Sequoia Capital partner Julien Bek published a thesis in early March that has been making the rounds in VC circles for a good reason: it tells every copilot founder they're building the wrong thing. The original essay, titled "Services: The New Software," argues the next trillion-dollar company won't sell tools. It will sell the work itself.
The core provocation is simple. For every dollar a business spends on software, it spends six on services: consultants, agencies, outsourced ops teams. AI agents are now capable enough to start eating that six-dollar side of the ledger directly. Why sell lawyers a faster contract review tool when you can sell the business a finished NDA?
Intelligence work is already cooked
Bek splits professional work into two buckets. "Intelligence" is rule-bound complexity: translating a spec into code, mapping clinical notes to ICD-10 codes, reconciling ledgers. "Judgement" is the messy stuff that requires taste, experience, pattern recognition built over years. Which feature to build next. Whether a candidate fits the culture. When to ship something half-baked.
His claim is that AI crossed the intelligence threshold already. Software engineering got there first, which tracks with the data point he cites: developers account for over half of all AI tool usage, while every other profession is still in single digits. That ratio feels right, though it's worth asking who measured it and how. Sequoia doesn't cite a source for the figure, and "AI tool usage" could mean a lot of things depending on how you count it.
The copilot trap
Here's where it gets uncomfortable for a lot of funded startups. If you built a copilot in 2024 or 2025 and now want to become an autopilot, you have a problem: selling the work directly means cutting your own customers out of doing it. Bek frames this as a classic innovator's dilemma. The law firm using Harvey to review contracts faster is Harvey's customer. If Harvey starts selling contract review directly to businesses, that law firm becomes a competitor.
"Every founder building an AI tool is asking the same question: what happens when the next version of Claude makes my product a feature?" Bek writes, and the bluntness is refreshing for a VC essay. He's basically telling copilot founders they're in a race against the model providers, and every model improvement erodes their moat.
The escape hatch, per Sequoia, is to sell outcomes from day one. Bek names a handful of companies already doing this. Crosby sells drafted NDAs to businesses, not tools to law firms. WithCoverage sells insurance directly to CFOs, not software to brokers. The common thread: they're replacing the outsourced vendor, not augmenting the professional.
Where the money actually sits
The essay includes an opportunity map plotting services verticals by intelligence-to-judgement ratio and outsourcing prevalence. A few stand out. Insurance brokerage ($140-200B) is the biggest dollar market listed, and Bek argues the broker's value-add in standard commercial lines is "essentially shopping across carriers and filling forms." That's a bold claim that a lot of brokers would dispute, but for commodity policies he's probably right.
Accounting ($50-80B outsourced in the US alone) has an acute structural problem: the US has lost roughly 340,000 accountants over five years while demand grew, and 75% of CPAs are approaching retirement. The profession can't replace itself fast enough, which makes it less about AI disrupting unwilling incumbents and more about AI filling a vacuum that already exists.
Recruitment and staffing ($200B+) is the largest market on the list but also the trickiest. Screening and matching is pure pattern recognition. Closing a candidate is not. Bek acknowledges the split but doesn't dwell on how hard it is to sell "we hired your engineer" as a service when the hiring manager wants to feel like they chose someone.
What the thesis leaves out
The essay is clean. Maybe too clean. One criticism that's been circulating, as noted by a Medium analysis from Han Heloir Yan, is that Bek "writes from 30,000 feet" and the terrain looks smooth from up there. The piece doesn't address liability. When the autopilot drafts a bad NDA or miscodes a medical claim, who's responsible? SaaS vendors don't carry malpractice risk. Services companies do.
There's also the quiet conflict of interest. Many companies cited in the essay are Sequoia portfolio investments. Crosby, Rillet, WithCoverage, Mercor. Bek is essentially publishing a thesis that describes his own deal pipeline, which doesn't make him wrong but is worth noting when evaluating how inevitable he makes this all sound.
The 6:1 services-to-software ratio is a compelling framing device, but the source isn't cited. Is that a global figure? US-only? Does it include all professional services or just tech-adjacent ones? The number does the rhetorical work of a fact without the footnotes of one.
So now what
Bek's call to action is direct: if you're building a pure-play autopilot, email him. The thesis reads less like analysis and more like a deal memo shared publicly, which is arguably the most honest thing about it. Sequoia is placing bets that the copilot-to-autopilot shift will define 2026 the way copilots defined 2025, and they want founders to know their door is open.
Whether or not the "next $1T company" framing proves accurate, the underlying shift feels real. AI models will keep getting better, and every improvement helps the company selling work more than the company selling tools. The founders who internalized this a year ago are already ahead. The ones reading Bek's essay now are playing catch-up, which is probably exactly the urgency Sequoia intended.
