These days we hear a lot about how AI is going to allow companies to operate with fewer writers, designers, developers and other specialists.
And you know what - maybe that’s true… for now. Companies focused on short-term efficiency are already looking at specialist roles and asking how much of the work AI can do instead. If the answer they come up with is enough to reduce headcount and costs, the immediate business case can look pretty compelling.
However, I think there’s a longer-term consideration that many people are missing:
What happens to specialist expertise when fewer people can make a sustainable living developing it?
Can companies really get by without specialists in their workforce indefinitely? And even if they can for a while, what happens in the meantime? What might those decisions cost them - as well as the broader knowledge base that powers our economy - a few years down the track?
Specialist expertise has to come from somewhere
While many specialists produce outputs - whether that be a piece of written content, a design, a piece of code, or anything else that companies are trying to hand over to AI - they also produce expertise. And that expertise takes years of sustained practice to develop.
I know this from my own experience as a writer. I’ve written since childhood and have since spent more than 20 years practising professionally. The writer I am today is the cumulative product of thousands of hours spent doing the work.
If there had been no viable way for me to make money from writing when I entered the workforce, I obviously would have found another career. I might still have written for pleasure when I had time, and I might still have had the underlying talent. However, I wouldn’t have had two decades of practice producing written content in a specialised commercial context.
For many specialists, your skills can wax and wane depending on how often you use them. I’ve experienced this during my career. As I moved into senior leadership roles, management took up more of my time and I wrote less. I could feel the difference. When I eventually returned to writing every day, it took time to get back into the rhythm. My skills certainly had certainly not progressed while I was using them less. If anything, they’d gone backwards a little.
Expertise needs practice. For most of us, paid work is what gives us the opportunity to practise at that level.
Short-term productivity, long-term cost
This is where I think the current corporate productivity equation starts to become more complicated. When a company decides it can operate with fewer specialists because AI can produce more of their work, it’s usually measuring the output.
How many articles can we produce? How quickly can we generate code? How much design work can we get through? How cheaply can we do it if we eliminate humans from the process?
It’s harder to calculate everything else that happens when people spend years developing deep expertise inside an organisation. If I’m writing every day, I’m not just putting words on a page. I’m also learning what works in a particular context. I’m encountering problems I haven’t seen before and working out how to solve them. I’m always learning from other writers and applying what I learn to the work I do.
That’s how much specialist capability develops. You do the work, encounter something difficult and become better at dealing with it next time.
In my experience, AI can produce an acceptable version of something based on knowledge that already exists. However, somebody still needs to know what good looks like. Somebody needs enough expertise to recognise when the answer is wrong or mediocre, or when the way something has always been done could be improved.
Those people don’t magically appear when you need them. They develop over years, and they can fall out of practice just like anybody else if they’re not being used.
Where do the next experts come from?
None of this would be particularly concerning if only a small number of companies were reducing their reliance on specialists. AI is different because the same calculation is being made across industries at the same time.
We’re already seeing concern about the disappearance of junior roles as AI takes on work traditionally given to people at the beginning of their careers. The immediate impact is fewer jobs, but I feel the longer-term consequence deserves more attention.
Junior work is how senior expertise begins. Nobody enters the workforce with ten years of professional experience. People develop it by doing real work in real organisations, initially with guidance and eventually with enough experience to make decisions themselves. If you remove enough of those early opportunities, the effects may not become obvious for years because today’s experienced specialists are still available. But they will become obvious at some point.
The same applies to established specialists. If AI makes their profession less economically viable, some will leave. Their knowledge doesn’t immediately disappear, but they are no longer spending their working lives extending it.
People will continue to write, paint, design and code because they enjoy doing it, however there’s a substantial difference between practising something when time allows and spending thousands of working hours solving increasingly complex problems. Paid work creates the conditions for that depth of practice.
Who keeps producing knowledge in the AI era?
This raises a bigger question about what happens to our collective stock of expertise when AI changes the economic conditions that support its development. In a 2026 working paper, AI, Human Cognition and Knowledge Collapse, economists Daron Acemoglu, Dingwen Kong and Asuman Ozdaglar examine how increasingly capable AI could change people’s incentives to acquire knowledge themselves.
Their work has a broader focus than specialist employment. They model a world in which people can increasingly use AI to access knowledge and make decisions without investing as much effort in learning for themselves.
Under certain conditions, this can eventually produce what the researchers call a “knowledge-collapse” state, where people continue to benefit from AI, even as the stock of general human knowledge declines.
It’s an interesting concept when applied to specialist work. Employment is one of the mechanisms that supports the development of expertise. A scientist can spend years conducting research because there’s an economic structure that supports that work. A writer or developer can spend thousands of professional hours solving problems and improving their craft for the same reason. If you take away enough of that paid practice, you potentially change the conditions under which new expertise gets created.
For organisations, the consequences may take time to become visible. Companies can reduce specialist headcount today while continuing to draw on experienced employees and decades of accumulated knowledge. But what happens, over time, when fewer people inside the organisation are developing the expertise required to improve on that knowledge?
Could organisations eventually create their own form of knowledge collapse?
The AI knowledge problem
There’s another piece of research that’s worth examining here. In 2024, researchers including Ilia Shumailov published a study in Nature examining what happens when generative models are repeatedly trained on data generated by other models.
They identified a phenomenon known as model collapse. Across successive generations, AI models tend to accumulate errors and and information from the original data gets lost. One of the key findings of the study is that continued access to original human-produced data remains important for future models.
Model collapse and the human knowledge collapse examined by Acemoglu and his colleagues are different phenomena, but taken together they expose an interesting dilemma for the future of knowledge and corporate work.
AI has been built using an extraordinary accumulation of human-created knowledge. Future AI systems still need reliable human-produced information to stay viable. At the same time, AI may reduce some of the opportunities and incentives that allow people to spend their careers developing specialist knowledge.
To put it simply, the key question is “who creates what comes next?”.
Fast-forward to 2036
A company that reduces junior specialist recruitment in 2026 may still have plenty of senior expertise available today. By 2036, the people who would have accumulated a decade of experience during that period may never have entered the profession.
Across an industry, fewer entry points mean a smaller future pool of experienced practitioners. If established specialists also leave because their work becomes less commercially viable, that pool narrows further. That means there’s no one to recognise weaknesses in existing approaches and encounter difficult problems that lead to new solutions.
For me, this also raises an uncomfortable question about the so-called corporate commitment to continuous improvement. If an organisation has become so reliant on AI that it’s more focused on reproducing what already works while becoming less capable of developing what comes next, can it really claim a commitment to continuous improvement?
What’s the true cost of cutting specialists?
I believe that company leaders making workforce decisions today must also think beyond the immediate saving.
Where will their specialist capability come from in five or ten years? Who will have enough experience to recognise when AI-generated work is mediocre or outdated? Who will understand the discipline deeply enough to improve on what already exists?
Right now, AI gives organisations access to an extraordinary amount of accumulated human knowledge. For that knowledge to keep developing, people still need the opportunity to become experts.
Cutting specialist capability may be relatively easy in the short-term. Rebuilding it a few years could be considerably more expensive.
One thing I’m thinking about this week
If you watch this week’s video, you may notice a chunk of hair sticking out from the side of my head.
Somehow, I managed to sit in front of a camera and talk for 15 minutes without noticing it. I only spotted it afterwards, and by then I didn’t really have the time nor the inclination to film the whole thing again.
So the hair chunk stayed in. And given what I’ve been talking about for the past couple of weeks, it actually feels quite appropriate.
AI avatars don’t tend to have random pieces of hair sticking out at odd angles. Humans do.
There’s plenty about AI that’s smoother and more polished than we are. I’m increasingly comfortable with the fact that my online presence doesn’t really need to be.
Sometimes the hair sticks out - and that’s perfectly okay.
Things I’ve made (and you can buy)
If you enjoy my content and would like to support my work, here are a few things I’ve created.
🛒 Track Changes On Shop - Coffee mugs and drinkware for corporate rebels
📖 Do Give Up Your Day Job - Guide to corporate exits and transitions to self-employment
