Last week I spent a day at Go Behaive, an event built entirely around the behavioural side of AI. The room was a heavy one: Rory Sutherland on behavioural science, Graham Fink on creative, Thomas Hirschmann on AI, with Victoria Russell steering the day. The conversations were some of the sharpest I've had in a while, a lot of serious minds circling the same problem from different directions.
By the afternoon, one pattern had surfaced repeatedly, and it was less about the technology. Some companies are quietly removing the first step of the career ladder and calling it an AI strategy. Others are openly announcing plans to replace thousands of employees with AI. Standard Chartered CEO Bill Winters recently spoke about replacing 7,000 roles of what he described as "lower-value human capital".
The logic is easy to follow. Graduate and entry-level work, the basic analysis, the first drafts, the routine modelling, is exactly what AI handles most comfortably. So the boardroom question has become blunt. Why spend years training a young mind when you can hand a director a supporting agent that does all those ‘lower-value human tasks’ today?
Well, it is already happening. A participant from one of the major consultancies told me graduate hiring had been cut heavily to help fund AI implementation. The data suggests they are far from alone.
Across the UK and US, entry-level hiring is falling while senior hiring remains largely intact. Companies are not reducing the size of the ladder. They are removing its first step.
The premise of the day made the contradiction sharper. The whole argument was that AI is only as good as the humans using it. If AI is the tool, behavioural science is the operating manual for the people holding it. So if the human side decides whether AI works, why is the human side the first thing being cut?
This is where I think the logic breaks down.
AI itself is unlikely to be a lasting competitive advantage. Over time, it will become a commodity, much like the internet. Being online stopped being an edge years ago. It became a given. The same shift is happening with AI, faster than many organisations expect.
The models are converging. Prices continue to fall. Capabilities that once felt exclusive are increasingly available through the same APIs and platforms. When every competitor can access similar tools, the tools themselves stop being the differentiator. They become infrastructure.
That makes the current enterprise response feel strange when you say it plainly. Organisations are racing to own something that is rapidly becoming accessible to everyone. To fund that race, many are cutting the thing that may become genuinely scarce: developed human judgement.
The real advantage will not come from having AI. Everyone will have AI. It will come from how effectively people apply it, challenge it, direct it, and combine it with experience, context, taste, and commercial instinct.
One idea from the day stayed with me because it reframed the problem entirely. Instead of thinking about artificial intelligence as a replacement technology, think about it as intelligent amplification. The goal is not to remove people from the system. The goal is to increase the capability of the people already in it.
AI that replaces people pushes organisations towards the same destination as everyone else using the same tools. AI that amplifies people compounds the one thing competitors cannot easily copy.
The entry level was never just cheap labour. It is where judgement is formed. It is where people learn how organisations work, how decisions get made, what good looks like, and what happens when they get it wrong. Those lessons cannot be downloaded. They are accumulated.
People learn taste, commercial instinct, and decision-making by doing the basic work, making mistakes, and being corrected by people who already know. Automate that away and you do not just save a salary this year. You stop producing the seniors you will need in ten. You can measure the cost of a graduate. It is much harder to measure the cost of never having developed one. We talk about this in our article The New Role of the Junior Designer.
None of this is anti-AI. Some junior work should be automated. Nobody develops judgement from reformatting the same spreadsheet four times. The point is to be deliberate about the difference between drag and development. Use AI to remove the drag from junior work, not the juniors themselves. The question that matters is not what can we automate. It is what capability are we eroding by automating it.
If you lead in an AI-exposed industry, this is the trade you are making. Strip out the junior layer and you improve this year's margin. You also weaken your ability to produce the judgement, expertise, and leadership you'll need a decade from now.
Scarcity creates value. In a market where everyone has access to the same tools, the organisations with the deepest bench of trained judgement will use those tools best. That is the moat. Not the AI.
If that diagnosis is correct, then the logical response is not to stop developing people. It is to develop them faster and more deliberately.
The bet is simple. When the tools are a given, the people who can think will be the difference. You cannot generate experience. You grow it.
Sources
- UK graduate hiring (46% fall in 2024, further 53% projected): Institute of Student Employers data, via The Register, October 2025.
- Junior versus senior hiring under AI adoption: Hosseini Maasoum and Lichtinger, Generative AI as Seniority-Biased Technological Change, Harvard University, August 2025.
- AI model convergence and falling inference costs: Stanford HAI, 2025 AI Index Report.