Copilot sells the tool. Autopilot sells the work — and services firms are rebuilding around it

The distinction that matters is not whether a services firm uses AI — nearly all of them now do — but whether AI changes what the firm sells and how it prices. Bolting a copilot onto an unchanged billable-hour engagement is a productivity tweak. Rebuilding delivery and pricing around AI doing the work, with people supervising and handling exceptions, is a different business. The firms making that second move are the ones Sequoia is describing, and they are starting to show up in law, accounting, and compliance — which is to say, in the exact markets a firm like ours sits next to.

Bek's argument starts from a market-size observation that is easy to state and hard to unsee: for every dollar enterprises spend on software, they spend roughly six on services — consulting, agencies, back-office labor, professional firms. SaaS spent two decades taking share of the software dollar. Bek's claim is that AI capable enough to complete a task end-to-end, rather than assist a human completing it, can now compete for the much larger services dollar instead — insourced and outsourced labor combined, not just software budgets.

The copilot/autopilot split is the operating distinction underneath that claim. A copilot suggests, drafts, and recommends; a human still does the work and is accountable for the output. An autopilot closes the books, resolves the ticket, drafts and files the brief, and a human supervises and handles exceptions rather than doing each step. Bek's line on the risk of staying a copilot: sell the tool and you're exposed every time the underlying model gets better and someone rebuilds your feature in a weekend; sell the outcome and you're delivering something a client already budgets real money for, at a price tied to results rather than seats.

Legal is the case Sequoia points to because the numbers are hard to argue with. Harvey, the legal-AI platform, raised $200 million at an $11 billion valuation in a round co-led by Sequoia and GIC that closed 25 March 2026 — up from an $8 billion valuation just three months prior. More than 100,000 lawyers across roughly 1,300 organizations, including most of the AmLaw 100, now run substantive legal work through it. Sequoia has co-led three rounds in the same company, which is unusual for a firm that typically diversifies risk across a portfolio, and reads as high conviction that legal services, not legal software, is the market being won.

What makes Harvey a services-as-software company rather than a legal-tech tool is what it's priced and measured against: hours of associate and partner time on document review, drafting, and research, not a per-seat SaaS license. That's the pattern Bek says generalizes — the total addressable market for an autopilot is the entire labor line item in a category, and revenue is bounded by how many engagements the system can run, not by how many software seats a client is willing to buy.

The same restructuring is showing up in accounting and audit, and not purely because the AI got good enough. CPA firms are heading into 2026 with roughly three-quarters of partners eligible to retire within a decade and thin entry-level pipelines behind them — a workforce shortfall that makes agentic delivery a capacity fix as much as an efficiency one. Purpose-built autopilot products (tax-prep agents like Black Ore's Tax Autopilot, for instance) now handle document intake, classification, and first-pass preparation end-to-end, with review and sign-off the human step that remains, mirroring Harvey's model almost exactly.

Legal regulators are already testing what a fully AI-native firm looks like structurally rather than just operationally: the UK's Solicitors Regulation Authority authorized Garfield.Law in 2025 as a firm delivering legal services entirely through AI under human supervision, and 2026 commentary frames the AI-native firm as a restructuring of how legal service is delivered and regulated, not a faster version of the same firm. Compliance functions — audit trails, exception handling, regulatory reporting — are a natural next target precisely because that work is procedural and auditable, which is what makes agentic systems governable there in the first place.

The uncomfortable question this thesis puts to any professional or technical services firm — including one like ours — is not "are we using AI" but "does AI change what we sell and how we charge for it." A firm that adds an AI copilot to an unchanged, hourly-billed engagement has improved margin on the old business model. A firm that restructures delivery so AI does defined units of the work end-to-end, with experts supervising and priced against the outcome rather than the hour, has built a different business — one that can serve more engagements without linearly adding headcount, which is exactly the scale constraint Bek says autopilots break.

For law firms, compliance functions, and other services businesses watching this shift from outside, the practical takeaway is the same one that keeps surfacing in agentic AI generally: the hard part isn't picking a model, it's redesigning the workflow, the review points, and the pricing around what the model can now finish unsupervised. That's a scoped, buildable project, not a research problem — which is the part worth acting on before a leaner, AI-native competitor does it first.