5 Ways Managers Can Stop AI Fatigue Burning Out Their Team

AI was sold to most teams as a way to do the same work in less time. In many offices it has become a way to do more work in the same time. AI fatigue is the exhaustion that follows: a steady build-up of extra review, constant re-learning and rising expectations, carried by people whose working day is still eight hours long.

In our work with senior leadership teams across East Africa over the past decade, we have seen every new tool arrive with a rollout plan that covers licenses, training dates and usage targets, and says nothing about what the team will stop doing to make room. AI tools follow the same pattern, only faster, because they change every few months and each change brings something new to learn.

This article explains what AI fatigue is, why AI adds work before it saves any, the signs that a team is running on it, and five ways managers can prevent it. It also looks at how AI fatigue shows up in Kenyan workplaces and the three mistakes that make it worse.

What AI Fatigue Is

AI fatigue is the tiredness that builds up when AI tools change a job faster than the job is redesigned around them. It is not a medical diagnosis. It is a description of workload, and it usually has three sources:

• Extra work around the tool. Someone has to check what the tool produces, correct it and take responsibility for it. That checking rarely appears in anyone's job description.
• Constant learning. The tools change every few months, so people are always learning the next version, often in their own time.
• Higher expectations. Once a task can be done faster, the faster pace becomes the new normal, and the volume rises to match it.

AI fatigue overlaps with change fatigue, the weariness that comes from absorbing one initiative after another. The difference is that an AI rollout is rarely declared finished. A restructure ends; a new tool is followed by a new version, a new feature and a new tool. The people who use AI most, and who are often its most enthusiastic users, carry the most of this load.

Why AI Adds Work Before It Saves Any

The evidence from the last two years points the same way: AI tools add work first, and whether they save any depends on decisions most organizations have not made.

The clearest survey evidence comes from the Upwork Research Institute, which surveyed 2,500 workers and executives in the US, UK, Australia and Canada in 2024. Among employees using AI, 77% said the tools had decreased their productivity and added to their workload in at least one way. The extra work took three main forms:

• Reviewing output. 39% spent more time reviewing or moderating AI-generated content.
• Learning the tools. 23% spent more time learning how to use them.
• More work. 21% were being asked to do more work.

A closer look comes from Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye, who spent eight months inside a US technology company of about 200 people where staff had broad access to generative AI. Ye observed daily work and conducted more than 40 interviews. Their findings, published in Harvard Business Review in February 2026, showed AI intensifying work in three ways:

• Wider jobs. People took on tasks that used to belong to colleagues, or that nobody would have attempted before.
• Fewer pauses. Because a prompt is easy to send, work moved into lunch breaks, the minutes before meetings and the evening.
• More parallel work. People kept several threads running at once, with AI working in the background while they reviewed code, drafted documents or sat in meetings.

Much of this was never asked for. In the researchers' words, as reported by Berkeley Haas, employees worked faster and longer "often without being asked to do so." In the moment, the work felt exciting. Looking back, people described feeling busier, more stretched and less able to switch off. That is the trap: what starts as extra effort quietly becomes the standard everyone is measured against.

Where the saved time goes is the deciding question. BCG's 2025 AI at Work survey asked 7,492 people who use generative AI at work what they did with the time it saved them.

The most common answer was simply more tasks, at 54%. Fewer than three in ten used the time for their own satisfaction and well-being. BCG also found that 47% of these users save more than an hour a day, but only about one-third receive any guidance on how to use that time. Time saved by AI that nobody assigns is not saved. It is spent, usually on more of the same work.

The Signs of AI Fatigue in a Team

AI fatigue rarely announces itself. People who are excited about a tool keep saying so long after the excitement has turned into strain, and the strain shows up in the work before anyone mentions it. Five signs are worth watching for:

• Review queues that never empty. Drafts, summaries and code produced by AI pile up waiting for a person to check them, and the checking happens late or not at all.
• Work in the gaps. Messages and prompts arrive at lunch, late in the evening and on weekends from people who used to keep those hours free.
• Learning on personal time. People mention that they spent the weekend working out a new feature, and nobody asked them to.
• More output, more corrections. Volume rises, but so do errors, rework and complaints about quality.
• Your strongest users are often the most tired. The people who adopted the tools first and use them most have usually absorbed the most new work.

None of these needs a survey to spot. A manager who reads message timestamps, looks at the review backlog and asks people what they did with their last free afternoon will see most of them.

5 Ways Managers Can Prevent AI Fatigue

Most advice on AI fatigue is addressed to the individual: take breaks, go for a walk, talk to friends. That advice helps, but it puts the fix with the person who has the least control over the workload. The causes in the evidence above are management decisions, so the fixes are too.

1. Decide what the saved time is for

Before a tool goes live, agree with the team what the time it saves will be used for. Some of it can go to more output. Some should go to the work that keeps being postponed: the client review, the process fix, the training nobody has had time for. Some should simply return to people as a shorter or calmer day. Write the split down and check it after a month. If the answer is "more of the same tasks", the team has been given a faster treadmill rather than a better job.

2. Pace the tool rollouts

Every new tool and every major update carries a learning cost, even when it is free to install. Introduce one change at a time and let it settle before the next one arrives. A team that is still learning last quarter's tool should not be handed a new one because a vendor released it. A simple test helps: if the team cannot yet say what the last tool changed in their work, they are not ready for the next one.

3. Budget learning time inside working hours

If learning a tool is part of the job, it belongs in the working week. Put it in the calendar, protect it like a client meeting and expect it to take longer than the vendor's demonstration suggests. Learning that only happens on evenings and weekends tells people the organization values the tool more than their time.

4. Count review work as work

Checking what an AI tool produces is real work, and it carries real responsibility, since the person who signs off owns the result. Make it visible. When you plan a week, count the time it takes to review AI output alongside the time it takes to produce it. When you set targets, set them on finished, checked work rather than on drafts generated. A target that counts only output rewards people for skipping the check.

5. Keep the boundaries the tools blur

AI makes it easy to keep working from anywhere, at any hour. The Berkeley researchers recommend building in deliberate pauses before major decisions, batching updates that are not urgent and protecting time when people are not interrupted. A manager can go further by setting the example: no requests in the evening, no praise for the reply sent at midnight, and a clear statement that a prompt sent at 10pm is not a sign of commitment. A manager who praises the colleague answering prompts at 10pm has just written the team's new job description.

How AI Fatigue Shows Up in Kenyan Workplaces

None of the research above was carried out in Kenya, and we have found no comparable published study of AI fatigue here. The pressures are present all the same, and some local features make them harder to see.

• Overload is reported late. In many Kenyan organizations, seniority carries real weight, and junior staff rarely tell a senior manager that a new expectation is too much. They absorb it. A manager who waits to hear about AI fatigue will hear about it last, often through a resignation or a mistake.
• Work already reaches people after hours. In Microsoft's 2025 data, messages per person rose 6% in a year worldwide and by more than 20% in regions including Central and Eastern Europe, the Middle East and Africa. Where work also runs through WhatsApp groups on personal phones, as it does in many Kenyan teams, the evening boundary was thin before AI arrived.
• Learning often comes out of personal time and money. Where employers provide no paid tools or training, people learn on their own phones and data bundles, at night. That is commitment, but it is also unpaid work, and it adds to the load.

The practical answer is to ask directly. In one-to-one conversations, ask what the new tools have added to each person's week, not only what they have saved. Make it safe to say "too much" by acting on the first person who does.

Three Common AI Fatigue Mistakes

Measuring adoption instead of workload

Dashboards that track logins, prompts and active users show whether people are using a tool, not what it is costing them. High usage can mean the tool works, or it can mean people are working longer to keep up. Cure: track the work alongside the usage. Look at hours, review backlogs and rework, and ask people every quarter whether the tool has made their week lighter or heavier.

Raising targets before the gains are real

Some organizations raise output targets as soon as a tool is announced, on the assumption that it will make everyone faster. The gains arrive later and less evenly than the forecast, so the gap is filled with extra hours. Cure: measure what the tool actually changes over a full quarter, then set new targets from that evidence and agree them with the team.

Treating fatigue as resistance

When a tired team slows its use of a new tool, it is tempting to label the slowdown as resistance to change and push harder. Often it is a reasonable response to a workload that has grown with no end in sight. Cure: before pushing adoption, ask what the tool has added to people's work. If the answer is extra review, constant learning and evening messages, fix the workload first.

Frequently Asked Questions

Conclusion

AI fatigue is not a sign that a team cannot cope with technology. It is what happens when tools speed up the work and nobody decides what the speed is for. Managers who make that decision keep the gains and keep their people.

Think of the last hour an AI tool saved someone on your team. Do you know where that hour went, and did anyone decide it?

If you want your managers to bring in AI at a pace their teams can sustain, our Leadership and Organizational Change Management program can help. Request a proposal and we will shape it around the changes your teams are absorbing now.