Competence at AI implementation is not a durable differentiator
Firms reaching for AI implementation excellence as their competitive position are betting on a skill that commoditises on the same curve as the expertise it applies to.
When a firm accepts that AI is going to reshape its market, the most common first move is to decide it will be the best in its segment at implementing AI. The reasoning is intuitive: if the technology is the threat, master the technology before your competitors do.
The move does not work, and the reason it does not work is the same reason the firm felt threatened in the first place.
Why it fails
Implementation skill is knowledge. It is knowledge about how to structure a problem for a model, how to assemble context, how to configure a tool against a particular domain — and it is codifiable, publicly discussed, and improving fastest in exactly the places the models are trained on. It commoditises on the same curve as the domain expertise it is applied to, and not far behind it. If knowing how to do trust accounting is heading towards free, then knowing how to configure an AI to do trust accounting is heading towards free too.
The advantage a firm builds this way is also unusually short-lived, because the thing that erodes it is not a competitor catching up but a vendor shipping. Capability that took a quarter to assemble arrives natively in the platform, available to every firm at once. This is the demand-side consequence of the supply-side movement described in Retrieval middleware is being absorbed into platforms at mid-tier scale: being six months ahead on tooling is an advantage that expires with the next release.
What it does not mean
It does not mean implementation quality is unimportant. A firm that implements badly will get worse results than one that implements well, and the gap is real while it lasts. The claim is narrower and about strategy rather than operations: implementation competence is table stakes, not a position. It is something to be good enough at, not something to build a market identity on.
The same distinction appears one layer down in Hire for durable AI judgement, not transient AI mechanics, which separates durable judgement from transient mechanics at the level of individual skills, and one layer across in Architect AI around principles, not vendors, which makes the case for durability in architecture. This note is the firm-strategy version of the same argument.
Where the durable position is instead
If implementation skill decays, the question becomes what does not. The answer runs in the opposite direction — see Captured knowledge is the one AI advantage that compounds. The useful reframe for a partner group is to stop asking how to use AI better than competitors and start asking what the firm knows that the AI does not.