Services are the new software. We got there by accident.

Sequoia published an argument called “Services: The New Software.” The core of it: for every dollar spent on software, roughly six dollars are spent on services. AI does not just improve the tools people buy. It absorbs the work people pay other people to do. A copilot sells the tool. An autopilot sells the work.

We did not read that essay and build a strategy. We ran a services business and the strategy fell out of it.

Flowleads is a paid services arm: AI agents, infrastructure, custom systems for enterprise clients in the US and Kazakhstan. Every engagement is somebody paying to have work absorbed by software. Which means every engagement is also a market signal. When three different clients pay to solve the same operational problem, that is not a coincidence. That is a product telling you it wants to exist.

This is an old pattern, older than the current AI cycle. 37signals built Basecamp as an internal tool for its own web agency; clients asked for it, and within about a year it out-earned the consulting business. Slack was the internal tool of a game studio whose game failed. The studios and agencies were not distractions from the product. They were the discovery mechanism.

The venture-studio numbers say the same thing. eFounders, now Hexa, launched around forty SaaS companies out of one operating engine and produced three unicorns with a failure rate far below the startup norm. Building inside an operating engine is not a compromise. It is a recognized, fundable model.

What the AI cycle changes is the economics of the loop. a16z’s 2026 notes describe the winning AI companies as orchestration plus domain-specific interfaces plus extreme specialization, not model wrappers. Specialization requires domain exposure. Domain exposure is exactly what a services book gives you, and it pays you while you accumulate it.

So our model is the essay in reverse. Sequoia’s argument runs from thesis to market: the TAM is labor spend, go absorb it. Ours ran from market to thesis: clients paid us to absorb work, the repeated work became products, and only later did we find the essay that explains why the loop compounds.

Kwanta is the worked example. It started as a client problem, shipped in production with a large client, and became standalone software we operate. Hackroda is the concentrated bet: the services arm already does QA work, so the product carries real client experience into a market where verification, not creation, is becoming the constraint.

One dollar of software, six dollars of services. We invoice on the six and build toward the one.