The 30% Rehire Problem

Whenever someone tells me a workforce decision “isn’t about the money,” I have a stock reply.

As always, “it’s never about the money… until it’s about the money.”

Gartner has just given that line a number.

By 2029, Gartner predicts, 30% of employees laid off due to replacement by AI will need to be rehired – often at a significantly higher cost.

Read that again.

Not “some roles may need revisiting.” Nearly a third of AI-driven redundancies, undone, at a premium.

30%. Nearly a third of AI-driven redundancies, undone, at a premium. Gartner predicts this by 2029 for employees laid off due to replacement by AI.

A few weeks ago I askedwhether we’re about to make exactly the same mistakes with AI that we made with the last generation of big technology.

Gartner’s own research suggests we’re already partway through answering that question – and not in our favour.

Gartner sets out three ways organisations are actually using AI, each carrying a different risk.

Use it to help people do their existing jobs – a smarter assistant sitting alongside them, not replacing them (“augment work,” in Gartner’s terms) – and the risk is morale, trust and quality: people either don’t use it properly, or lean on it too heavily.

Use it to genuinely redesign how work gets done – new workflows built around it, not just a tool bolted onto the old process (“reengineer work”) – and the risk is that skills quietly fade, fewer junior roles get created, and the usual route into more senior positions breaks down.

Use it to do something that didn’t exist before at all – new roles, new services, a genuinely new way of operating (“invent new work”) – and the risk Gartner names is excessive layoffs, reputational damage, and a culture that remembers.

Whatever the mechanism, the 30% rehire figure is the part that’s solidly sourced – and it sits next to another figure from Gartner’s research: 76% of CEOs expect AI to have the biggest impact on their industry within three years. Nobody’s arguing AI matters less than they think. The gap is between believing AI will change everything and actually preparing the workforce for how.

We’ve seen that gap before. ERP, shared services, self-service, robotic process automation – decades of technology implemented well, work barely redesigned.

Self-service was meant to remove work; it just moved who did it. Automation was meant to release capacity; the capacity often stayed exactly where it was, quietly absorbed into an already full day, because nobody had decided in advance where it should go.

Here’s what makes that history worth sitting with. Large organisations know exactly how expensive ERP and HRIS systems are. Get the contract wrong, and the true cost – licensing, implementation, integration, ongoing support – can run into the tens of millions, sometimes considerably more, depending on the product and the organisation. What’s rarer is anyone properly acknowledging the sunk cost it takes to get anywhere near the ROI in the original business case, let alone measuring whether they ever arrived.

Which is why I’m genuinely surprised how many people seem to assume AI is either free, or close enough not to matter. It isn’t. And on the evidence so far, it won’t be for some time yet.

Nvidia’s own VP of applied deep learning said this year: “For my team, the cost of compute is far beyond the costs of the employees.” Uber burned through its entire 2026 AI tools budget in four months. Big Tech capital expenditure hit $740bn for 2026, up 69% on 2025, with no clear productivity payoff yet visible in the numbers.

Even the heaviest AI-spending companies are paying around $7,500 per employee a month on AI, against roughly $16,000 a month for an average software engineer – the median company spends just $11.38. Unit prices are expected to fall sharply, but Gartner’s own analysts add the catch: the higher-value uses organisations actually want need five to thirty times more tokens per query, and Goldman Sachs forecasts a 24-fold rise in usage by 2030. Cheaper tokens and a cheaper total bill are not the same thing.

On Sunday morning Claude told me I’d used 99% of my usage for the week. That was probably a welcome stop, a chance to focus on other things, rest and recuperate, but it points out that while AI can work 24/7, which everyone is excited about, particularly CFOs, it does cost, and it costs for every transaction we use it for – in the AI business cases, I’d love to see where this cost is articulated and how it’s actually calculated.

Every one of those programmes – ERP, self-service, and now AI – was pitched as being about capability. Agility. Getting ahead of the curve. And every one of them eventually landed on someone’s balance sheet as a number.

I wrote about this from the other side inBefore You Delete the Role, Do You Understand What You Are Deleting?

The observation there was that job descriptions rarely capture the informal work that keeps an organisation functioning – resolving exceptions, joining up disconnected processes, checking unreliable data, maintaining relationships, remembering why a control exists. Most of it stays invisible until the person doing it is gone.

That’s the same blind spot showing up again here, just earlier in the process.

If you don’t understand the detail of the activity you’re removing, how do you measure it?

You can measure a salary. You can’t measure work you never actually mapped.

Which is presumably at least part of why 30% of these decisions end up reversed – the business case only ever priced what the role cost, not what it did.

The AI version of that pattern isn’t subtle. Cut the headcount before you’ve established what the work actually was. Before you know what AI genuinely replaces, and what still needs a person doing it, just differently. Get that sequence wrong, and you’re rehiring – at a worse rate, having shown your own people exactly how much thought went into the first decision.

Gartner’s own explanation for the rehire premium, in fact, has nothing to do with AI’s capability and everything to do with the basics: cutting a team depletes talent pipelines and erodes institutional knowledge, and in a labour market where growth is flat or declining, competition for the people you need pushes up recruitment, training and onboarding costs. In other words – the cost was never really about the technology.

Gartner’s other piece, published the same week, points at the alternative. Four shifts, in their language: expand human capability through human-AI collaboration rather than replacement; build an AI-ready workforce through continuous learning rather than a one-off training push; keep human judgement and context embedded in AI-driven processes rather than letting it quietly disappear; and – the one I keep coming back to – by 2027, organisations reinvesting AI’s productivity gains into new value are predicted to outpace the ones simply taking cost out.

None of this is an argument against using AI to change headcount. Sometimes that’s exactly the right call, made well, for the right reasons.

It’s an argument against skipping the harder work to get there faster.

Understanding what a role really does, not just what it costs. Understanding what AI genuinely costs, not just what the vendor demo promised. Deciding deliberately what stops, what gets automated, and what a human still needs to own under a different name.

Call it capability. Call it agility. Call it the future of work.

Just don’t be surprised when, a year or two later, it turns out it really was about the money all along – you just hadn’t done the sum yet. And you can’t do the sum on work you never understood in the first place.

So here’s the question I can’t put down: if nearly a third of AI-driven layoffs are heading for a rehire, at a premium, whose business case is that showing up in? And is anyone currently measuring it?

Related insight

Are We About to Repeat the Same Mistakes With AI?

Before You Delete the Role, Do You Understand What You Are Deleting?

Sources

• Gartner, “Gartner Identifies Four Shifts Shaping the Future of Work”, press release, 9 September 2026: the 30% rehire prediction, the reasons for the rehire premium, the four shifts and the 2027 reinvestment prediction.
• Gartner, “Lay the Talent Groundwork for AI Success” (AI talent risks): augment, reengineer and invent new work, and the 76% of CEOs figure.
• Fortune, 28 April 2026: Bryan Catanzaro, Nvidia VP of applied deep learning, on the cost of compute versus employees.
• Axios, 26 April 2026: Uber’s 2026 AI tools budget (first reported by The Information) and Big Tech’s $740bn 2026 capital expenditure.
• Ramp AI Index, June 2026, summarised by IndexBox: AI spend per employee versus engineer salaries.
• CIO Dive, 25 March 2026: Gartner analyst Will Sommer on falling inference costs and 5–30x more tokens per query.
• Fortune, 22 May 2026: Goldman Sachs forecast of a 24-fold rise in token use by 2030.