Briefing · 12/06/2026

Why employees hesitate to use AI at work

AI adoption stalls when leaders fund tools without changing incentives, workflow ownership, review points, and staff trust. COO and IT leaders need to solve the operating model behind the rollout.

Many COO and IT leaders are stuck between two bad signals.

The business is spending money on AI. Boards, owners, and executives expect productivity gains. Staff are experimenting, but the use cases that looked obvious in a demo often fail to become normal work.

At the same time, the workforce is nervous. Some people fear they will be judged for not using AI. Others fear they will be judged for using it badly. A third group sees the uncomfortable part more clearly: if they use AI well, they may prove that part of their job, or part of another team’s job, no longer needs the same structure.

This is showing up in ordinary work, far from the strategy workshop.

A support team uses AI to summarize tickets, but nobody has decided whether the summary becomes part of the customer record. A sales coordinator drafts quote notes with AI, but finance still owns price control and margin checks. A practice manager uses AI to turn referral letters into recall tasks, but privacy and clinical review rules are unclear. A project team uses AI to prepare status reports, then spends the same meeting arguing over whether the report can be trusted.

That is the adoption problem most AI strategy decks skip.

Two questions matter.

First: “Why aren’t people using the tools?”

Second: “Why would a rational employee want to expose that their work can be redesigned before the organization has explained what happens next?”

What The Data Shows

Microsoft’s 2026 Work Trend Index surveyed 20,000 AI users across 10 markets and analysed anonymized Microsoft 365 productivity signals. The headline is useful for operators: people are often ahead of the systems around them.

Microsoft states the problem directly: “People are ready. The systems around them are not.”

Only 19% of AI users sit in Microsoft’s “Frontier” zone, where individual AI capability and organizational readiness reinforce each other. About half sit in the emerging middle. Ten percent fall into what Microsoft calls “blocked agency”: skilled AI users inside organizations that have not built the conditions for them to apply those skills properly.

Microsoft also found that organizational factors such as culture, manager support, and talent practices accounted for 67% of reported AI impact, compared with 32% from individual mindset and behaviour. The strongest signal was the environment around the worker, not the worker alone.

The incentive problem is visible in the same report. Sixty-five percent of AI users said they fear falling behind if they do not use AI to adapt quickly. Forty-five percent said it feels safer to focus on current goals than redesign work with AI. Only 13% said they are rewarded for reinventing work with AI even when the results do not immediately land.

Those numbers explain why AI can be everywhere in conversation and thin in actual operations.

IBM’s 2026 adoption-challenges analysis points in the same direction from the technology side. Organizations are moving from generative AI experiments toward agentic AI systems, but common blockers include data quality, governance and security, ROI pressure, skills gaps, and workflow integration. IBM’s phrasing is blunt: AI capability is advancing faster than organizational capability.

IBM’s June 2026 CIO and CTO study makes the same pressure visible inside IT. Two-thirds of surveyed technology executives said they are accountable for AI systems they do not fully control. Seventy percent said business teams are deploying technology faster than IT can track. Only 11% said they are fully ready for expected agent deployment at scale. AI spend is projected to rise from just under 15% of IT budgets in 2025 to nearly 25% by 2027.

McKinsey’s 2026 AI Trust Maturity Survey adds the risk-management layer. It found that responsible AI maturity has improved, but strategy, governance, and agentic AI controls lag. Only about a third of surveyed organizations reached level three or higher in strategy, governance, and agentic AI governance.

Gallup’s February 2026 survey of 23,717 U.S. employees adds the employee view. Half of U.S. employees now use AI at least sometimes. Among employees in AI-adopting organizations, 65% said AI improved productivity and efficiency. Yet only about 1 in 10 strongly agreed that AI had transformed how work gets done in their organization.

The pattern is consistent: AI tools have arrived faster than the operating model needed to absorb them. Individual task gains are real. Organizational redesign is lagging.

This Is What Normal Looks Like Right Now

If the current state feels messy, that does not mean your organization is uniquely behind.

It means you are in the middle of the adoption gap.

The common pattern looks like this:

the board or owner wants visible AI leverage;
the COO needs measurable operational improvement;
IT is asked to enable tools while reducing risk;
managers are told to find use cases without being given workflow authority;
staff try AI privately because the official path is unclear;
strong users get faster, but their improvements stay local;
anxious users wait because the personal upside is vague;
nobody owns the redesigned process.

That is why a company can have AI licenses, training sessions, internal champions, and Slack enthusiasm while still seeing weak traction.

People are not necessarily resisting the technology. Many are resisting ambiguity.

The ambiguity is practical. Can AI touch customer data? Can it draft external messages? Can it classify an exception? Can it update the CRM? Can it summarize a complaint? Can it write policy material? Can it prepare a quote? Can it recommend a roster change? Can it generate board-pack commentary?

Each question crosses a different boundary: privacy, customer trust, legal risk, pricing control, workforce planning, compliance, or management accountability. A general “use AI responsibly” policy does not answer enough of them.

The Employee’s Incentive Problem

From the employee’s point of view, AI adoption can look like a trap.

Use AI well and you may raise the performance bar for everyone. Use it poorly and you may create errors, data leaks, or embarrassing outputs. Use it quietly and you risk breaching policy. Use it openly and you may invite scrutiny of your role.

For many workers, the safest move is cautious, private, low-stakes use: summarize notes, rewrite emails, clean up drafts, ask for ideas, and avoid touching core workflow decisions. That looks like weak adoption from the boardroom. From the worker’s seat, it is often rational.

This explains the strange split many managers are seeing.

The employee who uses AI to rewrite a difficult email is comfortable. The same employee may hesitate to use AI to draft a customer response, because now the output has a consequence. The analyst who uses AI to explain a spreadsheet formula may avoid using it to prepare financial commentary, because mistakes will carry their name. The coordinator who uses AI to tidy meeting notes may avoid using it to redesign the handoff they inherited, because that might step on another team’s work.

Team adoption has the same problem at a larger scale.

A whole team will lean into AI when the upside is shared and the rules are clear. They need to know what counts as good use, which work can be redesigned, who owns the output, how mistakes will be handled, and whether productivity gains will translate into better work or fewer people.

If those answers are missing, AI becomes a political object. The ambitious people use it to pull ahead. The anxious people avoid it. Managers get uneven results. IT inherits risk without control. Operations gets pressure to produce efficiency without a stable method for changing the work.

This is also why fear is not solved by a motivational launch meeting. BCG’s 2026 analysis estimates that 50% to 55% of U.S. jobs may be reshaped by AI over the next two to three years. Gallup found that 18% of all U.S. employees, and 23% of employees in AI-adopting organizations, think their job is very or somewhat likely to be eliminated within five years because of AI or automation.

That fear is part of the operating environment. Leaders can work with it directly or let it become quiet resistance.

Why Solo Operators Move Faster

Solo operators have the cleanest adoption path.

A founder, consultant, or independent operator can use AI to draft copy, write code, summarize calls, clean spreadsheets, design assets, prepare client material, and build small automations. The same person owns the request, the tools, the judgment, the output, and the consequences.

That does not make the work risk-free. It makes the coordination problem small.

An organization works through roles and handoffs. Marketing depends on operations for lead processing. Sales depends on finance for quoting. Service teams depend on compliance for review. Managers rely on those handoffs because they create visibility, accountability, and control.

AI changes the boundary. A person in one team can now complete parts of a workflow that used to pass through another team. Sometimes that produces faster and better work. It can also remove the review point, record trail, or team ownership that previously made the workflow governable.

That is where adoption starts to hurt.

The COO And IT Manager Squeeze

The COO feels the pressure to turn AI spend into operating leverage. That means fewer manual steps, faster throughput, better service, lower rework, and cleaner management information.

The IT manager feels a different pressure. They need to enable useful AI without creating a shadow stack of unapproved tools, copied customer data, unmanaged agents, unclear retention, and workflows nobody can audit.

Both are being asked to make AI real while the organization is still unsure what “real” should mean.

The pinch is visible in five common scenes.

In customer support, AI can summarize a ticket and propose a response. The support manager wants speed. IT asks where the customer data went. Legal asks whether the response was reviewed. The agent asks whether their judgment still matters.

In sales operations, AI can draft qualification notes and quote assumptions. Sales wants less admin. Finance wants margin control. Operations wants clean CRM data. The coordinator wonders whether the new process removes the work that made them valuable.

In finance, AI can turn month-end commentary into a cleaner board-pack draft. The CFO wants a faster close. The analyst worries that one hallucinated number will be treated as their mistake. IT worries about spreadsheet uploads and retention.

In HR, AI can summarize performance notes, draft role descriptions, and sort policy questions. HR wants consistency. Staff fear surveillance. Managers need help, but they do not want a machine making sensitive judgments.

In field operations, AI can convert job notes into service records, safety observations, and follow-up tasks. The operations team wants cleaner handoffs. Frontline workers do not want another reporting burden wrapped in AI language.

The weak version of AI adoption gives every team a tool, runs training, announces acceptable-use rules, and waits for use cases to appear. That approach produces scattered productivity stories. It rarely produces durable operating change.

The stronger version starts with the work.

Choose one workflow where the pressure is already visible: inbound lead handling, support triage, quote preparation, complaint response, client onboarding, roster notes, monthly reporting, procurement checks, or knowledge-base maintenance.

Map the current steps. Name the owner of each step. Identify the data involved. Mark the review points. Then decide where AI may assist, what it may produce, and who checks the result before it affects a customer, a record, a payment, or a compliance obligation.

This is where COO and IT interests meet. Operations gets a practical redesign. IT gets boundaries, access control, logging, and supportability. Staff get a reason to participate beyond “be more productive or be left behind.”

What Good Adoption Asks From Employees

Employees need a better bargain than vague encouragement.

The organization should be able to say:

which parts of the work are open for redesign;
which AI uses are approved;
what data must stay out;
who reviews AI-assisted output;
how mistakes will be handled;
how improvements will be credited;
what happens when a workflow becomes materially faster.

That last point matters. Redundancy fear is not irrational. Some jobs will change. Some tasks will disappear. Some teams will need fewer people doing the old version of the work.

Pretending otherwise makes adoption harder. People can smell the evasion.

The practical leadership move is to separate task change from person value. If AI reduces the need for manual status updates, duplicate data entry, first-draft writing, or inbox sorting, say so. Then show where human judgment, customer context, exception handling, relationship work, risk review, and process ownership become more valuable.

That is not a guarantee that nobody’s role changes. It is a more honest basis for participation than pretending AI is only a harmless assistant.

The better bargain sounds more like this:

We are going to remove low-value manual work where we can. We are also going to name the work that still needs human ownership: judgment, customer context, exception handling, review, escalation, process improvement, and accountability.

That message will not remove every fear. It gives people a framework that is more believable than pretending nothing important is changing.

A Better First Pilot

The first serious AI pilot should be narrow enough that a manager can answer six questions:

What workflow is in scope?
What decision or output will AI support?
Which data can the tool access?
Who reviews the output?
What record is kept?
Who can stop or change the workflow?

For example:

For inbound service requests, AI may summarize the request, extract missing fields, classify urgency, and draft an internal handoff. A coordinator reviews before any customer response or system update. The ticket records AI assistance, reviewer, changes made, and final routing.

That pilot is not glamorous. Good.

It gives employees a clear place to use AI without guessing the rules. It gives IT a bounded support and governance surface. It gives operations a measurable process change. It gives leadership a way to discuss impact without pretending the org chart is untouched.

The same pattern works for other first pilots:

Quote preparation: AI drafts assumptions and missing-information questions. Finance owns pricing rules. Sales owns customer context. A manager reviews before anything leaves the business.
Complaint response: AI summarizes the complaint and extracts facts. A human writes or approves the response. The record shows what AI produced and what was changed.
Monthly reporting: AI drafts commentary from approved source data. The analyst verifies numbers, adds interpretation, and marks any uncertainty.
Knowledge-base maintenance: AI proposes article updates from support trends. Support leads approve changes. IT controls where source data is pulled from.
Roster notes: AI summarizes availability, constraints, and conflicts. Managers make staffing decisions. Sensitive worker data stays inside approved systems.

The common feature is not the tool. It is the operating wrapper around the tool.

The Current Read

AI adoption is stalling because many organizations are asking employees to change work before they have changed the system around work.

The investment pressure is real. The staff anxiety is real. The failed-pilot fatigue is real. The way through is not another general training session or another list of use cases.

The useful unit is a named workflow.

For that workflow, a COO and IT manager should be able to say: this role may use this AI, with this data, for this step, under this review, with this log, and this off-switch.

That is the operating model staff can trust. It is also the one leaders can manage.

Sources

Was this useful?

Quick signal helps Rob sharpen future briefings.