Primary Source: Services: The New Software by Julien Bek, Sequoia Capital — Published March 5, 2026 Reading Time: 7 minutes (original article ~1,400 words) Research Compiled: April 2026
The headline claim: "The next $1T company will be a software company masquerading as a services firm."
Sequoia's argument is that the AI era's biggest winners won't sell tools to professionals — they'll sell the work itself directly to end-customers. The distinction is existential: if you sell the tool, every model improvement from Anthropic, OpenAI, or Google threatens to commoditize your product into a feature. But if you sell the work (the outcome), every model improvement makes your service faster, cheaper, and harder to compete with.
The concrete example Bek gives: A company might spend $10K/year on QuickBooks (software) and $120K on an accountant to close the books. The next legendary company will just close the books.
This is not about building better software. It's about replacing the entire service delivery layer — the accountants, brokers, adjusters, coders, recruiters — with AI-native firms that look like services companies on the outside but run on software economics on the inside.
The ratio that underpins the entire thesis: For every $1 spent on software, $6 is spent on services.
This is the critical economic insight. The global software market is roughly $700B. The global professional services market is $6T+. The entire SaaS era — from Salesforce to Snowflake — captured value from the tool side of this ratio. Sequoia is arguing the AI era will capture value from the services side.
Why this matters:
- SaaS companies sell seats/subscriptions (the $1)
- Services companies sell hours/outcomes (the $6)
- AI autopilots can capture the $6 at software margins (~80%+ gross margin vs. ~30% for services firms)
- The TAM for "selling the work" is 6x larger than "selling the tool"
Bek's outsourcing wedge: The easiest markets to attack are those already outsourced. If a company already pays an external firm to do the work, the buyer has already accepted that the work can be done by someone outside the organization. The AI autopilot just needs to do it better, faster, and cheaper. No organizational change management required — just a vendor swap.
The article identifies specific characteristics of ideal autopilot markets:
- High intelligence-to-judgement ratio — the work is rule-following, not taste-making
- Already outsourced — buyer is used to paying for outcomes, not tools
- Fragmented incumbents — no single player controls the customer relationship
- Aging workforce — structural labor shortages accelerating adoption
- Standardized work product — quality is verifiable without deep domain expertise
This is the conceptual backbone of the article. Bek draws a sharp line between two types of work:
Intelligence work: Complex but rule-based. Translating a spec into code, testing, debugging, translating clinical notes into ICD-10 codes, shopping insurance quotes across carriers. The rules are complex but they are rules.
Judgement work: Requires experience and taste. Deciding which feature to build next, whether to take on tech debt, when to ship before it's ready, strategic recommendations, culture-fit assessments in hiring.
The key insight: AI has crossed the threshold where it can do most intelligence work autonomously and leave the judgement to humans. Software engineering got there first — "more tasks are started by agents than by humans" in Cursor. Every other profession is still in single digits for AI tool usage. The reason: software engineering is primarily intelligence work.
The convergence: Today's judgement will become tomorrow's intelligence. As AI systems accumulate proprietary data about what good judgement looks like in their domain, the frontier shifts. This creates a compounding advantage for early movers — they're not just serving customers, they're building the training data that will let them handle judgement-heavy work in the future.
A copilot sells the tool. An autopilot sells the work.
| Dimension | Copilot | Autopilot |
|---|---|---|
| Customer | The professional (lawyer, accountant, broker) | The company that needs the outcome |
| Revenue source | Tool budget (~$10K/yr) | Work budget (~$120K/yr) |
| Value prop | Makes professionals more productive | Replaces the engagement entirely |
| Model risk | Every model improvement threatens the tool | Every model improvement strengthens the service |
| Responsibility | Professional takes responsibility for output | Autopilot company takes responsibility |
| Pricing | Per-seat SaaS | Per-outcome (per claim, per filing, per hire) |
The innovator's dilemma for copilots: Existing copilot companies (Harvey for law, Rogo for investment banking) have the product and customer knowledge. But transitioning to autopilot means cutting their own customers out of doing the work. This is the classic innovator's dilemma — your best customers are the professionals you'd be replacing. That's the opening for pure-play autopilots.
The convergence thesis: Copilots and autopilots will eventually converge. Copilots accumulate data about professional decision-making, which eventually lets them handle more judgement. Autopilots start with high-intelligence work and gradually expand into judgement-adjacent tasks. But starting position matters because it determines where you can win customers now and begin compounding data.
Bek maps out 10 specific markets with named companies. This is effectively Sequoia's public deal pipeline:
| Market | TAM (US) | Intelligence Ratio | Named Companies | Key Dynamics |
|---|---|---|---|---|
| Insurance Brokerage | $140-200B | High | WithCoverage, Harper | Largest dollar market. Pure intelligence work (shopping carriers, filling forms). Incredibly fragmented — tens of thousands of small brokers. |
| Recruitment & Staffing | $200B+ | Mixed | Juicebox, Mercor, Jack & Jill | Largest services market. Top-of-funnel is pure intelligence, closing/culture-fit is judgement. Wedge: high-volume, low-judgement roles. |
| Management Consulting | $300-400B | Low (mostly judgement) | TBD | Huge market but hardest to automate. Question: can consulting be disaggregated into intelligence (data gathering) and judgement (strategy)? |
| Supply Chain & Procurement | $200B+ | High (long-tail) | Magentic, AskLio, Tacto | 80% of suppliers get zero negotiation attention. Contract leakage runs 2-5% of procurement spend. Wedge: "abandoned work" — found money. |
| IT Managed Services | $100B+ | Very High | Edra, Serval | Patching, monitoring, provisioning, alert triage. Nobody sells "your IT runs" as an outcome yet. Existing tools (ConnectWise, Datto) sell to the MSP, not the company. |
| Accounting & Audit | $50-80B outsourced | High | Rillet, Basis | US lost ~340,000 accountants over 5 years. 75% of CPAs nearing retirement. Structural shortage forcing faster AI adoption than any other profession. |
| Healthcare Revenue Cycle | $50-80B outsourced | Very High | Anterior | Medical coding = translating clinical notes into ~70,000 ICD-10 codes. Rules are complex but they are rules. Outsourcing already mature and outcome-based. |
| Claims Adjusting | $50-80B incl. TPAs | High | Pace, Strala | Adjuster workforce aging out. Market massively outsourced to independents and TPAs (Crawford, Sedgwick). Same industry as insurance, but distinct autopilot surface. |
| Tax Advisory | $30-35B | High (80-90% intelligence) | TaxGPT, Skalar, Ravical | CPA licensing creates regulatory moat. Multi-jurisdiction complexity is exactly what SMBs outsource. Every jurisdiction deepens the data moat. |
| Legal (Transactional) | $20-25B | High | Harvey → autopilot, Crosby, Lawhive | Contract drafting, NDAs, regulatory filings. Standardized enough that quality is verifiable. Harvey moving from copilot to autopilot; Crosby and Lawhive are autopilot-native. |
Total addressable market across these 10 verticals: ~$1.4T-1.7T in the US alone.
Critical observation: Sequoia is not just theorizing. They've named 20+ companies across 10 verticals. Several are Sequoia portfolio companies (Edra was announced as a Sequoia investment on March 18, 2026, days after this article). This is as much a deal sourcing manifesto as an investment thesis.
This thesis doesn't exist in isolation. It's part of Sequoia's broader "Cognitive Revolution" framework, articulated across multiple publications:
"The $10 Trillion AI Revolution" (August 2025) — by Konstantine Buhler
- Positioned AI as bigger than the Industrial Revolution
- Drew a timeline: GPU (1999) → AI Factory (2016) → Cognitive Assembly Line (202X)
- Industrial Revolution compressed from 144 years to 17 years for AI equivalents
- The question isn't IF but WHO will be the Rockefeller/Carnegie of AI
AI Ascent 2025 Keynote (May 2025) — Partners Pat Grady, Sonya Huang, Konstantine Buhler
- AI represents a market opportunity "at least 10x larger than cloud computing"
- Cloud computing created ~$1T in enterprise value → AI opportunity is $10T+
- 5.6 billion people live online, making distribution physics fundamentally different from prior tech revolutions
- The value is shifting rapidly to the application layer
- The race is on for vertical, agent-first solutions
- Key quote from keynote: "You are in a run-like-hell business right now. The rails are in place. The opportunity is wide open."
"2026: This is AGI" (January 2026) — by Pat Grady and Sonya Huang
- Declared that AGI is effectively here (or imminent)
- This provides the technological foundation for the "autopilot" thesis — models are now capable enough to handle intelligence work autonomously
The intellectual progression:
- AI is a $10T revolution (May/Aug 2025)
- AGI is here or imminent (Jan 2026)
- The biggest companies will sell services, not tools (Mar 2026)
Each builds on the previous. The services thesis is the applied investment strategy derived from the broader AI revolution thesis.
Current Role: Partner, Sequoia Capital (Europe), Seed/Early stage — since September 2023 Previous: Vice President & Principal at Accel Partners (2018-2023), Investment Associate at Global Founders Capital Education: Graduate program at Cambridge University Location: London, United Kingdom Motto on Sequoia profile: "Discomfort is a privilege."
Background: Bek comes from a family of entrepreneurs. Parents were a recruiter and an astrologist — he learned "a lot about reading people." He had a successful startup in high school and a less successful one at Cambridge. Prior to Sequoia, he spent 5 years at Accel as a Principal focused on European venture investments.
Relevant context: Bek founded "Smart Outsourcing SL" in 2013. The name is revealing — his personal entrepreneurial history is literally in outsourcing, which is the exact wedge he identifies for autopilots. He's been thinking about this space for 13+ years.
Investment focus at Sequoia: Seed/early stage, European market. The companies named in the article skew heavily toward European startups (Skalar, Ravical, Lawhive are Europe-based), consistent with his geographic mandate.
🔴 The Margin Trap (Linas Beliūnas, "Linas's Newsletter")
- Title: "Sequoia's Thesis Will Mint Billionaires and Bankrupt Copycats"
- The "$0.03 problem": AI inference costs are declining rapidly. If the work can be done for $0.03, what's the sustainable margin?
- Commoditization risk: Every autopilot in a vertical is using the same foundation models. Differentiation comes from data and workflow, but those moats may be thinner than assumed.
- Many companies will try to execute this thesis simultaneously, creating a bloodbath in each vertical.
🔴 Surviving to Scale (N6 Finance, Ryan Gaines)
- "The autopilot thesis is compelling. But for most founders, the harder question isn't what to build, it's whether you can survive long enough to earn the right to sell outcomes."
- Three underappreciated challenges:
- Sales complexity: Selling outcomes requires trust that selling tools doesn't. You're asking the buyer to let AI do the work, not just assist.
- Error cost asymmetry: When a copilot makes an error, a human catches it. When an autopilot makes an error, the customer eats it. The liability surface is fundamentally different.
- Capital intensity: Autopilots must front the cost of service delivery while building trust. This requires more capital than SaaS businesses, creating a potential valley of death.
🔴 The "Building" Problem (Han Heloir Yan, Medium/Data Science Collective)
- "Services Are the New Software: Building Them Is the Hard Part"
- Thesis oversimplifies the intelligence/judgement distinction. Many tasks that look like intelligence work have embedded judgement that's invisible until the AI gets it wrong.
- Domain expertise is harder to encode than Sequoia implies. The "rules" in insurance, healthcare, and legal are often ambiguous, jurisdiction-specific, and constantly changing.
- The article assumes AI accuracy is "good enough" for autonomous operation. In many of these verticals (healthcare, legal, insurance), the error tolerance is extremely low.
🔴 Regulatory Risk
- Healthcare (HIPAA), legal (unauthorized practice of law), accounting (CPA requirements), insurance (licensing) — each vertical has regulatory frameworks designed around human practitioners.
- An "autopilot" that closes the books may trigger regulatory questions about whether it's practicing accounting without a license.
- Several verticals named (tax, legal, insurance) have regulatory bodies that could slow or block adoption.
🔴 Incumbent Response
- Accenture, Deloitte, WTW, and other services giants are not standing still. They have client relationships, domain expertise, and massive AI budgets.
- The thesis assumes incumbents will be slow to adopt AI. But large services firms are among the fastest enterprise AI adopters — they have the most to gain (or lose).
- Professional services firms (McKinsey, BCG, Big 4) are already building proprietary AI tools internally.
🟡 The Innovator's Dilemma Cuts Both Ways
- Bek argues copilots face an innovator's dilemma in transitioning to autopilot. But autopilots face their own dilemma: they need to be dramatically better than the human-plus-copilot alternative from day one.
- A copilot that makes an accountant 3x more productive is immediately valuable. An autopilot that replaces the accountant needs to handle 100% of the work, not just the 80% that's intelligence.
🟡 Pricing & Buyer Behavior
- Companies may not be willing to pay outcome-based pricing at the levels implied. If AI cuts the cost of accounting from $120K to $12K, will companies pay $60K for an "autopilot" or will they expect the cost savings to flow through?
- Race to the bottom: Multiple autopilots in each vertical will compete on price, potentially compressing margins toward infrastructure costs.
For Founders:
- 🔴 Choose autopilot or copilot — don't straddle. The article is explicit: selling the work means cutting your own customers (professionals) out of doing it. You can't serve both masters.
- 🔴 Start with already-outsourced markets. The wedge is vendor displacement, not organizational transformation. If the work isn't already outsourced, you're fighting two battles: proving AI can do the work AND changing how the company operates.
- 🟡 Intelligence ratio is your selection filter. Map every task in your target profession to intelligence vs. judgement. If < 60% intelligence, it's probably too early for autopilot.
- 🟡 Data compounding is the moat. The autopilot that accumulates the most data about what good judgement looks like wins the convergence. Speed to deployment > perfection.
For Investors:
- 🔴 The TAM math works. $1.4T+ in addressable US services markets across just 10 verticals. Even modest penetration rates justify large outcomes.
- 🔴 Watch for the copilot-to-autopilot transition. Existing copilot companies with distribution and product will try to make this pivot. Some will succeed (Harvey), most will struggle with the innovator's dilemma.
- 🟡 Vertical specificity matters. The winners won't be horizontal AI companies. They'll be deeply vertical — owning the workflow, the domain data, and the customer relationship in a single industry.
- 🟡 Europe is underrated. Bek's article names several European companies (Skalar, Ravical, Lawhive, Tacto). Many of these markets are more fragmented in Europe, creating potentially easier entry points.
For Operators (Services Firms):
- 🔴 This is an existential threat to mid-market services firms. The small brokers, accounting firms, staffing agencies, and IT MSPs are most vulnerable. They lack the resources to build AI and the scale to compete on price.
- 🔴 The aging workforce accelerates displacement. Bek specifically identifies CPA retirement (75% nearing retirement) and claims adjuster attrition. The labor supply problem creates a window for AI adoption even among reluctant buyers.
- 🟡 The judgement layer is the moat for humans. Services professionals should move up-stack into judgement-heavy work: strategic advice, complex negotiations, regulatory interpretation, relationship management. Pure intelligence work is indefensible.
"The next $1T company will be a software company masquerading as a services firm." — Julien Bek, opening line
"Every founder building an AI tool is asking the same question: what happens when the next version of Claude makes my product a feature? They're right to worry. If you sell the tool, you're in a race against the model. But if you sell the work, every improvement in the model makes your service faster, cheaper, and harder to compete with." — Julien Bek, on the existential risk for tool companies
"A company might spend $10K a year for QuickBooks and $120K on an accountant to close the books. The next legendary company will just close the books." — Julien Bek, the core economic argument
"A copilot sells the tool. An autopilot sells the work." — Julien Bek, defining the framework
"For every dollar spent on software, six are spent on services." — Julien Bek, the TAM insight
"The higher the intelligence ratio in any field, the sooner autopilots will win." — Julien Bek, the selection filter
"In 2025, the fastest-growing AI companies were copilots. In 2026, many will try to become autopilots. They have the product and the customer knowledge. But they also face the innovator's dilemma: selling the work means cutting their own customers out of doing it. That's the opening for pure-play autopilots." — Julien Bek, closing argument
"Today's judgement will become tomorrow's intelligence." — Julien Bek, on the convergence
"You are in a run-like-hell business right now. The rails are in place. The opportunity is wide open." — Sequoia AI Ascent 2025 Keynote
"The autopilot thesis is compelling. But for most founders, the harder question isn't what to build, it's whether you can survive long enough to earn the right to sell outcomes." — Ryan Gaines, N6 Finance (critic)
What's right about this thesis: The $1 tool vs. $6 service ratio is real and underappreciated. The SaaS era was built on a $700B TAM. The services disruption is 6x larger. The intelligence/judgement framework is genuinely useful for evaluating which markets are ready. And the outsourcing wedge (vendor displacement instead of organizational transformation) is a smart go-to-market insight.
What's missing: Bek underplays three things: (1) the regulatory surface area in most of these verticals, (2) the liability question (who's responsible when the autopilot gets it wrong?), and (3) the speed at which foundation models are commoditizing the application layer. If Claude 5 can do most of what these autopilots do out of the box, the autopilot companies need to compete on workflow integration and data moats, not AI capability.
The real insight that most commentators missed: The article is as much about Sequoia's deal pipeline as it is about investment philosophy. Bek names 20+ companies across 10 verticals. Several are clearly Sequoia portfolio companies or deal targets (Edra was announced as a Sequoia investment 13 days later). This article is a public signal to founders: if you're building an autopilot in these verticals, Sequoia wants to talk. It's dealflow marketing disguised as thought leadership — which, fittingly, mirrors the thesis: a VC firm masquerading as a publisher.
Bottom line: The thesis is directionally correct and will define the 2026-2028 AI investment cycle. The companies that win will look nothing like today's SaaS companies. But execution risk is extreme — most autopilot startups will fail not because the thesis is wrong, but because selling outcomes is fundamentally harder than selling tools. The best founders will read this article, identify the verticals where intelligence ratios are highest and regulatory risk is lowest, and move with extreme urgency. Sequoia is essentially publishing the map. The question is who can navigate the territory.
- Services: The New Software — Julien Bek, Sequoia Capital (March 5, 2026)
- Sequoia Capital: The next $1T company won't sell software — Guillermo Flor, AI Market Fit (March 7, 2026)
- Sequoia Capital: The Next Trillion-Dollar Company Sells Results, Not Software — Odaily (March 11, 2026)
- Sequoia Capital: The Next Trillion-Dollar Company Won't Sell Software—It'll Sell Outcomes — TechFlow (March 11, 2026)
- Sequoia's Thesis Will Mint Billionaires and Bankrupt Copycats — Linas Beliūnas (March 13, 2026)
- "Services Are the New Software" Building Them Is the Hard Part — Han Heloir Yan, Data Science Collective (March 2026)
- The $10 Trillion AI Revolution — Konstantine Buhler, Sequoia Capital (August 28, 2025)
- AI's Trillion-Dollar Opportunity: Sequoia AI Ascent 2025 Keynote — Sequoia Capital (May 7, 2025)
- Sequoia Is Right That AI Is Coming For Services, But Three Things Don't Get Talked About Enough — Ryan Gaines, N6 Finance (March 13, 2026)
- Sequoia Partner Urges Founders: Sell Services, Not Tools — Blockport (March 11, 2026)
- Julien Bek Profile — Sequoia Capital
- Partnering with Edra: Context for Agents at Scale — Luciana Lixandru, Sequoia Capital (March 18, 2026)