Services: The New Software
Why the next trillion-dollar company will sell not software, but the work itself — and what that means for Axevil's deal flow.
Source: Julien Bek, Partner at Sequoia Capital. Essay "Services: The New Software," March 5, 2026. This briefing was prepared by Axevil Capital for internal and partner distribution.
A software company disguised as a services firm.
The next $1 trillion company will be a software company disguised as a services firm. It won't sell a tool for professionals — it will sell the finished result of the work: closed books instead of an ERP, an issued policy instead of a brokerage platform, a finished NDA instead of lawyer access.
The economic lever is simple: the budget for the work itself, in any profession, is many times larger than the budget for tools. By capturing the labor budget instead of the software budget, an autopilot plays in a market 6x larger from day one.
Copilot vs Autopilot.
One sells a tool. The other sells finished work. These are different business models and different bets for the decade ahead.
Sells a tool
AI in the hands of a professional; the human decides how to use it and remains accountable for the result. A continuous dialogue between human and system.
Vulnerability: competes with every new model release; hands domain knowledge back to the human.
Sells the work
The client buys the finished result directly. The process is automated "like in aviation": a human monitors, takes over the hard parts, and stands ready to intervene.
Advantage: gets cheaper to run with every model improvement, instead of more vulnerable.
"Autopilot" ≠ a full replacement of the human expert by AI. The analogy is an aircraft autopilot: most of the process is automated, but a human stays on standby and handles the hardest parts ("takeoff and landing").
Intelligence versus judgment.
What gets automated first, what stays with the human, and where the ideal autopilot target lives.
Intelligence
Complex but ultimately procedural tasks. Coding to spec, running tests, translating clinical notes into ICD-10 codes, screening resumes. The rules are complex — but they are rules. Automatable.
Judgment
Experience, intuition, accountability. Stays with the human. The higher the share of judgment in a task, the slower autopilot wins.
Example: medical billing sounds like "medicine = judgment," but the coding layer is almost pure intelligence in a mature outsourcing market.
TAM in the US labor market.
The largest verticals where autopilot competes not with the SaaS budget, but with the budget for the work itself.
Bek maps roughly a dozen verticals with TAM estimates. Above are the largest and most-cited from the essay.
Where the defensibility lives, and why the window is months, not years.
The moat is data, not the model
An accumulating dataset of domain decisions: what "good" means in brokerage or medical coding. Every completed task teaches the autopilot something a copilot never learns — because a copilot hands that knowledge back to the human.
The moat is trust
The window to become a trusted AI services firm in a vertical is measured in months, not years.
The 2025 copilot dilemma
They have a product and a customer, but selling the work means cutting their own customers out of the process. This is a window for pure 'day-one' autopilots (Rillet vs. Basis).
Unresolved
Go-to-market: enterprise services sales don't scale like software sales; margins are lower than pure SaaS (~70% versus ~90%) due to inference costs.
A new class of pre-IPO assets — AI services.
One entry filter for deal flow, one watchlist for Market Intelligence.
Screen any AI startup at intake with one question: does it capture the labor budget, or only the software budget? An autopilot in an already-outsourced vertical is a category for priority screening.
Watchlist for the Market Intelligence Portal — first-tier companies in each relevant vertical:
WithCoverage
Autopilot brokerage — sells insurance directly to CFOs.
Harper
AI-native brokerage in a $140–200B market.
Crosby
Sells finished NDAs to companies, instead of access to lawyers.
Rillet
AI-native ERP that closes the books.
Pace · Strala
AI-native claims-settlement operations.
Sierra
The margin benchmark for the autopilot economy.
Related context: both OpenAI and Anthropic are building sales channels around PE firms; PE shops are betting on rolling up non-software businesses, making them AI-efficient, and reselling them at a higher price.