AI Is Doing Your Juniors' Work. Where Will Your Future Seniors Come From?

Nearly two decades ago, in the middle of the Global Financial Crisis, I spent my first day in HR filing a cabinet full of leave application forms.

It was not the job I wanted. But it was the only job in town, in a market full of more qualified, more senior people who had suddenly found themselves out of work. So I filed.

Today's new joiners are mostly spared that kind of task. Hand someone a stack of paper now and a scanner, a text-recognition app and an AI assistant will clear it in an afternoon. I am genuinely glad about that.

Here is what I did not appreciate at the time. That filing cabinet taught me how the business actually worked. Who took leave when, which managers approved everything and which approved nothing, where the policy said one thing and practice did another. The boring work was the training. AI is now removing both the chore and the lesson, and most businesses have not noticed that the second one is gone.

The entry-level squeeze is already measurable

In October 2026, Anthropic, the company behind the Claude AI models, published a report on how AI has changed employment inside its own business. It is one of the most AI-heavy workplaces in the world, which makes it a useful early signal. Two numbers stood out to me:

• The share of junior employees in its technical departments fell from 21% to 14% since early 2025.
• Among new technical hires, the junior share fell faster, from 27% to 9%.

The report also found that hiring managers now look less for specific technical expertise and more for "autonomy and judgment". And it is candid about the cost: employees are "continually automating their own tasks away", and in some roles they are "increasingly removed from the object-level work that originally built their skills."

A fast-growing AI lab is not a 30-person Singapore company, and the authors themselves warn that much of their situation cannot be copied. But the direction is the point. If the leading edge has stopped hiring juniors, and is asking for judgement instead, then the question every SME owner should be asking is: where does judgement come from, if nobody does the junior work any more?

The honest answer is that there were never many quality entry-level roles that properly engaged and trained the next generation of leaders. AI is making a thin pipeline thinner.

Handing out AI subscriptions is not a strategy

The easy response is to buy AI subscriptions for the team and hope for the best. In my experience, that is where most AI adoption quietly fails.

Failing at AI is mostly a skills problem, not a tools problem. I have given the same task to different people in two ways: once with clear, pointed guidance on how to use AI for it, and once with the standard "learn together with AI" encouragement. The results could not have been more different. In this session, I have coached a few members in the client team to generate a dashboard based off what I have showed. We had spent time discussing about prompt engineering and providing instructions to the team members, but the output of the dashboards could not be more different.

Think of AI as an intern. You would not hand a new intern to a staff member and say "work it out". The staff member needs to know how to brief the intern, check the intern's work and catch the confident mistake before it reaches a client. That is a leadership skill. And it exposes the paradox at the heart of this article: you cannot check AI's work on a task you have never done yourself. Supervising AI takes judgement, and judgement is exactly what juniors have not yet had the chance to build.

Your managers are already stretched

All of this pushes up what we expect from managers. They now review far more output than before, much of it produced by AI, and they are also expected to coach people on how to use it.

The Anthropic data fits that picture. The median employee there spends 20% more time in meetings than a year ago, managers' own code contributions have doubled, and 20% of staff described their role as shifting "from actually doing work to managing Claudes." Research cited in the same report finds that AI-native startups and heavy AI adopters tend to be flatter, and in our experience many businesses have removed a management layer expecting AI to fill the gap. In practice, the remaining managers are busier than ever, and there is less room for the informal teaching that used to turn juniors into seniors. (What you lose when you cut a layer deserves its own article, and we will come back to it.)

Rethink the org chart: human plus AI as 1.5

So what do we do next? Start with the org chart.

As a simple rule of thumb, which is our own working figure and not a measured one, treat each person working well with AI as about 1.5 people's worth of output. Some will do far better than that. But you only get to 1.5 if three things are in place.

a) A clear, human-centred AI strategy for the business. Decide, in writing, what AI is for in your company: which work it should take on, which work stays with people, and what you will do with the time it frees up. "Human-centred" means the plan starts from the work and the people doing it, not from the tool. Without this, every team invents its own rules and the gains scatter.

b) Proper AI onboarding and training for existing and new staff. Every new joiner gets a laptop password policy; very few get an AI policy. Cover the basics: which tools are approved, what may and may not be pasted into them, how to brief AI well, and how to check its output. For juniors, add one more thing: build in real practice on the underlying work, not just on the tool. Let them do a task by hand first, then with AI, then compare. That is how judgement gets built now that the filing cabinet is gone. Our guide to AI literacy as an L&D problem goes deeper on this.

c) A foundational leadership programme for managers. Managers need to learn how to manage and review the combined output of a person plus AI: how to set the brief, what to spot-check, and when to send work back. They should also learn to use AI themselves to help review the sheer volume of output now coming their way, without handing over the final judgement. Most first-time managers were never trained even before AI arrived, which is the problem we cover in The First-Time Manager Problem.

Then decide where the extra 0.5 goes. This is the part most businesses skip. If one person plus AI now produces 1.5 people's output, the tempting conclusion is that you need fewer people. Some of that gain should be reinvested in training the people who will run the business in ten years. If every bit of it goes to cost, you are borrowing from your future leadership bench.

Keep hiring the next generation

The next generation brings ideas, energy and creativity that no tool supplies. They are also your next customers and clients: they know how the people you will be selling to in five years think, buy and work. Shutting them out of your business means losing that insight as well as the future seniors.

The Anthropic report makes the point bluntly: protecting existing staff "will not benefit those not hired to begin with."

I understand the objection. In a slowdown, with tight margins, a junior hire who needs months of training feels like a luxury. But you do not have to choose between a full graduate programme and nothing. Internships, fractional roles and contract work are all lower-cost ways to bring younger people in, give them real work and see who grows. Training grants can take some of the cost off (our guide to training grants for Singapore SMEs in 2026 covers what you can claim). And if the training budget is the first thing to go when times get tight, it is worth reading why that cut costs more than it saves.

Secure your data first

More AI use means more of your company's information flowing into AI tools, often through many different people using many different tools, each in their own way. That is a data protection risk before it is a productivity gain. A staff member who pastes a client's details, a candidate's CV or a colleague's medical note into a public AI tool has created a PDPA exposure for the company, whether or not they meant to.

Singapore's Personal Data Protection Commission (PDPC) issued proposed advisory guidelines on personal data in generative AI in June 2026. They are not yet final, but the direction is clear: the organisation is accountable, not just the individual. Settle which tools are approved and what data may go into them before you roll AI out more widely, not after.

We have partnered with technology, data privacy and security organizations who are very qualified to answer these questions.

Find a partner who will get into the trenches with you

None of this is a software purchase. It is talent strategy and business strategy at the same time: what work your people should do, how they learn it, and who you hire next. It helps to have a partner who is willing to get into the trenches with you and understand your business deeply before recommending anything.

That is the work we do as a fractional CHRO: senior people strategy, part-time, alongside the founder. And when it is time to bring in the next generation, our talent acquisition team can help you find them.