The AI Race to the Bottom Nobody Is Talking About

The most dangerous AI strategy isn’t the one that fails. It’s the one that succeeds. Just long enough to cause irreversible damage.

There is a quiet consensus forming in boardrooms right now. AI is arriving, costs must fall, and the most obvious cost to cut is people. If you can automate a function, the logic goes, why keep paying a human to do it?

It sounds like clear thinking. It isn’t.

What it actually is, is one of the oldest traps in business strategy: optimising for what you can measure today, at the expense of something harder to quantify that will hurt you tomorrow. And I think it is worth naming this clearly — not to score moral points, but because the consequences are entirely avoidable if leaders are willing to think one step further.

The replacement narrative and why it is so seductive

The case studies are landing thick and fast. Klarna replaced the equivalent of 700 customer service agents with AI between 2022 and 2024 and declared it a triumph. IBM announced plans to replace roughly 7,800 back-office roles with automation. BT Group is targeting 55,000 job reductions by 2030, with AI doing a significant share of the work. The numbers are clean. The savings are real. The investor response is predictable.

What the press releases don’t cover is what happens next.

Klarna, to their credit, provided us with an unusually honest follow-up. Between 2022 and 2024, the company eliminated around 700 customer service positions, replacing them with an AI assistant built in partnership with OpenAI. At its peak, the system was handling roughly three quarters of all customer interactions. The CEO declared it a triumph. Investors applauded.

Within a year, the story had changed. Customer satisfaction had deteriorated on complex interactions. The cost savings projected in the original announcement had not fully materialised. By early 2025, Klarna began rehiring human agents — and Sebastian Siemiatkowski made a public admission that is rare in the world of tech leadership: “We went too far. We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable.”

The AI could handle volume. It struggled with nuance, with escalation, with the kind of conversation where a frustrated person needs to feel heard before they need to be helped.

The humans, it turned out, were not just cost centres. They were carrying something the model couldn’t replicate.

This is not an argument against AI. It is an argument against the framing.

The question beneath the question

Before any organisation commits to a large-scale replacement strategy, I think it is worth asking a more fundamental question: what is this organisation actually for?

If the honest answer is purely shareholder return, then the logic of replacing humans with cheaper AI agents is internally consistent. Uncomfortable, but consistent.

But most organisations claim something broader. They talk about creating value for customers, for communities, for employees. Many have sustainability commitments, diversity goals, and purpose statements on their walls.

And here is the tension that too few leaders are willing to sit with: a strategy that systematically depletes human employment at scale is in direct conflict with the idea that organisations have a role in sustaining human dignity.

I believe they do have that role. I also know that being on the moral high ground while your competitors undercut you is not a viable strategy. There is no virtue in going out of business. So this is not a call to ignore competitive reality — it is a call to think more rigorously about what the real trade-offs are, and over what time horizon.

The cost nobody is modelling

One consequence rarely appears in the AI investment case: adaptive capacity.

When organisations hollow out their human workforce in pursuit of short-term efficiency, they are making a bet that the current operating model is sufficiently stable to justify the trade. I was reminded of how deep this framing runs when a client recently asked us to help their people learn to work alongside their “AI colleagues.” A revealing choice of words — because a colleague is a peer, someone who shares the work. What they were describing was something closer to a replacement arriving in stages, and the humans were being asked to manage the handover of their own roles. They meant well. But the language told the whole story.

And they are far from alone. A recent IBM survey of 2,000 CEOs found that just one in four AI projects delivers on the return on investment it promised. Yet companies press on regardless — because nearly two thirds of CEOs admit that the fear of falling behind drives them to invest in technology before they fully understand the value it will bring. That is not strategy. That is herd behaviour with a budget attached.

But technology does not stand still. The organisations that navigate disruption well do so because they have people who can think, experiment, reframe problems and carry new ideas across the business. That capability does not live in a process map. It lives in people.

When the next wave of technology arrives — and it will — who pivots? You cannot prompt an AI agent to reimagine your business model. You cannot automate your way to a new strategic direction. That work requires human judgment, human relationships, and human courage. If you have spent the previous five years systematically reducing your human capability in the name of efficiency, you will arrive at that moment poorly equipped.

This is the race to the bottom nobody is advertising. Not a race to the lowest wage.

A race to the lowest organisational intelligence.

The alternative is the steeper path — and fewer are taking it

The organisations getting this right are not doing so because it is easier. They have simply chosen not to pretend that a complex problem has a simple solution.

Microsoft’s internal rollout of Copilot across its workforce is perhaps the most cited example of genuine augmentation thinking. The intent was not to reduce headcount but to increase what each person could do, tracking productivity gains at the individual and team level.

Toyota, whose philosophy of respect for people has guided its approach to automation for decades, has consistently treated technology as something that assists human judgment on the production line, not something that eliminates the need for it.

It is also worth noting that the consulting firms advising organisations on AI strategy are not immune to this tension themselves. Their published positions consistently favour augmentation over replacement. Their own workforce decisions tell a more complicated story — with significant reductions in junior hiring across the major firms in the same period they were rolling out internal AI tools. The people selling the map are navigating the same terrain.

The augmentation model says: here is a tool that can handle what slows you down, so that you can concentrate on what really makes a difference. It asks people to learn, to experiment, to adapt. It accepts that the transition is uncomfortable and that not every role survives unchanged. But it positions people as partners in the improvement rather than casualties of it.

There is one more thing the augmentation model demands, and it may be the hardest: share the value that is created. If AI doubles the output of your workforce, the question of whether that gain flows entirely to shareholders or is shared in reward, in time, or in development is not just an ethical question.

It shapes whether people stay, whether they care, and whether they will bring their best thinking to work tomorrow or quietly update their CVs.

So here is the question I want to leave you with

If you are an HR or L&D leader reading this, you probably already see the effects. You are managing anxiety, outplacement, the slow erosion of team morale. You may feel that the strategic decision has been made above your head and your job is simply to manage the fallout.

But I think you have more standing to challenge the framing than you realise. Because the data on what organisations lose when they deplete their human capability is not soft data — it is strategic risk.

The question worth putting to your leadership team is this:

We are modelling what AI will save us. Are we also modelling what we will lose — and whether we can get it back when we need it?

I would genuinely like to know what you are seeing in your organisations. Are you navigating this well? Is the replacement conversation being tempered by anything? Or does the efficiency logic feel unstoppable right now?