Services: The New Software
Sequoia by Julien Bek
AI tools are turning into services that do the work, not just help with it. That could let software companies eat budgets once reserved for humans.
Based on reporting by Sequoia, by Julien Bek — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Sequoia’s latest thesis is blunt: the next giant software company may look more like a services firm. That sounds backwards until you follow the money. If a startup sells a tool, it ends up racing the model. If it sells the work, every model upgrade makes the service faster, cheaper, and harder to beat.
The cleanest example is bookkeeping. A company might spend $10,000 a year on QuickBooks and $120,000 on an accountant to close the books. Sequoia’s point is that the valuable business is not the software seat; it is the work itself. That same logic applies wherever the task is mostly execution and the judgment can stay with a human for now.
Software engineering got there first. A year ago, Cursor users mostly treated AI as autocomplete. Now, more tasks are started by agents than by humans, and software engineering makes up more than half of AI tool usage across professions. Sequoia argues that coding is mostly intelligence work — translating specs, testing, debugging — while judgment is what to build next, when to ship, and how much tech debt to accept. AI has cleared the first part.
That split matters because it changes how AI companies should enter a market. A copilot sells the tool. An autopilot sells the work. Sequoia says the best wedge is usually an outsourced task, because the budget already exists and the buyer already cares about the outcome. Crosby, for example, started with NDAs, a narrow task many companies already send to outside counsel. Replace the vendor, not the headcount, and the sale gets easier.
The firm’s priority map runs through insurance brokerage, accounting and audit, healthcare revenue cycle, claims adjusting, tax advisory, legal work, IT managed services, procurement, recruitment, and consulting. But the pattern is the same: start with intelligence-heavy work that is already outsourced, then move toward the messier insourced stuff as the system learns what good judgment looks like. Sequoia thinks 2025 belonged to copilots. 2026 may be the year a lot of them try to become autopilots.
My take — AI-written commentary, not fact-checked reporting
This is the least sexy AI thesis and probably one of the better ones. Selling outcomes instead of widgets is how you sneak past the model commoditization trap, which is a polite way of saying tool makers should stop building their own replacements. The market keeps rewarding companies that hide behind “productivity” when the real prize is taking over the work.
Read more about this at: Sequoia
