Anthropic expects 30,000 firms to need AI orchestration platforms. They are not building a sales team to reach them. They are not cold-calling managing partners. They are not running outbound campaigns.
Think about what that tells you.
The company that built Claude — one of the most capable AI systems on the planet — does not believe AI adoption is a sales problem. They believe it is an application problem. The firms that see the opportunity will come to it. The firms that do not, will not — and no amount of selling will change that.
This is not how technology usually works. Software companies sell. SaaS vendors have business development teams, SDRs, account executives, partner channels. The entire go-to-market playbook for B2B software assumes you need to convince people to buy.
AI is different. And understanding why it is different tells you everything about where the real value sits.
The value is in the application, not the sale
Every AI vendor is selling the same thing: models. Capabilities. Benchmarks. Context windows. Parameters. They compete on performance metrics that most buyers cannot evaluate and do not care about.
None of that matters to a 30-person accountancy firm. What matters is: can this help me do something I could not do before? Can it do something I already do, but faster and better? Can it make my team more capable without making them more expensive?
That is application. And application requires something that no AI vendor provides and no sales team can deliver: someone who understands the work well enough to see where AI fits.
AI cannot act independently
This is the part that gets lost in the excitement. AI cannot define a good outcome. It does not know what success looks like for your firm, your clients, or your operations. It does not know which problems matter and which are noise.
Even at the most sophisticated end — pharmaceutical companies spending millions on drug discovery, biotech firms modelling protein structures, defence contractors running intelligence analysis — somewhere, a person or a team is saying: focus on this. If we achieve that, it is a good outcome.
The AI does not make that call. A human does. Always.
The question is which human. And the answer is not who most people expect.
The man who kicks the machine
In a production line, machines run routines. They follow programmed sequences. They process inputs and produce outputs, reliably, at scale. But every production line has a person — usually someone who has been there for years — who knows where to kick the metal press when it jams after a certain sequence of operations.
He is not the engineer who designed the machine. He is not the board member who approved the capital expenditure. He is the operator. He knows what the problem is. He knows what leads to it. He knows what a good outcome looks like — not in theory, but from having seen it ten thousand times.
That person — bizarrely, counterintuitively — is the most useful person to AI. Not the strategist. Not the executive. The person with tacit knowledge of how things actually work.
This is because AI's limitation is not capability. It is context. AI can execute almost anything you ask it to do. What it cannot do is know what to ask for in the first place. The man who kicks the machine knows. He has spent years accumulating precisely the kind of knowledge that AI needs and cannot acquire on its own.
Knowledge-based execution is being industrialised
The person who could draft a standard letter, summarise a report, extract data from a filing, prepare a compliance return — that work is being automated. It is not gone yet, but the trajectory is clear. Anything that can be reduced to explicit steps and applied to structured information is moving to AI.
This is not a threat to the knowledge operator. It is a liberation.
The person who knows which letter to send and why. The person who knows what that report actually means for the client relationship. The person who can look at a filing and see not just the data but the story behind it. That person just became more valuable, not less.
Their execution burden is dropping. Their judgment value is rising. The ratio is shifting decisively in favour of the people who know things that cannot be written in a manual.
AI adoption is pull, not push
The firms that benefit from AI are the ones where someone sees the opportunity. Either way, AI does not arrive through a sales pitch. It arrives through recognition. Someone inside the firm looks at their work, looks at what AI can do, and connects the two. This is why the 30,000 firms Anthropic expects will not be sold to. They will arrive because someone inside each firm saw the opportunity.
Tacit knowledge is the only sustainable advantage
If AI is commodity then the model is not the advantage. If AI cannot define outcomes then capability is not the advantage. The advantage is the knowledge that tells AI what to do, how to do it, and what good looks like. That knowledge is tacit. It lives in people.
The platform makes it easy to capture that knowledge. Constitution captures how your firm thinks. Knowledge hubs capture what your firm knows — the domain knowledge, the client intelligence, the operational reality. Together, they turn the man who kicks the machine into the firm's most scalable asset.
The firms that get this right do not need a sales team. They need the people who already know where to kick.