in the session. A flat refusal once the policy, the training and the accounts are in place is a performance conversation. Most apparent refusals are an unanswered question about roles or an unwritten rule about data.

Should AI training be mandatory?

Expected rather than mandatory, in most cases. A LinkedIn survey reported by Fortune in August 2025 found 51% of professionals already call the volume of required AI training excessive. Compulsion produces attendance. A named business reason and a session worth the time produce use.

What is the most common objection in a corporate AI workshop?

Doubt about accuracy, often with job fear underneath it. Stack Overflow’s 2025 survey found 46% of developers actively distrust the accuracy of these tools against 33% who trust them. Accuracy is the objection a room will say out loud.

Is it a bad sign if people push back during the session?

The opposite. Silence usually means the room has decided to sit the session out. Pushback means people are measuring the tool against work they care about, the condition any behavior change needs.

How do you deal with a mixed room of skeptics and enthusiasts?

Same exercise, different jobs inside it. Ask the doubter to judge whether the output would survive their own review, and the fast adopter to produce the first draft. Pairing them as helper and helped leaves the doubter watching.

Should the executive who booked the session be in the room?

For the first five minutes, yes, to answer the job question with their own authority. For the exercises, usually not, because people will not admit what they cannot do in front of whoever writes their review. Say which in the invite.

Can one session change a resistant team?

One session can change what a team believes about the tool’s limits, and that part lasts. Daily habits need more exposure: BCG put regular use at 79% above five hours of training against 67% below, so plan the follow-up first.

How much does a session like this cost?

Team workshops run roughly $2,000 to $15,000 at market rates, set by room size, how much of the design is built from your material, and whether anything follows. Resistance handling is not a line item. It comes down to who opens the session and what the exercises run on.

Key takeaways

  • Three objections cover most rooms and need three answers: fear needs the sponsor’s authority, low confidence needs an exercise the person finishes, and evidence-based doubt needs the failure shown rather than hidden. Golgeci et al. (2025) make the same split, which is why one-size advice does not land in a live room.
  • The accuracy objection is correct. Stack Overflow’s 2025 survey put active distrust at 46% against 33% trust, with almost-right output the top frustration for 66%.
  • Schedule twenty minutes for the room to break the tool on real company material and collect the findings in writing. Dietvorst et al. found that watching an algorithm err costs it more confidence than the same error costs a person (2015), and that letting people modify it wins some back (2018).
  • The senior skeptic is the most valuable person to persuade: they own the workflow and their failure cases are the sharpest available.
  • Mandates buy attendance. BCG put regular use at 79% above five hours of training against 67% below, while a 2025 LinkedIn survey found 51% already call required AI training excessive.
  • An unannounced reduction plan, an unwritten tool policy and personal accounts instead of company ones get fixed before the room, not in it.

Before the next session goes on the calendar, read the invite you are about to send and ask what row three will hear in it. For a second opinion on that paragraph, and on which weekly tasks the exercises should run on, book a free AI-readiness call.

Last Updated: Nov 13, 2026

AW

Arthur Wandzel

SFAI Labs helps companies build AI-powered products that work. We focus on practical solutions, not hype.

Make your team fluent in AI — then automate what proves out

  • Hands-on training applied to your own documents and workflows
  • Often reimbursable through state small-business training programs
  • Built for 5–50 person firms with no IT department

Related articles