Sequoia Capital published one of the most consequential frameworks for understanding the AI economy in March 2026: "Services: The New Software," written by partner Julien Bek. The central claim is provocative and data-backed — the next trillion-dollar company will not sell software. It will sell the work itself.
The essay identifies legal work ($60 billion), accounting and audit ($50-80 billion), and healthcare revenue cycle management ($50-80 billion) as the largest near-term opportunities for AI "autopilots" — systems that don't assist workers, but complete the work autonomously and deliver outcomes.
The Core Thesis
Sequoia's essay frames the opportunity around a simple observation: for every dollar spent on software, six dollars are spent on services. The global services market — outsourced work of all kinds — is approximately six times larger than the software market. If AI can capture even a fraction of that spending by delivering completed outcomes instead of tools, the TAM is enormous.
The key distinction Bek draws is between:
- Copilots: AI tools that make workers more productive by assisting with tasks (drafting, research, coding, etc.)
- Autopilots: AI systems that take ownership of complete outcomes — delivering the work itself, not just assisting with it
In 2025, the fastest-growing AI companies were copilots. Bek's thesis is that 2026 marks the transition where the most ambitious companies try to become autopilots — selling completed work rather than productivity tools.
The business model implication: an autopilot charges per outcome (per contract reviewed, per tax return filed, per claim processed) rather than per seat or per API call.
The Three Big Targets
Legal: $60 billion addressable
Artificial Lawyer's coverage of the Sequoia analysis breaks the legal opportunity into two buckets:
- Paralegal and LPO work: $36 billion — document review, legal research, contract management, e-discovery, compliance monitoring
- Legal transactional work: $20-25 billion — contract drafting, due diligence, regulatory filings, standard transactional document production
The underlying logic: companies that already outsource legal work to law firms or legal process outsourcing (LPO) providers are already in the right mindset to see that work move to autopilot. The outsourcing relationship established the precedent; AI autopilots extend it.
Accounting and audit: $50-80 billion addressable
The US has lost approximately 340,000 accountants over five years while demand has grown. This supply-demand imbalance creates structural pressure toward automation. Sequoia estimates $50-80 billion in US accounting and audit work is currently outsourced — and that the intelligence layer of that work (not the judgment and advisory layer) is highly automatable.
Healthcare revenue cycle: $50-80 billion addressable
Medical billing and coding represents Sequoia's cleanest autopilot example: ~70,000 standardized ICD-10 and CPT codes, rule-based adjudication logic, and enormous outsourced volume make it structurally ideal for AI automation. The intelligence work is almost entirely codifiable — if not perfectly, then well enough that human review of AI-generated outputs is 80-90% faster than human-first workflows.
Insurance brokerage: $140-200 billion
The largest identified opportunity: standard commercial insurance lines are highly standardized, volume-intensive, and already heavily outsourced — making them prime autopilot targets.
Why "Autopilot" Is the Right Frame for 2026
The copilot-to-autopilot transition matters for a specific reason: pricing model shift. Copilots are priced per seat — $20-$50/user/month for Copilot licenses, regardless of what work actually gets done. Autopilots are priced per outcome — $X per tax return, per contract reviewed, per claim processed.
For buyers, this is a fundamentally different risk profile. You pay for outputs, not tools. You know your cost per outcome before committing to the service. The ROI calculation is direct.
For providers, the risk is that AI must reliably deliver the promised output quality. Unlike a copilot (where the human is responsible for output quality and the tool is responsible for productivity), an autopilot takes responsibility for the result.
Sequoia's AGI Claim
Separately, Sequoia published an essay titled "2026: This Is AGI" — arguing that 2026 AI systems already satisfy a practical definition of artificial general intelligence for knowledge work contexts. The claim is not that machines have human consciousness, but that in the domain of knowledge tasks, AI can now reliably outperform the median human worker on speed, consistency, and breadth.
This claim is directionally consistent with the services autopilot thesis: if AI has crossed the competence threshold for standard knowledge work, the case for paying human LPO providers, offshore bookkeepers, and medical billing services is weakening for the specific task categories Sequoia targets.
What This Means for BPO and VA Industries
The Sequoia thesis is both a threat and an opportunity for the outsourcing and virtual assistant markets:
The threat: If AI autopilots can reliably produce legal document reviews, tax returns, and medical billing at scale for a fraction of human labor cost, the volume-based LPO and BPO businesses that depend on handling standardized work are at risk.
The opportunity: The transition from copilot to autopilot requires someone to implement, monitor, QA, and manage the AI systems. Human expertise doesn't disappear — it concentrates in oversight, exception handling, and the judgment layer that autopilots genuinely cannot serve.
Bolster Legal's analysis captures this well: "The future of legal outsourcing lies in hybrid models, blending AI-driven technology with offshore human expertise and in-house attorney oversight." The same principle applies across accounting, healthcare billing, and insurance.
For virtual assistant services, the Sequoia thesis reinforces the shift toward specialized, judgment-intensive VA roles — and away from volume-based administrative work that AI autopilots will increasingly serve.
Market Adoption Timeline
The autopilot transition is not simultaneous across all service categories. The realistic adoption curve:
- First movers (2026-2027): Highly standardized, rules-based work with clear quality metrics — medical coding, document classification, contract redlining on standard templates
- Mid-term (2027-2029): More complex document work where judgment matters but workflows are clear — due diligence, audit sampling, claims adjudication
- Long-tail (2030+): Work requiring genuine contextual judgment — complex litigation strategy, bespoke M&A structuring, novel regulatory interpretation
The vast majority of BPO and VA revenue sits in the first-mover category or the long-tail — not the clean middle. The most vulnerable providers are those operating purely in the standardized task categories with no human judgment differentiation.
The Takeaway
Sequoia's "Services: The New Software" is the most important strategic framework for understanding the AI economy in 2026. The billion-dollar business model is not tools that assist workers — it's services that deliver outcomes. The $60 billion legal and $80 billion accounting opportunities are the clearest near-term targets.
For businesses that consume outsourced services, the question is: which of your current outsourced functions are about to get dramatically cheaper? For providers of those services, the question is: what is your plan for the part of your value proposition that an autopilot will deliver at 80% lower cost? The human judgment layer that autopilots cannot replace is exactly what virtual assistant services are designed to deliver.
Sources: