The AI Layoff Reversal: Companies Are Rehiring the Engineers They Cut. Here's How to Rebuild Without Repeating the Mistake.

The Reversal Is No Longer Anecdotal

For most of 2025 and early 2026, the story about AI and jobs ran in one direction. Executives announced that agents and copilots would let them do more with fewer people, and the layoffs followed. The correction has now arrived, and it is showing up in survey data rather than in isolated news stories.

In February 2026, outplacement firm Careerminds surveyed 600 HR professionals at organisations that had made AI-led layoffs in the previous twelve months. The results are stark: 35.6% had already rehired more than half of the roles they cut, and another 32.7% had rehired between 25% and 50%. Of those rehires, 52.1% happened within six months of the original reduction. Only 8.4% of respondents said the restructure delivered what was promised, and 90% said they would approach AI layoffs differently if they could do it again.

This was not a surprise to everyone. Back in April 2025, Orgvue reported that 39% of business leaders had made employees redundant as a result of deploying AI, and that 55% of those admitted they had made wrong decisions about the redundancies. What has changed since then is that regret has turned into job requisitions.

The early high-profile case was customer service. In May 2025, Klarna's chief executive Sebastian Siemiatkowski acknowledged that an AI-first approach to support had produced "lower quality" service and that the company would invest in human support again. In 2026 the same pattern reached engineering, manufacturing and IT, the functions where companies had assumed the case for automation was strongest.

Where the AI Layoff Wave Actually Stands

The reversal does not mean the cuts have stopped. According to the Challenger, Gray & Christmas August 2026 report, US employers have cited artificial intelligence in 116,175 job cut announcements so far in 2026, approximately 22% of all cuts, and AI remains the leading reason year to date. The technology sector alone has announced 155,126 cuts through August, an increase of 52% on the same period of 2025 and 29% of all announced cuts.

But the same report shows the momentum shifting. In August, AI fell to the fourth most cited reason, with 3,462 cuts, its lowest monthly total since December 2025, ending a five-month run in which it led every month. Technology also leads all industries in announced hiring plans for the year, with 19,751. The sector is cutting and hiring at the same time, which is exactly what a correction looks like from the inside.

For engineering leaders, the headline numbers matter less than the composition. The roles cut first were often the ones that looked most automatable on a spreadsheet: junior developers, manual QA, support engineering, internal tools. In the Careerminds data, entry-level roles were the most affected group at 31.5%. Those are also the roles that, in a working software organisation, do much of the verification, documentation and operational follow-through that AI-generated output now depends on.

Ford's Lesson: The Knowledge Left Before the Model Learned It

The clearest engineering case of 2026 came from Detroit. Ford installed 900 AI-assisted cameras to catch quality defects on its production lines, expecting automation to raise quality while reducing reliance on experienced staff. It did not work as planned. As CBT News reported, citing Bloomberg, the company has rehired 350 veteran engineers, who now run mandatory quality meetings and reprogram the AI tools to flag problems earlier.

Charles Poon, Ford's vice president of vehicle hardware engineering, put the core problem simply: "The technology is only as good as the data used to train it." Ford had let experienced engineers go before their judgement had been captured in the systems meant to replace them. The automation had nothing to learn from.

The result of reversing course was measurable. Ford rose from tenth place to become the top mainstream brand in the 2026 J.D. Power Initial Quality Survey, behind only the luxury brands Porsche and Genesis. The veterans did not replace the AI. They made it useful.

Software teams face the same dynamic with less visible consequences. A coding agent working in a large codebase depends on the same kind of tacit knowledge Ford's cameras lacked: why a module is shaped the way it is, which integrations are fragile, which customer depends on an undocumented behaviour. When the engineers who held that knowledge leave, the agent keeps producing code at the same speed. It just stops being right as often, and there is nobody left who can tell.

Why Engineering Teams Feel the Reversal Hardest

The most revealing number in the Careerminds survey is not about rehiring. It is about why the rehiring happened: 54.6% of organisations said AI required more human oversight than they anticipated. In the same survey, 32.9% said they had lost critical skills and expertise, and 28.1% said they lacked the internal capability to fill the resulting knowledge gaps.

In software, oversight is not a side task. It is review, testing, incident response, security triage and architectural judgement, and the volume of all of it rises when AI increases the volume of code. A team that halves its headcount while doubling its output has not become twice as efficient. It has moved the bottleneck from writing code to verifying it, and removed the people who did the verifying.

Some large employers have already drawn the opposite conclusion from their peers. In February 2026, IBM's chief HR officer Nickle LaMoreaux said the company would triple entry-level hiring in the US, including for "software developers and all these jobs we're being told AI can do". IBM's junior developers now spend less time on routine coding and more time working directly with customers. Her reasoning was blunt: the companies most successful three to five years from now will be those that doubled down on entry-level hiring.

The underlying point is that an engineering organisation is a system for producing judgement, not just code. Cut the people who generate and pass on that judgement, and the AI tools you bought to replace them lose the context that made them productive.

The Rehire Often Costs More Than the Layoff Saved

Layoffs are justified on a simple calculation: salaries removed minus severance. Rehiring breaks that calculation. Careerminds found that 30.9% of organisations that brought employees back spent more doing so than they had saved from the original cut, and a further 42.37% only broke even. For a large majority of companies that reversed course, the layoff produced no net saving at all.

The reason is that replacing people has always been expensive. Gallup estimates that the cost of replacing an individual employee ranges from one-half to two times their annual salary, and describes that as a conservative estimate. For software engineers, the visible costs of recruiting fees, interview hours and signing premiums are only part of it. The larger cost is the months a new engineer spends learning a codebase before they can be trusted to change it safely.

There is also a cost that never appears in the HR budget: the work that did not happen. Every month a team runs below the capacity it needs, features slip, incidents take longer to resolve, and technical debt accumulates in the AI-generated code nobody had time to review properly. By the time the rehiring decision is made, the backlog of deferred work is often larger than the team being rebuilt.

None of this means AI investment was wasted. It means that the business case for many AI-led layoffs counted the savings and left out the cost of the oversight, context and capacity the cuts removed.

The Quiet Rehire, and Why It Can Become a Second Mistake

Few companies will announce that their AI layoffs were a mistake. Analysts expect the reversal to happen quietly. Forrester's Predictions 2026 research forecast that half of AI-attributed layoffs would be reversed, with much of the work given to lower-wage workers, "offshore or at lower salary". Gartner made a parallel prediction in February 2026: by 2027, 50% of companies that attributed headcount reduction to AI will rehire staff to perform similar functions, but under different job titles.

That is where the second mistake happens. A company under pressure to restore capacity fast, without admitting the first decision was wrong, tends to reach for the cheapest option available. Often that is a distant offshore vendor with a low hourly rate, little overlap with the working day and high turnover. The capacity shows up on paper. The judgement, context and accountability that were lost in the layoff do not come back.

The irony is sharp. The original cuts failed because they removed people who held context and provided oversight. Rebuilding with a team that is many time zones away, rotates engineers frequently and is managed through tickets rather than conversations reproduces the same gap in a different form. The organisation ends up paying for headcount it still cannot rely on, and the AI tools remain under-supervised.

Rebuilding is a strategic decision in its own right. It deserves the same scrutiny the layoff should have received: what capability is missing, who needs to hold the knowledge, and what working model will keep that knowledge in place.

How to Rebuild Engineering Capacity on Purpose

Start with a capability audit, not a headcount target. List the work that has slipped since the cuts: code review latency, incident response times, test coverage on AI-generated code, deferred upgrades, onboarding of new services. That list tells you which skills you need back. In most teams it points to senior engineers who can review and architect, and to QA and platform skills that keep AI output safe to ship.

Next, decide where the knowledge should live. The Ford lesson is that expertise has to be captured before it can be automated. Whatever capacity you rebuild, make knowledge transfer an explicit deliverable: architecture decision records, runbooks, repository-level agent instructions, and test suites that encode business rules. That way the next restructure, whatever its cause, does not wipe out the context again.

Then choose a capacity model that can flex without whiplash. Pure in-house hiring is slow and makes every future adjustment another layoff. Freelancers are fast but rarely accumulate context. Low-cost offshore vendors save on rates but often lose the overlap and continuity that oversight depends on. A dedicated nearshore team sits between these options: stable named engineers who stay on your product, work in your repositories and tools, and share most of your working day, with the ability to scale up or down through a contract rather than a redundancy process.

Where a Dedicated Nearshore Team Fits

This is the situation we see most often at Stepto in 2026. A client has invested heavily in AI tooling, reduced its engineering team, and discovered that review queues, incidents and quality problems have grown faster than output. The client does not need a vendor to take over the product. It needs experienced engineers who can work inside its existing team, supervise AI-assisted work and rebuild the context that left with the people it lost.

Our dedicated development teams are built for exactly that. Engineers from our Belgrade team join your repositories, your stand-ups and your review process, and stay on your product rather than rotating across accounts. Serbia shares most of the working day with Western Europe and gives several hours of overlap with the US East Coast, so code review and incident response happen in conversation rather than in an overnight ticket queue. When your needs change, the team scales by agreement, not by another round of layoffs.

If the gap is specific, we can fill it specifically: AI developers who know how to build evaluation and oversight into AI features, QA automation engineers who restore verification capacity, and senior engineers through staff augmentation who can take ownership of review and architecture. Each engagement includes deliberate knowledge transfer, so the context we build stays with your organisation.