A 21-year-old founder named Finlay Ekins wrote the sharpest short piece I've read this month on where business models are going. His argument: for twenty years, software was the prize and services didn't scale, because delivery scaled with headcount. AI flips both sides. Software learns the user instead of the user learning the software. And services — accounting firms, agencies, consultancies — can now run their admin and operations on agents, which means the delivery itself gets productised: the first engagement is custom, the fifth is mostly recipe, and the recipe becomes the product. What wins, he says, is the hybrid: a business that looks like a service from the outside and runs like a product inside. "We sell outcomes, not software." He quotes Sequoia: for every dollar spent on software, six are spent on services — and the next trillion-dollar company will be a software company masquerading as a services firm.
He's right. And from where I sit — I've bought these services as a CIO and I sell this model now — he's right in insurance first. The market evidence landed last month: in ISG's study of 150 senior insurance decision-makers, buyers ranked the routes they'll use to acquire AI capability, and new AI managed service providers came second of six — ahead of dedicated AI platforms, with incumbent BPO providers last. Not software. Not bodies. A service that runs like a product. The masquerade Sequoia predicted is, in this industry, already the buyer's stated preference.
The correction from the operator's chair
But Finlay's version has a hole in it, and the hole is exactly where regulated industries live. In his telling, the service layer is the wrapper — the human face, the relationship, the custom feel — and the product is the machine underneath. In insurance it's the other way round. The wrapper is not packaging. The wrapper is accountability, and accountability is the moat.
Here's why. When an agent working inside a carrier gets something wrong — a mispriced risk, a mishandled claim, a submission declined for the wrong reason — somebody has to own that outcome in front of a regulator, a broker, and a customer. Software can't own it. ISG's buyers said this plainly: 75% would trust only insurance-specific or governed AI for consequential decisions, and their conditions were controlled evidence, explicit permissions, and accountable human oversight. What the client is buying from the service firm is not the workflow. It's a named human who stands behind the workflow. Productised delivery is table stakes — everyone will have agents doing the work within two years. Governed accountability is the part that doesn't commoditise.
In regulated markets, the machine is not the moat. The signature under the machine is.
His "recipe" point needs the same translation. What gets codified by the fifth delivery in insurance isn't just playbooks and verification steps. It's decision-authority design: what the agent may execute on its own, what needs referral, where the limits sit, what evidence every action leaves behind. Agents pre-qualify, a licensed person confirms, the ledger keeps the tape. That codified judgment — earned engagement by engagement, failure by failure — is the product behind the service. It's also the part a competitor can't clone by copying your software, which is more than can be said for the software.
The friction he skips
Two honest taxes on the model, from inside it. First, enterprise insurance doesn't buy at startup speed: procurement, security review, model governance, state AI rules. The sales cycle is the price of the accountability premium — clients pay more precisely because you submit to all of it. Second, "we sell outcomes" means you carry outcome risk. Price an outcome without a baseline and the contract will eat your margin by the second quarter. Outcomes are a measurement discipline before they're a pricing model.
And his growth-over-cuts argument — Jensen Huang's point that AI means doing twelve times the work, not a tenth of the people — has a literal underwriting form here. Roughly one in nine broker submissions currently goes unquoted because of operational capacity, not appetite. Released capacity in a carrier doesn't just save cost; it converts into quoted and bound business. In this industry, efficiency done right is a growth line, and the firms structured to deliver it are services on the outside, products on the inside.
Finlay ends with advice to the young: pick one niche skill, sell the outcome, deliver it yourself, then build the machine behind you. Twenty years into doing some version of that in one industry, I'd only add the sentence his decade hasn't taught him yet: in regulated markets, the machine is not the moat. The signature under the machine is.
The wrapper is the moat.
— Krishna