For every dollar spent on software, $6 goes to services. If AI is going to make real money, it needs to sell outcomes — not tools.
What Is It?
Sequoia Capital partner Julien Bek published a piece in March 2026 called "Services: The New Software" — and it got people talking. Ben Tossell (Ben's Bites) picked it up in his newsletter and zeroed in on the key fork in the road for AI business models.
The core framework is straightforward. AI products split into two camps:
Copilot sells an AI tool to lawyers. Autopilot sells finished contracts to companies. The key is that Copilot is after your software budget, while Autopilot is after your services budget — and services budgets are 6x bigger.
Sequoia goes one level deeper. They split all work into "intelligence" (rules-based tasks) and "judgment" (tasks that require human discernment). The more intelligence-heavy the work, the faster Autopilot moves in.
What Changes?
| Copilot (selling tools) | Autopilot (selling outcomes) | |
|---|---|---|
| Customer | Professionals (lawyers, developers) | Companies (businesses that need a legal team) |
| Pricing model | Per-seat subscription | Outcome-based pricing (per task, per resolution) |
| Budget targeted | Software budget | Services/outsourcing budget |
| As models improve | Risk of being replaced by competitors | Faster and cheaper → competitive advantage |
| TAM | Software market ($1T) | Services market ($6T+) |
Two cases make this distinction crystal clear.
Harvey (Legal AI) — Moving from Copilot to Autopilot
Harvey started by selling Copilot tools to law firms — helping lawyers research faster and draft documents. Now it's moving toward directly delivering finished contracts, NDAs, and legal filings. The results? As of March 2026: $11B valuation, $190M ARR, and over 100,000 lawyers on the platform. Most of the AmLaw 100 and 500+ corporate legal teams are customers.
Intercom Fin — The Textbook Case for Outcome-Based Pricing
Intercom's AI agent Fin charges $0.99 per resolved customer inquiry. Not per seat. Fin only gets paid when it actually solves a customer's problem. No resolution, no charge. That's the Autopilot pricing model in practice. It's a pretty easy sell for customers — if you don't get results, you don't pay.
Sequoia's "Autopilot opportunity map" is worth a look too. Work that's high in intelligence and already being outsourced is the number-one target for Autopilot:
Deloitte backs up the trend. Their 2026 predictions report says seat-based subscriptions will shift toward usage- and outcome-based pricing. Gartner forecasts that more than 40% of enterprise SaaS spending will move to usage- or outcome-based models by 2030.
Getting Started
If you want to position your AI product or service as an Autopilot:
- Start with work that's already outsourced
Work that's already being outsourced proves three things: there's no cultural resistance to external delivery, a budget exists, and customers buy on outcomes. Replacing an existing outsourcing contract with AI is a "vendor swap." Replacing internal headcount is "organizational restructuring." The first one is a much easier sell. - Measure the intelligence-to-judgment ratio
The more rules-based repetitive work (intelligence) involved, the faster the Autopilot transition. Think: NDA drafting, insurance quotes, tax filings. Save the judgment-heavy work for later, once you've built up enough domain data. - Charge for outcomes, not access
Design a model like Intercom's: "$X per resolution". Per-seat pricing is the Copilot game. Outcome-based pricing lowers your customer's resistance to buying — and the better your product performs, the more you make. - Build a structure where better models become your advantage
Copilot products are threatened every time a new foundation model ships — your tool could get absorbed into ChatGPT's default feature set. Autopilots get better as models improve: same outcome, cheaper and faster to deliver. That structural difference is the whole game. - Starting as a Copilot is fine — just have a migration plan
The Harvey playbook — start as a Copilot, build your customer base and domain data, then gradually shift to Autopilot — is a legitimate strategy. Here's the thing though: Sequoia flags a real innovator's dilemma here. When you make the shift from Copilot to Autopilot, you risk alienating your existing users (the professionals) whose jobs you're now automating.
Heads Up
Most of the companies Sequoia mentions — Crosby, Rillet, WithCoverage, Magentic — are Sequoia portfolio companies. The analytical framework holds up, but keep in mind that the specific case studies reflect an investor's perspective.
