Short answer
What is service as a software and where does it come from?
Short answer
Service as a software means selling the work done with AI instead of the tool to do it: the closed books, not the accounting software.
Julien Bek of Sequoia Capital set out the idea on 5 March 2026.
Sequoia’s article «Services: The New Software» starts from one figure: for every dollar spent on software, six are spent on services.
His example is vivid. A company might pay 10,000 dollars a year for QuickBooks and 120,000 to an accountant to close the books. The big budget is not in the tool, it is in the work.
$6
spent on services for every dollar spent on software
Sequoia, 05-03-2026
×12
what the accountant can cost compared with the accounting software
Sequoia, 05-03-2026
340,000
fewer accountants in the US over five years, while demand has grown
Sequoia, 05-03-2026
75%
of US certified public accountants (CPAs) are nearing retirement
Sequoia, 05-03-2026
Sequoia’s thesis is that the next great tech company will be a software company disguised as a services firm. You will also see it written as service as software or services as software: it is the same concept.
There is a practical reason behind it. If you sell an AI tool, every new model can turn it into a mere feature of another product. If you sell the work, every model improvement makes your delivery cheaper.
Service as a software, AI as a service, agent as a service: are they the same?
Models
Copilot or autopilot: what does each one sell?
Sequoia distinguishes two ways of selling AI, and the difference lies in who answers for the outcome.
«A copilot sells the tool. An autopilot sells the work.»
This is how it looks in practice:
| Criterion | Copilot | Autopilot |
|---|---|---|
| What do you buy? | A tool for your team or your adviser | The finished work |
| Who does the work? | A person, helped by AI | An AI agent, with a person who reviews |
| Who answers for the outcome? | You or your service provider | The autopilot provider |
| How is it charged? | Per user or licence | Per task done or per outcome |
| Example | Software that helps your accountant prepare tax returns | A service that hands you the tax returns filed |
The important nuance: an autopilot does not remove people. It puts them where they add judgement, reviewing and approving, and leaves the rule-based work to AI.
Context
Why now? Intelligence versus judgement
Sequoia splits professional work into two parts. «Intelligence» is what follows rules: reconciling accounts, classifying documents, applying a regulation. «Judgement» is deciding what to do when the rules are not enough.
His argument is that software engineering was the first profession where AI absorbed most of the rule-based part. And the more weight that part has in a trade, the sooner autopilots arrive.
The way in: what is already outsourced
Sequoia’s advice to founders is to start with tasks that companies already contract out. The budget exists, the scope is well defined and switching provider is easy.
His example: Crosby started with non-disclosure agreements (NDAs), a well-bounded document that many companies already send to outside law firms.
| Criterion | Size Sequoia estimates | Why is it a candidate? |
|---|---|---|
| Accounting and audit | $50–80bn outsourced, in the US alone | Lots of rule-based work and a shortage of professionals |
| Tax advisory | $30–35bn | Clear rules and plenty of repetitive work |
| Transactional legal work | $20–25bn | Repetitive documents, such as NDAs |
| Managed IT services | $100bn+ | Standard support and maintenance tasks |
| Procurement and supply chain | $200bn+ | Small suppliers that nobody manages today |
| Recruitment and staffing | $200bn+ | Screening and sourcing candidates at scale |
For a small or mid-sized business, the translation is direct: what you pay your accountant, your agency or your adviser today is exactly where AI-run services will arrive first.
Pricing
Outcome based pricing: how some providers already charge
Outcome based pricing is the natural way for an autopilot to charge: you pay for each task the AI completes, not per user. There are already real examples in customer service:
- 28 Aug 2024
Zendesk announces outcome-based pricing for its AI agents
Only the issues the AI resolves on its own, without human involvement, are charged. It includes a starter usage level at no extra cost and a dashboard to track spending.
- 5 Mar 2026
Sequoia publishes «Services: The New Software»
It names the trend and turns it into an investment thesis: sell the work, not the tool.
- 12 Mar 2026
Intercom switches to charging per Fin outcome
It charges when its Fin agent completes the action it was set up for: resolving a query, handing over a case or qualifying a sales lead. Its published pricing starts at $0.99 per outcome (checked on 26 October 2026).
The small print worth reading
There is a paradox few people mention: with outcome based pricing, the better the AI works, the more you pay each month. Above a certain volume, a flat fee can be cheaper. The calculator in this article shows you where that point is.
Does this mean SaaS is dead? I don’t think so. Software is still the foundation. What changes is the unit being charged for: more and more, the work done. That is why AI pricing models are shifting from seats to tasks.
Impact
What does service as a software mean for your business?
It depends on which side you are on: if you buy services, you gain bargaining power; if you sell them, your competition changes.
If you buy servicesHigh impact
You will be offered more and more AI-run services priced per outcome. Compare them on cost per task done, not on licence price, and ask to see how each outcome is measured.
If you sell servicesHigh impact
Accountants, advisers and agencies will compete with autopilots. The way out is to standardise your delivery with AI agents and sell a measurable outcome, with your judgement as the differentiator.
LeadershipMedium impact
Review which outside services are mostly rule-based work. That is where the savings are, and also the suppliers who are about to change their model.
IT and dataMedium impact
An autopilot needs access to your systems. Ask for minimal permissions, a log of every action and an exit plan if you change provider.
New AI business models will not always mean inventing a product. Many profitable ones will be familiar services, delivered in a different way.
How I apply it
Plan
What should you do now?
This is the plan I suggest, from least to most effort:
This week: list your outside services
Write down what you pay outside the company every month (accounting, support, agency, translation, recruitment) and mark the ones that are mostly rule-based work.
This month: measure what each task costs
Divide what you pay by the tasks you get: cost per invoice booked, per ticket resolved or per candidate screened. Without that figure you cannot compare outcome-priced offers.
This month: ask your suppliers about their AI plans
Ask whether they will use AI, how the price changes and who reviews the work. If you sell services, prepare your answer before your clients ask.
This quarter: pilot one process with human approval
Pick a measurable task and test an agent that proposes while a person approves. To map it before you automate anything, use the MAPA method.
Before you sign: run the full numbers
Compare outcome based pricing with a flat fee using your real volumes. And if your case needs its own agent, I explain what drives the price in how much an AI agent costs.
Tool
Should you pay per outcome or a flat fee?
Enter your volumes and the prices you are offered. The calculator shows which is cheaper this month and from what share of resolved cases the answer flips.
Key points
- Service as a software means selling the work done with AI, not the tool to do it.
- Sequoia backs it with one figure: for every dollar spent on software, six are spent on services.
- A copilot sells the tool; an autopilot sells the work and answers for the outcome.
- Zendesk and Intercom already charge per outcome: read carefully what counts as one and who measures it.
- With outcome based pricing, the better the AI works, the more you pay: work out the break-even point.
FAQ
Frequently asked questions about service as a software
What is service as a software?
Service as a software is the idea of selling the work done with AI instead of selling the tool to do it. Julien Bek of Sequoia Capital popularised it in an article published on 5 March 2026. His argument: for every dollar companies spend on software, they spend six on services, and AI makes it possible to go after that bigger budget with a standardised outcome.
What is the difference between an AI copilot and an AI autopilot?
A copilot sells the tool to a professional, who still does the work and answers for the result. An autopilot sells the outcome directly to the end client: for example, the closed books instead of the accounting software. According to Sequoia, autopilots will win first in the work with the most rules and the least judgement.
What is outcome based pricing in AI?
It means charging for each task the AI completes, not per user or licence. Zendesk announced it for its AI agents on 28 August 2024, and since 12 March 2026 Intercom charges per outcome of its Fin agent, from $0.99 in its published pricing (checked on 26 October 2026). The key is how each outcome is defined and measured.
Is SaaS dead because of AI?
I don’t think so. Software is still the foundation of any AI-run service. What changes is what gets charged for: more and more providers sell the finished work or charge per outcome, rather than only per user. For buyers, that means comparing offers on cost per task done, not on licence price.
Which businesses are most exposed to service as a software?
Sequoia points to services that companies already outsource and that are mostly rule-based: accounting, tax advisory, claims handling, procurement and simple legal documents such as non-disclosure agreements. It also fits accountants, advisers and agencies that can standardise their delivery with AI agents and a person who reviews.
What should I check before buying AI with outcome based pricing?
Four things: what exactly counts as an outcome and who measures it, whether assumed outcomes are charged (when the customer does not reply), whether there is a monthly spending cap and what happens to your data and your process if you change provider. Also run the numbers the other way round: if the AI works very well, you will pay more each month.
References
Sources
Checked and verified on 26 October 2026.
- Intercom. Pricing (checked on 26 October 2026).
- Intercom. Understanding Fin outcomes (effective from 12 March 2026).
- Sequoia Capital (Julien Bek). Services: The New Software (5 March 2026).
- Zendesk. Press release on outcome-based pricing for its AI agents (28 August 2024).
This article was created with the help of AI and reviewed by José Galán. I take great care over every post and every translation, but the odd mistake can still slip through. If you find one, write to me: you will be helping me improve.
Custom AI agents
Want to sell or buy the finished work? Let’s design the agent.
I design custom AI agents that analyse your data, propose the next action and explain why. After launch I support them every month with a report on what they have done. If your case can be solved with a simple automation, I will tell you that too.
See how I design agents