Over the past two years, a certain kind of announcement became routine. A company would tell its staff or its investors that AI would let it run with fewer people, and a round of roles would disappear. In 2026, a quieter sequel has started to appear in the data: some of those roles are coming back.
A Robert Half survey of 1,365 Canadian hiring managers found that 34 per cent of those who had cut positions after adopting AI have since restored those roles, or very similar ones. Robert Half calls the pattern an "AI overreaction in the labour market," and Staffing Industry Analysts reported in August that the same pattern may be reaching the UK, where job postings rose between the first and second quarters in IT (42 per cent), logistics (25 per cent), utilities (21 per cent) and manufacturing (18 per cent).
This does not mean AI failed. It means the first round of decisions was made on a forecast, and the forecast is now meeting the work.
What the rehired roles tell us
The reasons managers gave for restoring positions are more useful than the headline figure. In Robert Half's survey, as reported by Canadian HR Reporter, the most common answers were:
AI needed more human oversight and quality control than expected (38 per cent)
Business demand grew and required more capacity (38 per cent)
The roles involved relationship management AI could not replicate (37 per cent)
AI tools were not fully or consistently adopted across teams (37 per cent)
The roles carried institutional knowledge or context AI could not replace (36 per cent)
Productivity gains fell short of expectations (35 per cent)
Risk or compliance concerns emerged without a human in the role (35 per cent)
Remaining team members experienced burnout or workload strain (33 per cent)
Read together, these fall into three groups, and they call for different responses.
The work changed shape rather than disappearing.
Oversight, quality control and compliance review are tasks that did not exist, or existed at small scale, before the tool arrived. Someone has to check the output, and that someone usually needs the expertise the cut role used to hold.
The business moved on.
More than a third of managers said demand grew. A headcount plan built on last year's volume will not fit next year's, with or without AI.
Uneven adoption and missing productivity gains point to implementation problems, which are fixable but not by removing people first.
The hidden workload: checking the machine
One finding from the same research explains much of the first group. In a separate Robert Half survey of 1,005 Canadian workers, employees reported spending 33 per cent of their working time checking accuracy and refining the final output when they use generative AI.
That is one-third of the working time, spent on a step that rarely appears in a business case. When a company projects savings from AI, the usual calculation covers the time a task takes to produce. It tends to leave out the time the result takes to verify. If checking absorbs a third of the saving, the case may still hold, but the number is smaller and the people doing the checking have to be somewhere on the payroll.
Cuts came first, and some leaders now regret it
Robert Half is not the only source suggesting second thoughts. Orgvue's 2025 survey of more than 1,000 senior leaders across several countries, including Canada, found that 39 per cent had made job cuts because of AI and that 55 per cent of those leaders said the cuts were the wrong decision. The "over half" figure applies to that 39 per cent subgroup, not to all leaders surveyed, and Orgvue did not report any rehiring. It did find that 80 per cent planned to reskill staff for AI collaboration, a plan rather than a result.
Canadian employers are also still in motion. Littler's 2026 Canadian Employer Survey, published October 21 and based on 200 executives, in-house lawyers and HR professionals, found that 44 per cent had reassessed or were reassessing job responsibilities in response to actual or anticipated increased AI use. Thirty-six per cent had reduced or were reducing hiring, and 33 per cent had made or were making workforce reductions. These figures describe responses to "actual or anticipated" AI use, which matters: much of the change is still being planned, so the pattern of cutting first and correcting later could repeat.
The cost nobody budgets for
A cut followed by a rehire is not a neutral round trip. Several costs sit outside the payroll line:
Institutional knowledge.
Thirty-six per cent of managers in the Robert Half survey said the restored roles required context AI could not replace. Once a person who held that context has left, it does not come back with a new hire.
Burnout in the team that stayed.
One-third of respondents reported strain among the remaining staff. Colleagues who absorbed extra work after a cut remember it when the replacement arrives.
Hiring in a worse position.
A company that restores a role months after cutting it is recruiting from scratch, often in a market where the people it let go have already moved on.
Employer reputation.
Candidates read postings closely now, and a role that was eliminated and then reposted raises questions about how the company plans.
None of these appear in the original savings estimate, and all of them can be reduced by sequencing the change differently.
What a better sequence looks like
Employers that are expanding AI use, or reviewing earlier cuts, can draw several practical lessons from the data.
Pilot before you cut.
Run the AI-assisted process alongside the existing team long enough to measure error rates, review time and exceptions. Reduce roles once the numbers come in, not before.
Count the checking.
Put verification time into the business case explicitly. If one-third of working time goes to review, the model should say so.
Protect context-heavy roles.
Positions that rely on client relationships, regulatory judgment or knowledge of how the organization really works are the ones managers most often restored. They deserve a higher bar before they are removed.
Redeploy before releasing.
Where a task can be automated, the person doing it often has the context needed to oversee the tool. Moving them into a quality or oversight role keeps that knowledge in the building.
Keep capacity flexible.
Demand rose for 38 per cent of the managers who rehired. Uncertainty about volume and about how well a tool will perform is a case for contract and project-based talent that can be added when the work appears and scaled back when it does not, without the commitment of a permanent role in either direction. That is a practical middle path between committing to headcount and cutting it, and it is where staffing partners such as SRA Group are often brought in.
The bigger picture
The first phase of workplace AI was driven by what the technology might do. The second phase is being driven by what it has done in daily use, and the early evidence suggests the gap between the two is wide enough to cost real money. About a third of the Canadian managers who made cuts have already corrected course, which is a quick reversal by the standards of workforce planning.