Key Takeaways
- The real global figure is 61%. The 60% doing the rounds is BCG's Gulf regional cut, not a worldwide number.
- Only people who already knew what AI agents were and used AI regularly were asked — 8,849 of the survey's 11,749 respondents.
- Half of all respondents say they have a limited understanding of what an AI agent actually is.
- Managers are more convinced than the people doing the work: 65% against 52%.
- BCG's own economists forecast 10–15% of US jobs eliminated over five years — a serious number, and a much smaller claim.
- The finding that matters is time. A day a week is already being saved, and two thirds of workers are told nothing about what to do with it.
A statistic has been travelling fast this week: most workers think AI could do half their job. It comes from BCG's AI at Work 2026, the fourth edition of its annual global survey, and the real figure is 61%.
The number is accurate. The way it is being read is not. It is being passed around as a forecast of what AI can do. It is a measure of what a particular group of people expect — and once you look at who was asked, it says something considerably more interesting.
Where does the “AI could do half my job” stat come from?
BCG surveyed 11,749 workers across 14 markets and published the results on 3 June 2026. In its own announcement, the firm put the finding plainly: six in ten respondents believe agents could do at least half their job within three years, “yet more than half (52%) still have a limited understanding of what agents are.”
Those two facts sit in the same sentence in BCG's press release. Most of the coverage kept the first half and dropped the second.
Who was actually asked whether AI could do half their job?
This is the part that changes the meaning. Read the footnote on the relevant slide in BCG's published deck and it says the base is:
n=8,849 respondents who have heard of AI agents and regular AI users
So the question was not put to all 11,749 workers. It was put to the roughly three-quarters who already knew what an AI agent was and were already using AI regularly. The people least exposed to the technology — the ones whose jobs a sceptic would most want counted — were filtered out before the question was asked.
That does not make the finding worthless. It makes it a different finding: among people already using AI, most expect it to take over half their work. That is a genuinely useful thing to know. It is just not “AI can do half your job.”
| What people hear | What BCG measured |
|---|---|
| AI is capable of doing half your job | Workers expect agents could, within three years |
| Workers, generally | People who know what agents are and use AI regularly (n=8,849) |
| An assessment of the technology | A forecast by non-specialists — 52% report limited understanding of agents |
| It is happening now | A three-year horizon, from a June 2026 survey |
Why do workers think AI can do half their job?
Because a real thing is happening to their work, and this survey question is the nearest available way to express it.
BCG found 74% of frontline employees are now regular AI users, up 23 percentage points in a single year. 72% say the skills expected of them have already shifted — but only 36% feel they have had sufficient training. And 47% report their role has moved toward managing and directing AI rather than doing the work.
Read those together and “AI could do half my job” starts to look less like a prediction and more like a description of how much the job has already changed underneath them. That is closer to the distinction we draw between an AI agent and plain automation, and it is why what agentic AI actually means is worth understanding before reacting to a headline about it.
Who thinks AI will take over their job?
The split is the most revealing number in the whole report.
| Group | Say agents could do half their job in 3 years |
|---|---|
| Managers and leaders | 65% |
| All respondents | 61% |
| Frontline employees | 52% |
The people doing the work are the least convinced it can be automated. The people furthest from it are the most convinced. Anyone who has scoped an automation project will recognise that pattern immediately — the tasks that look trivially automatable from two levels up are usually the ones carrying the exceptions, the judgement calls and the undocumented context.
What does BCG's own research say about job losses?
Notably, something much calmer. In separate modelling, BCG concludes that AI will reshape more jobs than it replaces, estimating that 10% to 15% of US jobs could be eliminated over roughly five years, with most roles changing substantially rather than vanishing.
So the same firm, in the same year, publishes both “61% think agents could do half their job” and “10–15% of jobs eliminated in five years.” Those are not contradictory, because they answer different questions. One is a survey of feelings. The other is an economic model. Only the first went viral.
What is actually changing about work right now?
The genuinely actionable finding in this report has nothing to do with the headline. It is about time.
42% of frontline employees who use AI regularly now save a full day or more per week. That is a large, measured, present-tense result. And then: 66% receive limited or no guidance on what to do with the time they save, and more than half do not redirect it to higher-value work.
That is the failure worth worrying about, and it is not a technology failure. A day a week returned to every regular AI user is a serious operational gain. Left unassigned, it dissipates into slightly longer versions of the same tasks.
Nobody has ever called us because a survey said AI might do half their job. They call because quotes are going out three days late, or the phone rings after hours and nobody answers. Pick the thing that is actually costing you money, automate that, and measure the hours it gives back. Expectations are not a plan.
— Daria Morrison, Co-Founder, Avelle SolutionsWhat should a business owner do about it?
Very little, urgently. Do not restructure headcount around a three-year expectation survey. The useful response is smaller and duller:
- Pick one task, not a department. Repetitive, high-volume, rule-heavy, and currently late or dropped. Our guide on how to decide what to automate first covers the selection properly.
- Decide where the hours go before you start. This is the 66% failure. Name the work the reclaimed time is for, in advance.
- Check the economics honestly. Whether AI automation is worth it at your size depends on volume, not on what a survey says about the future.
- Separate automating from hiring. They solve different problems — see when to automate and when to hire.
- Know who maintains it. A system nobody owns quietly stops working; naming a maintainer is not optional.
If you are still working out the ground floor of this, start with what AI automation actually is and what can realistically be automated before worrying about agents replacing half of anything.
What should you watch next?
Two things. First, whether the 2027 edition of this survey shows the expectation falling as understanding rises — if people who genuinely know what agents do are less convinced, that tells you the 61% was novelty. Second, whether any employer starts reporting what happened to the reclaimed day.
Until then, treat “AI could do half my job” as what it is: a measure of how quickly work feels like it is changing. That feeling is real and worth taking seriously. It is just not a capability spec.
Methodology and sources. All figures come from BCG's AI at Work 2026 (fourth edition, published 3 June 2026): the full survey base of 11,749 workers across 14 markets, and the n=8,849 sub-base for the agents question, both read directly from BCG's published slide deck rather than from secondary coverage. Job-elimination estimates come from BCG's separate AI Will Reshape More Jobs Than It Replaces analysis. Where coverage cited “60%”, we traced it to BCG's Gulf regional cut released 2 September 2026 and have not used it as a global figure. No statistic in this article is unattributed.
Frequently Asked Questions
Not today, and the 61% figure is not evidence that it can. It records what workers expect within three years. BCG asked the question only of people who had already heard of AI agents and use AI regularly, and 52% of respondents overall admit a limited understanding of what an AI agent actually is. It is a measure of expectation, not of demonstrated capability.
From BCG's AI at Work 2026, the fourth edition of its annual survey, published 3 June 2026. The full survey covers 11,749 workers across 14 markets. The “half my job” question was put to a narrower base of 8,849 respondents who had heard of AI agents and were regular AI users.
Both numbers are real and they are not the same survey cut. 61% is the global figure. 60% is BCG's Gulf regional cut, which circulated in early September 2026 and covers GCC frontline workers specifically. If you see “60% of workers” without a region attached, it is the Gulf number missing its label.
Far fewer than “half of everyone's job” implies. BCG's own modelling estimates 10% to 15% of US jobs could be eliminated over about five years, with most roles changing substantially rather than disappearing. That is a serious number, but it is a different claim from the survey headline.
Yes, and by a wide margin. 65% of managers and leaders said agents could do at least half their job within three years, against 52% of frontline employees. The people furthest from the actual task are the most confident it can be automated.
Because the time was never reassigned. BCG found 42% of frontline regular AI users save a full day or more per week, but 66% receive limited or no guidance on what to do with it, and more than half do not redirect it to higher-value work. The saving is real; the value leaks away without a decision about where the hours go.
Nothing dramatic. Do not restructure headcount around a survey of expectations. Pick one narrow, repetitive, high-volume task, automate it properly, measure the hours it returns, and decide in advance what those hours get spent on.
Find Out What AI Can Actually Do In Your Business
Book a free 30-minute call. We’ll look at where your time is actually going, pick the one task worth automating first, and tell you honestly if it isn’t worth it yet.
Book Your Free AI Assessment →