“A startup could rebuild our whole platform in a weekend.”
They're probably right, that's the unsettling part. Whatever it is you spent four years and a great deal of money constructing, some small team can now approximate before Monday, and they can do it because the thing that used to be expensive — writing the software — has become abundant, and almost free.
If anyone can build it, the people who used to be paid for building it are exposed. The moat was the code, the code is now abundant, therefore the moat is gone. It's the obvious trade – value drains out of what's suddenly plentiful – and as far as it goes, it's correct. The only trouble is where everyone assumes the value drains to.
In most industries, that claim still holds true. But it's worth asking whether it was ever true here — in a hospital, a pharma company, a clinic. Because a competent team could always build the app. A sharp teenager could always write the diagnostic logic. The code was rarely the wall anyone actually hit.
Which leaves an awkward question, and the rest of this piece is really just an attempt to answer it. If the building was never the hard part — what, exactly, was that decade spent on?
The Rate Limiting Step
To conveniently answer my own question…It was spent waiting. And proving. Running the trial, filing the submission, answering the regulator's questions, running the trial again. Getting the thing cleared, then getting it reimbursed, then getting a hospital to actually trust it enough to change what a clinician does on a Tuesday morning. The software was the easy 5%. The permission was the other 95%, and the permission is what took the years.
This is worth sitting with, because it changes what "cheap software" actually means. Every process has a slowest step i.e the one that sets the pace for everything else. Chemists call it the rate-limiting step. You can speed up every other reaction in the sequence, and it buys you nothing, because the slow step still governs how fast the whole thing runs.
Here it is, building was never the rate-limiting step in healthcare. Clearance was. Trust was. So making the building cheap does not make the medicine arrive faster, and it does not make the incumbent weaker. It speeds up a step that was never the bottleneck — which means the value that we imbue building with doesn't evaporate. Instead, it has to go somewhere else.
And so we now have our underlying thesis: cheap code doesn't drain the moat, it moves the moat to whatever the code still has to get past.
Moats you build, moats you're given
But not every slow step is slow for the same reason. Some bottlenecks are slow because the thing behind them is genuinely hard to accumulate. Proprietary data. A distribution network. The workflow habits of a hundred thousand clinicians who already use your system and don't want to relearn one. These are moats companies built — earned over years, expensive to replicate, and yours. Call them built complements.
The trouble with a built moat is that it can be out-built. It's scarce because it's difficult to accrue. Someone with more capital, more time, or a cleverer approach can accumulate what you accumulated. Cheap software makes this worse by making it easier for competitors to build. Abundance erodes built moats.
Then there's the other kind. Some bottlenecks are slow because someone has to let you through. A regulatory clearance. A prescribing licence. A reimbursement code. The legal standing to be the party held responsible when the model is wrong. These are not accumulated like built complements. Instead, they're issued or granted by an authority that has decided, deliberately, to restrict who gets one. We’ll call these granted complements.
You cannot out-build a granted complement, because effort is not the currency it responds to. You can be the most capable team in the world, sitting on the next breakthrough model, and it changes nothing until someone with a stamp decides you may proceed. The lone teenager who could write the diagnostic logic still cannot write themselves a licence to use it on a patient. That was true when code was expensive and it stays exactly as true now that code is free. This is exactly the point. Cheap code erodes every moat except the granted one.
Breaking Bad spends five seasons on this distinction, though it never calls it that. Walt is the best chemist in the American Southwest; the chemistry is the one thing that was never his problem. Everything that actually threatens him lives on the granted side of the ledger — territory, protection, a way to move product through channels other people control. His built complement is world-class and nearly worthless on its own, because the scarce thing was simply the permission to operate. Most of healthcare runs on the same asymmetry, minus the bodies (I hope).
There is an alternative universe where cheap generation completely dismantles where value falls, but gatekeepers are still important. Instead it dissolves everything that wasn't the gate and leaves the gate standing alone, now holding all the value that used to be spread across the whole process.
History repeating itself
This phenomenon mirrors the oldest patterns of what happens every time a means of production suddenly gets cheap.
Take the printing press. The usual platitude is that cheap copying shifted value to authors, and eventually it did. However, within a few generations of the press taking root in England, the Crown had handed a single company of stationers an absolute monopoly over what could legally be printed — and out of that monopoly, almost as an accounting device, came the very thing we now call copyright: the registered right of one member to a text that no other member could touch. And so it stands to reason, abundance of copying didn't abolish the gatekeeper. On the contrary, it actually helped manufacture one. The moment reproduction stopped being scarce, the right to reproduce became the thing worth owning
Closer to home now, in Pharmaceuticals. For most of the industry's history, the chemistry got steadily cheaper and more routine. Barriers to putting together a working production line dropped, while the thing that got harder was permission to sell it with a claim on the label. After thalidomide, the 1962 Kefauver-Harris amendment made proving efficacy mandatory, and in doing so relocated the entire industry's centre of gravity. The approval is now the competitive moat. Cheap chemistry, expensive permission. This is cheap code & expensive clearance wearing different clothes.
The lesson each time is the same, and it's counterintuitive: making one input abundant doesn't flatten an industry. It relocates the scarcity, usually toward whoever controls the permission that the newly-cheap input still has to pass through.
Sort your moats
So what do you do with this, if you're the one accountable for where your company's value actually lives?
The useful response is somewhat of an audit. Map every advantage you think you have and sort it into two piles. On one side, the things your company has built — your data, your integrations, the fact that forty thousand clinicians already know your interface and resent the idea of learning another. On the other, the things you were granted — your clearances, your reimbursement codes, your licences, the fact that you are the one legally standing behind the output.
In the former pile, consider that a huge chunk of these are exactly what cheap software erodes. A data moat is wonderful until a competitor's model needs less data. Workflow lock-in is real until switching costs collapse because the new system configures itself. These were solid moats in a world where building was expensive – this is no longer. Much of what felt like strategy was really just the incumbency that expensive software used to buy you for free.
The second pile is where the durable value now sits. This is where most incumbents make their actual mistake. They see a regulatory position, a hard-won clearance, a reimbursement pathway, and they file it under "moat" and stop thinking. But a granted advantage has a property a built one doesn't: you don't own it. You hold it at the discretion of whoever granted it, and that party can change the terms without asking you.
This is why durability and security are not the same thing. A granted moat is durable in the one sense that matters against competitors — no rival can out-build it, because effort isn't the currency it answers to. What it isn't, is secure. For instance, a regulatory authority can, with a single decision, widen the gate and let a crowd through by introducing an over-the-counter pathway, an AI-specific fast track or a reclassification, rendering your once scarce and valuable asset commonplace. Your best moat is the one most exposed to a stroke of someone else's pen.
This new way of thinking changes what "defending your position" means. A built moat you defend by building. A granted one you defend by watching the grantor — a different discipline entirely, and one most firms still run as a compliance function rather than a strategy.
If the durable value in your industry is increasingly granted rather than built, then influence over the terms of the grant stops being back-office housekeeping and becomes the main event.
The individual version of the same problem
The same sorting works one level down, on a single career.
A clinician's expertise splits into the same two piles. There's the built part — the diagnostic reasoning, the pattern recognition, the accumulated judgment of ten thousand cases, even the patient relationships they’ve nurtured. Then there's the granted part: the licence, the authority to sign, the standing to be the human answerable when a decision goes wrong.
It's tempting to read this piece as a warning that the built part is under threat, and in a narrow sense it is; the diagnostic reasoning is exactly what the models are climbing toward. But that's not the same as saying the skill has lost its value.
Skill has lost its scarcity, which is a different thing entirely. The reasoning still makes the clinician good. It's simply no longer the rarest thing they own. What stays scarce is the permission i.e the part a model cannot hold no matter how well it reasons, because you cannot grant a licence to something that cannot be held responsible.
Which is worth saying plainly, because the two halves are not rivals:
A licence defended by no real skill is just liability with a signature.
The granted authority is only worth anything because a built competence stands behind
it.
The clinician's position hasn't hollowed out, but its centre of gravity has shifted from the skill that used to be scarce toward the authority that still is. The work of the next decade is less about being the one who can reach the judgment, and more about remaining the one permitted to stand over it.
