Guess Who Becomes AI Native
The pyramid was the product
Last week I published a piece about what AI is doing to professional services. One short section in it said the entry level is where this lands first, and that most firms will quietly decide not to train people on work they can no longer bill.
One of my partners read it and came back with the hardest version of that argument I have heard from anyone.
He is not a skeptic. He is the most technically forward person I work with, he is halfway through building an order management system by himself, and he was not talking his book. He was telling me my optimism has a hole in it.
“There is no developer in the future. When every software developer is out of work, the scarcity goes away. And the rates plummet.”
The strongest case that the juniors are finished
Let me lay out his argument in his order, without softening it, because a case you have to weaken before you answer is not a case you have actually answered.
The offshore arbitrage is gone. AI eliminates the margin on offshore development. That spread was the engine underneath a very large part of this industry, and it does not come back.
The team got small. Two smart people. That is all this takes now. Maybe some QA. He is not speculating about that, he is living it, alone, halfway through a system that would have been a funded team two years ago.
So how do you bill the money required to grow? If the work takes two people, the revenue for that work supports two people. You cannot staff your way to scale on a base that no longer exists.
And the perverse part. Since he is the only one who knows how the thing was built, his personal value goes up. The individual wins while the model breaks. That is exactly the incentive that stops firms from fixing this.
Then the line that stuck with me for two days.
“The new children will be useless.”
Not cruel. Accurate, under his assumptions. And it follows logically from something I have already written.
The junior’s job was the rote work. That is how the apprenticeship functioned: you did volume under supervision, and the domain went in through your hands over about four years. Automate the rote work and you have not just removed a task, you have removed the classroom. Worse, it was a classroom the client paid for.
His scarcity argument is the one that actually bites, though, and I want to be honest about it. Right now being genuinely fluent with these tools is rare, so it pays. His prediction is that it stops being rare. Everybody has it, it becomes table stakes, and the rates fall to the floor. Being AI-native buys you nothing once it is the baseline.
I told him the truth, which is that I could not disagree with him.
“I cannot disagree with you. I just need to see.”
That was not a polite concession. He may be right. But underneath his case there is a prediction, and predictions can be tested.
The test, which is the only honest position for an investor
Here is where I landed with him, and it is deliberately unsatisfying.
We do not need to be the innovators here. We need to find the firms that have already figured it out. If they are out there, we will find them and we will back them. And if they do not turn up, that in itself is the answer.
That is not a forecast. It is a search with a falsifiable result, which is worth more than either of us being clever at each other over text. My whole thesis rests on those firms existing. If a year of looking turns up nothing, my thesis is wrong, and I would rather learn that from the market than defend it in an argument.
So the rest of this is the other side of the case. Not a rebuttal. The part I think his version leaves out.
Where I think the bear case breaks
It assumes everybody becomes AI-native at roughly the same time.
They will not. It is not close.
Most people with a job today do not have the luxury my partner has. He can disappear into a coding agent for a weekend because his role rewards exactly that. Most people cannot. They have a manager, a set of deliverables and a review cycle that measures the thing they were hired to do. Learning to work this way means being visibly slower for a while, in front of the people who rate you. Very few organisations make that a safe thing to do.
Most people also have lives. Everybody I know who got genuinely good at this did it on nights and weekends. That is not available to someone with small children, a commute, a second job or a parent to look after. It is not unwillingness. It is arithmetic.
And a large share of people simply do not want to change. That is not a character flaw. Most careers reward doing the known thing extremely well, and most people are correctly reading the incentives they actually face.
So subtract all three groups and ask who is left.
The people with no workflow to protect, no reputation built on the old method, and time. The current college graduates.
That is the inversion in this cycle, and it is the opposite of what everyone is saying out loud. The cohort with the least experience has the least to unlearn.
Every previous technology wave rewarded experience, because experience was pattern recognition and the patterns still held. This one splits the two. It rewards judgment, and it actively punishes the muscle memory that experience installs. A twenty two year old is not carrying twenty years of instinct about how the work is supposed to be done.
The anomaly problem, which is the real crux
Then my partner made a point that reframed the entire thing for me, and it worked against his own argument, which is how I knew it was the important one.
“I do not think many business experts will also become AI IDE experts. We are anomalies.”
He is right, and it is the most useful sentence in the whole exchange.
Last week I wrote about the forward engineer: someone who knows the business requirements because they lived them, and who builds directly without gathering, documenting or convincing anybody. That works because thirty years of domain knowledge and real fluency with the tooling happen to sit inside one head.
That combination is rare. You cannot staff a firm out of it. You cannot recruit for it, because the people who have it are already doing their own thing. You cannot train it in a year, because half of it takes a career.
Any model that requires anomalies is not a model. It is a lottery ticket with a business plan attached.
So stop requiring it. Split the pair.
One expert, many builders
The answer is not the forward engineer. It is the pairing.
One person who knows a business deeply. Several AI-native builders who do not need to know that business, because the expert supplies it, and who can produce far more than the expert could have directed under the old constraints.
I have run this myself rather than theorised about it. I built a real, working product using roughly five percent of my partner’s time. I did not need to know his domain. He supplied the judgment in small, high-value doses. I supplied the direction and the build.
Which means that in that arrangement, I was replaceable. Specifically, I could have been replaced by a recent college hire who is AI-native, working with the same five percent of the same expert.
That is not a threat to me. It is the whole opportunity.
The business expert is the new scarce good. Spread one of them across many problems, with AI-native builders attached to each, and you have leverage again.
Read that sentence back and notice what it describes. One expensive person at the top, several cheap fast people underneath, output multiplied across more ground than any of them could cover alone.
That is a pyramid.
What this actually does to the pyramid
Here is the part that matters for the industry rather than for any one career.
Consulting and systems integration have run on a pyramid for fifty years. One partner, a few managers, a lot of juniors. You billed the juniors at a healthy multiple of what they cost you, and the spread across that wide base was the margin. Growth meant hiring more base. That is not a side effect of the model, it is the model.
The bear case says the pyramid dies, because the base of it was rote work and the rote work is gone.
I think the base changes composition and the pyramid survives, but only at firms that rebuild it on purpose, and quickly.
The old base was cheap hours. Leverage over labour. You made money on the gap between what a junior cost and what a client paid for their time.
The new base is machine capacity, directed by fast, cheap, AI-native people. Leverage over judgment. You make money on the gap between what one expert’s judgment costs and how many problems it can now be applied to.
The ratios move violently. Instead of one partner over twenty juniors grinding through rote work, you get one domain expert over a handful of AI-native builders covering more ground than the twenty ever did.
Two consequences, and firms are going to hate both of them.
Headcount stops being the growth engine. If revenue no longer scales with bodies, it has to scale with revenue per person, which means the commercial model has to change. That is the pricing problem from last week’s piece, arriving from a different direction. The pyramid and the rate card were always the same machine viewed from two angles.
Seniority stops being the organising principle. The pyramid was a proxy for experience, and experience was a proxy for judgment. Now the thing you are levering is judgment directly, and it does not distribute neatly by tenure. Some of the best judgment in a firm sits at level three. Some people at level eight are carrying instinct that is now a liability. No title structure survives contact with that honestly.
And the training problem inverts rather than disappearing. The old apprenticeship taught the domain by making juniors do rote work inside it. The new one has to teach judgment while they direct machines, which is harder, needs to happen faster, and nobody has a curriculum for it. The firms that write that curriculum first will not have a talent problem for a decade.
The transformed and the transformers
The last thing worth separating is where all of this lands, because conflating the two is why this debate goes in circles.
The transformed are vertical services companies. A telehealth business, a specialty clinic group, a logistics operator, an insurance servicer. In those companies labour is a cost line, not a revenue line. Nobody bills a patient by the hour for the back office. So when AI takes cost out of the operation, it falls straight through to margin, and there is no pricing problem to solve at all. The only real question is how fast they adopt.
That is why I expect the first large, visible margin expansion from this wave to show up inside vertical services companies rather than inside consultancies. They get to keep the entire benefit. Nobody negotiates it away from them.
The transformers are the consultancies, agencies and integrators. Same capability, opposite economics. Their labour is the revenue line. For them every efficiency is deflation until they change how they sell, and they have to rebuild the pyramid while it is still carrying the whole business. It is the harder job by a distance.
It is also the bigger prize, because whoever solves it does not just fix their own P&L. They get to sell the solution to everyone in the first category.
The young professional sits in a completely different position in each. In a transformed company they are a cost centre that just got dramatically more productive, which is straightforwardly good for them. In a transformer they are the product, and the definition of the product just changed underneath them.
So what would I tell a twenty two year old
Four things, and the third one is the only one that compounds.
Your disadvantage is smaller than you have been told. You are not behind on twenty years of domain knowledge in a race that rewards domain knowledge. You are ahead on the one thing that is hard for everyone above you, which is having no habits to unlearn.
But my partner is right that fluency alone is a commodity. By the time you are thirty, being good with these tools will be like being good with a spreadsheet. It will not be a career. Do not build one on it.
What compounds is judgment. Knowing what is worth building, why it matters to the business, and whether the output is any good. That is the half the model does not have, and it is the half your employer cannot buy from a vendor.
So attach yourself to a domain expert and take everything they know. That pairing is the highest-leverage arrangement available in business right now, and here is the part nobody will tell you: they need you at least as much as you need them. Their knowledge is stranded without someone who can build at the speed it deserves.
The uncomfortable version of that advice, for the firms rather than the graduate: your best juniors’ optimal move is to attach to your best expert, absorb the domain in three years instead of ten, and then leave to do it for themselves.
Which is precisely the supply of new firms I described last week. The minnows are going to be staffed by the people the incumbents decided not to train.
Where I actually land
I am not certain, and I said so to him in the moment rather than afterwards.
But notice what his case requires. It requires a whole population to become AI-native at roughly the same time. Nothing in thirty years of watching technology arrive suggests that populations move like that. Adoption is always slower, lumpier and more political than the capability curve, and the gap between what is possible and what is normal is where every fortune in my career has been made.
So the scarcity he expects to evaporate probably lasts a good deal longer than he thinks. Not forever. Long enough for one cohort to build an entire career inside it.
The bear case is right about the technology and wrong about the people. That has been the shape of every one of these I have lived through.
The rest is not an argument, it is a search, and at Bambu Capital it is the one we are running. Find the Services-Tech firms that already work this way. Back them. And if a year of looking turns up nothing at all, then he was right and I will say so here.
The pyramid was the product.
The base of it just changed.
Whoever rebuilds it first owns the decade.
A six part series on what AI is doing to professional services.
One. AI Decides How. Humans Decide What, Why, and If. Governing AI inside a single firm, and where the human veto has to sit.
Two. Nobody Is Buying Hours Anymore. What happens to an industry that priced and sold the one thing the machine turned out to be best at.
Three. Guess Who Becomes AI Native. You are here. Whether this wipes out a generation of young professionals, and what becomes of the pyramid.
Four. The Offerings Transform Too. The second wave, where the platform makes your catalogue unnecessary, and why productising is the wrong answer.
Five. Services-Tech. Naming the category, and the line between the services businesses that get rebuilt and the ones that do not.
Six. Nobody Is Ripping Anything Out. The AI layer over the systems you already own, the migration nobody has named, and why it is the first good news in the series.