The VC Who Wants Credit for Noticing That Accountants Exist

Julien Bek at Sequoia Capital has a post making the rounds arguing that the next trillion-dollar company will be a services firm running on AI, not a software firm selling tools. He's right about the core idea. He's also writing a pitch memo dressed as analysis. Let's separate those two things.

The insight that holds up: if you sell outcomes instead of tools, you capture a larger and more defensible budget. The ratio he cites (six dollars spent on services for every dollar spent on software) isn't spin. It's the actual shape of enterprise spending. And his observation that every improvement in the underlying AI model makes an autopilot stronger rather than obsolete is the best point in the post. The tool company prays the next Claude doesn't eat their lunch. The services company orders dessert.

The intelligence/judgement distinction is also real. Some work is rules at scale. Some work is taste and experience and knowing which rule to break. AI crossed the threshold on the first category before the second, with software engineering being the clearest example. That framework is worth keeping.

Now the screwdriver.

The opportunity map is presented as if someone measured these things. They didn't. Bek plots insurance brokerage, accounting, legal, and claims adjusting on an intelligence-to-judgement spectrum, but the spectrum is an assertion dressed in a diagram. The problem with unexamined spectrums is that they hide the cases that matter most.

Take medical billing, which Bek describes as "almost pure intelligence" because you're mapping clinical notes to ICD-10 codes using rules. True for the routine cases. But the cases that generate audits, denials, and compliance liability are precisely the ones where documentation is ambiguous, the clinician made an unusual call, or the payer's interpretation of the rules differs from the provider's. The intelligence ratio in any profession isn't uniform across the work. It's bimodal: easy cases are mostly intelligence, hard cases are mostly judgement, and in regulated industries the hard cases are where the exposure lives. An autopilot that handles 80% of clean claims and escalates the rest is valuable. It isn't the same thing as "just close the books."

The outsourcing-as-wedge logic is his strongest practical argument. If a company already outsources a task, they've accepted external delivery, there's a budget line to substitute against, and you're doing a vendor swap rather than a workforce restructuring. That's a real distribution insight. What it skips is that those outsourced markets are already competitive. Law firms and accounting firms and staffing agencies aren't sleeping. Several of them are building the same tools Bek is funding. The displacement opportunity exists, but it's a race, not an open field.

The thing the article doesn't address at all is professional liability. Accountants, lawyers, and insurance brokers have durable businesses not primarily because they do intelligence work well, but because they carry professional responsibility for the output. There is a licensed human whose name is on the engagement letter. When it's wrong, there's someone to sue. An autopilot that closes the books needs to answer whose error it is when the books are wrong. That's not a rhetorical question. It's the question regulators, audit committees, and legal teams will ask before they sign the contract. The regulatory and liability infrastructure for AI-native services firms is unsolved, and Bek doesn't mention it once. That's a telling omission from a piece that otherwise reads like a sourcing post for Sequoia's next fund.

The opening line ("the next $1T company will be a software company masquerading as a services firm") is a hook, not a claim. You can agree with everything that follows and still recognise that the framing exists to make you feel like you're reading about category-defining companies rather than about billing process automation.

The intelligence/judgement framework is worth extracting and keeping. The trillion-dollar opener is worth ignoring. Most viral VC content works exactly this way: a genuine observation wrapped in stakes-raising that the observation doesn't actually support. The observation is good enough to stand without the wrapper. It usually is.