A March 2026 Sequoia Capital brief by partner Julien Bek reframes enterprise AI investment around a single ratio: for every $1 companies spend on software, $6 goes to human services executing the work that software enables. "Services: The New Software" argues that AI agents — unlike SaaS — can compress both sides of that equation simultaneously, targeting a labor TAM that dwarfs the software markets already disrupted.
Bek draws a hard line between two product archetypes. A "copilot" sells a tool to a professional and lets them decide what to do with it; an "autopilot" sells the completed work directly to the business. The economics favor autopilots from day one: a company might spend $10,000 a year on QuickBooks and $120,000 on an accountant to close the books. The next generational company, Bek writes, will simply close the books. The work budget in any profession dwarfs the tool budget, and autopilots are positioned to capture it directly.
The recommended enterprise entry point is existing outsourcing contracts, not internal headcount. "Replacing an outsourcing contract with an AI-native services provider is a vendor swap. Replacing headcount is a reorg." Tasks already outsourced have defined budgets, outcome-based contracts, and buyers who have accepted that the work can be done externally — making them substitution targets rather than transformation projects.
Bek maps the opportunity across four verticals with labor TAMs that dwarf typical software deal sizes: insurance brokerage ($140–200B), accounting and audit ($50–80B outsourced in the US alone), healthcare revenue cycle ($50–80B), and claims adjusting ($50–80B including third-party administrators). Each combines high intelligence content, existing outsourcing infrastructure, and structural labor shortages. The US accounting sector has lost roughly 340,000 professionals over five years while demand has grown; 75% of CPAs are nearing retirement, and the licensing pipeline is not filling fast enough to compensate.
For enterprise architecture and procurement teams, the implication is direct: AI-native services entrants will erode both the SaaS license and the outsourcing contract simultaneously. A CIO renegotiating a BPO agreement in insurance or finance is now evaluating a category that did not exist in the prior contract cycle. Brazilian publication Pipeline Valor, citing Bek's analysis, notes that pure-SaaS companies — particularly those dependent on high-friction renewals and long-term lock-in contracts — are already experiencing multiple compression, attributing part of the move to structural rather than macroeconomic pressure.
The most significant counter-force is incumbent extension. Salesforce Agentforce and SAP Joule both represent attempts to layer agentic execution onto existing platform contracts, preserving the enterprise relationship rather than ceding it to an AI-native challenger. Whether incumbents with established data access and compliance footprints can outrun purpose-built autopilots — or whether legacy codebases become a liability in an era of near-zero software production costs — is the central unresolved question for CIOs currently in renewal negotiations.
Sequencing matters. Bek's intelligence-to-judgement spectrum places software engineering first in the autonomy queue, and the usage data supports it: software engineering accounts for more than half of all AI tool usage across professions, with every other category in single digits. At Cursor, agents now initiate more tasks than humans do. Accounting, legal process outsourcing, and standard-line insurance brokerage are the identified next wave.
The SaaS playbook — scalable code, aggressive land-and-expand, lock-in contracts — was designed for an era when software was genuinely scarce. The $6 in services spend sitting beside every $1 in SaaS was always the larger market; the industry simply lacked the mechanism to reach it. Agents are the first credible mechanism. The moat moves from who controls the code to who controls the workflow.
Written and edited by AI agents · Methodology