Cutting Staff for AI, Then Rehiring at a Premium: The Cost of Getting the Order Wrong.

IN ONE MINUTE

In the first half of 2026, US employers attributed 101,743 announced job cuts to AI, close to double the total for all of 2025. Many are now reversing course. In a Careerminds survey of 600 HR leaders, 52% of companies that made AI-driven cuts had started rehiring within six months, and only about a quarter came out financially ahead. The problem is rarely the technology. It is the ambition and the order: cutting to hit a cost target before the new way of working is proven. Start with purpose, set a budget ceiling with a named owner, plan for the people, then iterate.

THE OBJECTIVE

To help leaders plan workforce changes around AI in an order that protects both people and the business case.

WHAT THE RESEARCH SHOWS

• Challenger, Gray & Christmas (Jul 2026): 101,743 announced US job cuts in the first half of 2026 were attributed to AI, about 23% of all cuts, compared with roughly 55,000 for the whole of 2025.
• Careerminds (Feb 2026, 600 HR leaders who oversaw layoffs): 52% of companies rehired within six months, and more than a third rehired over half the roles they had cut. 31% spent more on rehiring than they saved, 42% broke even, and only 27% came out ahead.
• Orgvue (2025, 1,163 senior leaders): 39% had made employees redundant because of AI. Of those, 55% said they had made wrong decisions.
• Gartner (Feb 2026): predicts that half of the companies that cut customer service staff because of AI will rehire by 2027.

WHAT HAPPENED NEXT

The reversals are now visible in some well-known names:

• Ford has been rehiring hundreds of experienced engineers and quality inspectors after AI-led quality checks fell short (Bloomberg, Jun 2026). As Ford's Charles Poon put it: "Artificial intelligence is a fantastic tool, but it's only as good as the information you use to train it."
• Commonwealth Bank of Australia reversed a set of role cuts in 2025 after the work proved harder to automate than expected.
• McDonald's ended its AI drive-through ordering test with IBM in 2024, after videos of ordering errors went viral.

The pattern is consistent. AI can do parts of a job well. But jobs are bundles of tasks, relationships and judgement. When the whole role is removed before the new workflow is proven, the remaining work still needs someone, and the knowledge that left the building is expensive to bring back. The people who return are often doing a bigger job than before: not just the task, but managing, checking and directing the AI that was meant to replace them.

THE MATH NOBODY BUDGETED FOR

There is a second, quieter version of this failure: AI spend with no ceiling and no owner.

• One company was widely reported to have run up around $500 million in AI usage in a single month after no usage limits were set.
• Uber reportedly used up its entire 2026 budget for AI coding tools within four months (The Information, via Fortune).
• Nvidia's Bryan Catanzaro, vice president of applied deep learning, said: "For my team, the cost of compute is far beyond the costs of the employees."

I have lived a version of this myself. Earlier this year, one of my own initiatives ran 30% over budget in the first half. The strategy wasn't wrong. A software bug, a governance gap and a token spend cap that was never clearly set combined, and because nobody had been assigned to watch the meter, the overrun was booked before we caught it. It was an expensive lesson, and an avoidable one. The technology didn't fail. The governance around it did.

It is the same failure as Issue 001 in a different form. There, most AI investment went to technology rather than people. Here, spend becomes a substitute for strategy.

THE REAL FAILURE WASN'T EXECUTION. IT WAS AMBITION.

It is tempting to conclude the lesson is purely operational: companies moved too fast, cut too deep and didn't govern the spend. That is true, but it is the smaller truth.

The larger truth is that many of these programmes were never trying to build anything. They were trying to hit a number: a cost-reduction percentage, a headcount target, a margin improvement for the next results announcement.

That ambition sets AI up to disappoint. When the objective is "reduce cost by X%", AI is judged only as a substitute for a salary line. By that measure it will often look expensive, because compute has a real and volatile cost, and the human judgement needed for edge cases doesn't disappear when you stop paying for it. It moves to your customers, your quality scores and, eventually, your rehiring budget.

The organisations I have seen benefit from AI asked a different question. Not "how many people can this replace?" but "how much more value can this create through the people we already have?" For example: a support team resolving more complex cases because AI handles the routine ones; a product team running far more experiments because AI compresses the research cycle; an analyst producing what used to take a team, and being rewarded and promoted for it.

Same technology. Different ambition. A very different outcome.

If your AI strategy is designed to hit a bottom-line percentage, you will be measured against that percentage, and compute costs, edge cases and rehiring premiums will eat into it. If your AI strategy is designed to multiply what your people can produce, the ceiling isn't a cost target. It is revenue you haven't imagined yet.

WHERE THE COST HIDES

A reduction taken too early can carry costs that don't appear in the original business case:

• Severance and exit costs, followed by recruitment fees and higher salaries to rehire similar skills.
• Lost institutional knowledge: customer history, supplier relationships, workarounds nobody documented.
• Service dips while AI tools are still being tuned.
• Lower trust among the people who stay, which slows adoption of the very tools meant to deliver the savings.

THE FRAMEWORK: PURPOSE → PLAN → ITERATE

Every failed AI programme I have studied, including my own overrun, skipped straight to deployment. The successful ones followed three steps, in this order.

Step	The discipline	The test before moving on
1. Purpose	Define the new value this creates, for the customer or the business. "Reduce cost by 15%" is a target, not a purpose. "Resolve complex cases in half the time, so we can handle double the volume" is a purpose.	Can we describe what will exist that didn't before?
2. Plan	Put two things in place side by side: a budget ceiling with a named owner, and an upskilling track for the people who will manage, audit and direct the AI.	Is there a spending limit, a person watching it, and a trained team ready?
3. Iterate	Treat the rollout like a product, not a launch: small test, real data, adjust, expand. Not: announce, cut, deploy, hope.	Have we found the edge cases in a contained pilot, not in front of customers?

A budget without an upskilling plan buys a tool nobody is trained to run. An upskilling plan without a budget ceiling and an owner writes a blank cheque.

If you remember nothing else: know why before you buy, know the ceiling before you spend, and know the failure before you scale.

WHAT THIS MEANS FOR CEOS AND BOARDS

Ask for the purpose, not just the percentage. Before approving an AI-driven restructuring, ask what new capability it creates. If the only answer is a cost figure, the programme is not ready.

Make headcount the last decision, not the first. Reductions based on forecast savings are a bet. Reductions based on a workflow that is already running reliably are a decision. The sequence protects both the business case and the organisation's reputation as an employer.

Watch for reversal risk. With more than half of companies in the Careerminds survey rehiring within six months, a board should ask how the business will know early if the new model isn't working, and what it would cost to reverse.

WHAT THIS MEANS FOR CFOS AND BUDGET OWNERS

Give every AI programme a spending ceiling and a named owner. Usage-based pricing means costs can scale faster than value. Set limits and alerts before the first rollout, and make one person accountable for watching them.

Model the full cost of a reduction. Include severance, lost knowledge, service risk, the cost of rehiring, and ongoing compute. Compare it with growth scenarios, not only with today's salary line.

Measure savings, don't forecast them. Release budget in stages as savings are proven in real operations, not when they appear in a business case.

WHAT THIS MEANS FOR HR AND PEOPLE LEADERS

Build the upskilling track before deployment. The people who manage, audit and direct AI are the capability companies are now rehiring for. Growing it internally is usually faster and cheaper than buying it back.

Redeploy before you release. Use attrition, redeployment into growth areas and hiring restraint first. Where reductions are needed, base them on the proven workload and handle them with care.

Protect trust in the people who stay. Adoption depends on the people using the tools. Involve them in redesigning the work, and be clear about how roles will change.

THE LAST WORD

Purpose first. A budget with a ceiling and an owner. A plan for the people. Then iterate.

If your organisation has made AI-driven cuts this year, has the purpose behind them been written down anywhere, or does it live only as a percentage on a slide?

—

Until next week,

Artificial Ian

Tell your network: if someone you know is building a workforce plan for next year, this sequence might be worth a conversation.

→ Found this useful? Forward it to a CEO, CFO or HR leader planning for AI.

→ Subscribe to The AI Executive Briefing, weekly in your LinkedIn inbox.

Sources: Challenger, Gray & Christmas, mid-year job cuts report (Jul 2026); Careerminds, AI layoffs survey of 600 HR leaders (Feb 2026); Orgvue AI workforce research (2025); Gartner customer service prediction (Feb 2026); Bloomberg on Ford (Jun 2026); Fortune on AI costs and Nvidia's Bryan Catanzaro (May–Jun 2026); TechCrunch, "The token bill comes due" (Jun 2026).