Cover photo: Rodeo Project Management Software / Unsplash. Illustrative photograph.
I think AI is about to make a lot of very experienced people look surprisingly average.
Not because they suddenly got worse at their jobs.
Because the thing that made them valuable just got a hell of a lot cheaper.
For most of my career, experience has worked a bit like compound interest. You spend ten, fifteen, twenty years sitting in meetings, making mistakes, watching projects go sideways, learning how organisations actually work, getting better at writing, analysing, presenting and generally knowing what to do when somebody drops a vaguely defined problem on your desk and says, “Can you have a think about this?”
Eventually, somebody pays you quite a lot of money for all of that accumulated scar tissue.
Fair enough.
Except frontier AI has started doing something rather awkward. It is giving people with considerably less scar tissue access to a surprisingly large chunk of the capability we spent twenty years acquiring.
And I’m not sure most companies have properly clocked what that means yet.
The intern has entered the chat
I had one of those moments recently where something I’d been thinking about in the abstract suddenly became very real.
An intern was challenging a piece of thinking that, historically, would probably have sat comfortably above their pay grade. Not being irritating for sport. Not doing that delightful graduate thing where somebody discovers one McKinsey framework and attempts to restructure the company before lunch.
They actually had a point.
What struck me wasn’t simply that they had access to more information than an intern would have had ten years ago. Google solved that problem a long time ago. It was that sitting beside them, metaphorically speaking, was a frontier model capable of interrogating the issue, helping them structure the problem, testing assumptions, researching alternatives and turning half-formed instincts into a coherent argument.
That’s a much bigger shift.
For most of corporate history there has been an enormous capability gradient between somebody in year one and somebody in year fifteen. The junior person hadn’t seen enough yet. They didn’t know which questions to ask, couldn’t synthesise information as quickly, hadn’t developed the same writing or analytical ability and had very few patterns to draw on when the situation got messy.
So we built hierarchies around that gradient.
The analyst researched. The manager synthesised. The director reviewed. The partner wandered in fifteen minutes before the client meeting, changed three words on slide seven and somehow acquired most of the billing rate.
I may be oversimplifying slightly.
But only slightly.
What those layers represented wasn’t just increasing responsibility and judgement. They also represented increasing ability to produce good knowledge work. The more senior you became, generally the better you got at turning ambiguity into research, analysis, presentations, recommendations and decisions.
AI is starting to separate those things.
And that is where this gets interesting.
Some of the experience curve has just been bottled
There’s already evidence of this happening.
Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied more than 5,000 customer-support workers using a generative AI assistant. Productivity rose by 14% overall, but the gains weren’t evenly distributed. Novice and lower-skilled workers improved by around 34%, while the most experienced workers saw relatively little benefit. The researchers found evidence that the system was effectively passing the practices of stronger workers down to newer ones. (National Bureau of Economic Research)
That result has stuck with me because it suggests something more interesting than “AI makes people faster.”
It suggests AI can compress parts of the experience curve.
Things that used to be acquired slowly, by watching stronger people work, making mistakes yourself and gradually developing patterns, can increasingly be made available on demand. Not all experience, obviously. Nobody has managed to download twenty years of office politics into a vector database yet, although God knows somebody in HR is probably trying.
But some of it.
And once you start looking through that lens, you see the same pattern everywhere.
The £100,000 PowerPoint problem
I’ve noticed it repeatedly in my own work.
Take a decent strategy deck.
Not twenty slides containing blue gradient boxes, a photograph of six suspiciously attractive people pointing at a Post-it note and the word TRANSFORMATION written somewhere near the top.
An actual strategy deck.
Research. Argument. Commercial thinking. Narrative. Options. Recommendations. Something you could credibly put in front of a board.
Historically, producing one well might involve several people. A researcher gathers information. An analyst turns it into something useful. A consultant structures it. A manager beats it into shape. A senior person adds judgement and ensures nobody has accidentally recommended acquiring North Korea.
There is a perfectly rational reason consulting teams developed that way. Good knowledge work took time, and experienced people were expensive.
Now I can sit down with a frontier model, throw quite an ambiguous business problem at it, interrogate the reasoning, challenge the assumptions, add the bits it couldn’t possibly know and get to remarkably high-quality raw material before the traditional project team has finished arguing about who owns the Teams folder.
That’s not because the machine suddenly has twenty-five years of my lived experience.
It doesn’t.
It’s because quite a lot of what we’ve historically bundled under the word experience turns out to have been capability.
Research capability.
Writing capability.
Analytical capability.
Structuring capability.
Presentation capability.
And capability is getting cheap remarkably quickly.
This isn’t just me being impressed by my own PowerPoints, either. In a Harvard Business School experiment with 758 Boston Consulting Group consultants, people using GPT-4 on tasks within AI’s capabilities completed work more than 25% faster and produced outputs rated more than 40% higher in quality. Crucially, the weaker performers improved more than the stronger performers. (Harvard Business School AI Institute)
There was an equally important sting in that research: on a task outside AI’s effective frontier, people using it were actually more likely to get the answer wrong. (Harvard Business School)
Which is rather inconvenient for anyone planning to replace the entire management team with ChatGPT by Thursday.
The model can give you capability.
Knowing when not to trust it is something else entirely.
I saw the same thing in HR
A separate example came up around an HR issue.
Something needed analysing and escalating. Historically, that sort of problem naturally moved up the organisation. A junior person identifies the issue, somebody more experienced interprets it, someone else shapes the response and eventually a sufficiently senior person arrives carrying the great corporate superpower of being authorised to say what everyone else has already concluded.
Again, there were sensible reasons for that structure.
But increasingly I find myself looking at these chains and asking a fairly basic question:
What exactly is being escalated here?
If Claude can read the material, identify the issues, compare them against policy, surface inconsistencies, structure the available options and draft a sensible response, the mechanical reason for moving that work through three levels of seniority gets rather fuzzy.
There may still be an excellent reason to involve the senior person.
They may know that policy says one thing but the organisation has always done another. They may understand the individuals involved. They may know that technically correct option B will cause absolute carnage because Susan in Finance and Ahmed in Operations haven’t spoken since the Christmas party in 2019.
That’s context.
That’s judgement.
That’s valuable.
But notice what’s happened. We’ve moved away from “the senior person can produce better work” towards “the senior person knows things and can make calls the system can’t.”
Those are not the same source of value.
And quite a lot of corporate hierarchy has been priced as though they were.
So what is actually scarce?
This is where I think most conversations about AI and jobs go slightly off the rails.
We immediately ask whether AI will “replace people”, everyone picks a side, LinkedIn fills up with inspirational posts about uniquely human creativity and somewhere a man with “AI Futurist” in his bio produces a pyramid diagram.
I’m more interested in a simpler question.
What becomes valuable when competent cognitive output stops being scarce?
The more I work with these systems, the more I think the answer sits around a handful of things.
Judgement is the obvious one. AI is astonishingly good at generating plausible options. Choosing between them is different. Judgement is recognising which apparently insignificant detail matters, knowing when the data is technically correct but practically useless, and understanding that the rational answer on paper may be completely impossible inside the organisation expected to deliver it.
A lot of judgement is accumulated consequence. You’ve made the wrong call before. You’ve watched clever strategies collapse when they met real people. You’ve discovered that the best answer and the answer that can actually be implemented are sometimes distant cousins.
That stuff doesn’t disappear because a model got clever.
Context becomes more valuable for the same reason. A frontier model knows an absurd amount about the world. It doesn’t automatically know why your CEO really wants the project, what the customer said privately last year, why the system everyone claims is being retired still processes half the company’s revenue, or which apparently minor stakeholder can kill six months of work with one annoyed phone call.
Generic knowledge is getting cheap.
Proprietary context isn’t.
And then there are decision rights. Somebody still has to own the call. Spend the money. Hire the person. Fire the person. Sign the contract. Ship the product. Explain to the board why the thing everybody agreed was definitely going to work has developed a small smoking crater where the business case used to be.
AI can make a decision better.
It cannot magically inherit accountability for it.
Execution is about to have a very good decade
The one I find most interesting, though, is execution.
Ideas are getting cheaper. Research is getting cheaper. Analysis is getting cheaper. Plans are getting cheaper. Software prototypes that used to require a team can increasingly be produced by one person over a slightly antisocial weekend and an unreasonable number of conversations with Claude.
Which means saying what should happen becomes progressively less impressive.
Making it happen becomes more valuable.
Can you actually move an organisation? Can you convince people? Can you sell the idea? Can you build the system around the model? Can you ship something customers use? Can you deal with the bizarre edge case at 11:47 on a Sunday evening when reality politely informs you that your beautiful architecture diagram was complete fiction?
That distinction matters because professional life has always contained a surprisingly large number of people who are very good at describing work that somebody else should now go and do.
AI is going to be magnificent at that.
The interesting people will be the ones who can close the loop.
And then there’s point of view
This might end up being the most underestimated one.
If everyone gets access to roughly the same frontier intelligence, asking the machine for “a strategy” becomes progressively less useful.
You’ll get competent strategy.
So will I.
So will your competitor.
So will the 22-year-old graduate sitting three desks away who has absolutely no idea what EBITDA means but has wisely asked Claude not to tell anyone.
The differentiator becomes what you think.
What you’ve noticed.
Which trade-offs you’re prepared to make.
What you believe everyone else is getting wrong.
Which bit of the conventional wisdom you’ve tested and found wanting.
The model can help sharpen a point of view. It can challenge one. It can tell you about fifty others.
But if all you bring to the table is the ability to ask it for an answer, you’ve got a problem.
Because so can everybody else.
The middle of the organisation suddenly looks quite expensive
This is why I think the biggest AI story may eventually be organisational rather than technological.
Companies traditionally needed a lot of people in the middle because information and cognitive work were expensive to move around. People translated, reviewed, coordinated, synthesised, prepared, analysed and converted things from one corporate format into another corporate format, usually while adding another meeting to the diary.
The middle connected the people who knew things with the people who could decide things.
That structure made sense.
But if research, analysis, synthesis and preparation become dramatically cheaper, the economics underneath some of those layers start to change.
Not overnight. Organisations are marvellously good at preserving structures long after everyone has forgotten why they exist. I have personally attended meetings to prepare for meetings which themselves existed to decide whether another meeting was necessary.
The cockroach and the steering committee will survive anything.
But the pressure is there.
I suspect we’ll see smaller teams containing unusually capable people, operating with AI systems that allow them to span work that previously required several layers of specialists and managers. The organisations that figure that out won’t simply “use AI more”. They’ll start questioning why work moves through the company in the shape it currently does.
That is a much more uncomfortable conversation than buying everyone Copilot licences.
An awkward question for people like me
There’s another side to this that anyone with a few grey hairs should probably think about.
Your experience isn’t automatically a moat.
Neither is mine.
If twenty years of experience has mainly made somebody faster at finding information, better at writing reports, more polished at building presentations and more confident at explaining what everybody should do next, they’re potentially in a difficult spot.
AI is becoming unusually good at precisely that bundle of work.
The valuable part of experience is everything underneath it: patterns you’ve seen, mistakes you’ve survived, relationships you’ve built, taste you’ve developed, questions you know to ask, and the instinct that something doesn’t smell right even though the spreadsheet insists everything is green.
It is also the willingness to take responsibility for a judgement when the answer isn’t obvious.
That remains valuable.
But experienced people are going to have to combine it with these tools, because the future comparison isn’t really experienced person versus AI.
It’s experienced person using AI versus inexperienced person using AI.
And, eventually, experienced person who has completely redesigned the way they work around AI versus experienced person who occasionally asks ChatGPT to tidy up an email.
I know which one I’d rather be.
The junior opportunity is enormous
For younger people, I think the same disruption looks completely different.
The traditional experience curve hasn’t disappeared. But somebody has installed a lift next to the staircase.
You can interrogate subjects you barely understand, ask questions you would have been too embarrassed to ask your boss, study expert work, test your arguments, build things outside your formal role and walk into a meeting substantially better prepared than somebody at your level could have been even a few years ago.
That does not make you senior.
This distinction matters.
AI can give an inexperienced person access to senior-looking output long before they have senior judgement. Sometimes it gives them senior-looking confidence as part of the bundle, which is even more entertaining.
The dangerous junior employee is the one who mistakes a beautifully structured AI answer for wisdom.
But the dangerous organisation is the one that sees a young person genuinely operating at a much higher level and tells them to spend eight more years waiting for their turn because everybody else did.
Both are going to happen.
One is arrogance.
The other is bureaucracy.
The part you can’t rent from the model
I’ve spent a fairly unreasonable amount of time building with these systems.
And one thing keeps becoming clearer.
The model itself is rarely the interesting part.
The interesting part is what sits around it. Your context. Your systems. Your judgement. Your processes. Your feedback loops. Your domain knowledge. The decisions you make about where to trust it and where not to. And, perhaps most importantly, your ability to recognise when an extraordinarily articulate machine has just handed you complete bollocks.
The models will get better.
Which means raw capability gets cheaper again.
And that doesn’t make human experience worthless.
It exposes which parts of it were actually valuable all along.
For years we’ve been able to confuse seniority with output because the two were strongly correlated. The more experienced person usually could research better, write better, structure better and present better.
That relationship is weakening.
Which leaves organisations, and quite a few of us individually, with a wonderfully awkward question:
What is this person’s experience actually buying us?
Sometimes the answer will be enormous.
Judgement. Context. Trust. Accountability. Execution. A point of view earned through years of getting things right and, more importantly, getting some things spectacularly wrong.
Brilliant.
But sometimes I suspect the answer will turn out to be:
They make a really lovely PowerPoint.
And that is no longer a £200,000-a-year skill.
