For twenty-five years, business software has made the same promise: we will make your people more productive. A CRM helps salespeople track deals. An accounting package helps accountants close the books. The software is a tool, and a person still does the work. Companies pay per seat, and the vendor's job ends when the tool works.
AI changes what software can promise. When a system can draft the contract, resolve the support ticket or prepare the audit file, the vendor can sell the result instead of the tool. Investors call this service as software. It is one of the most discussed ideas in enterprise AI, and it matters to founders and executives on both sides: as buyers deciding what to pay for, and as builders deciding what to sell. This article explains the idea, its origins, how it changes pricing, where it starts, what can go wrong, and how to buy it well.
Key takeaways
- Service as software means a vendor sells completed work, done by AI with human oversight, rather than a tool people operate.
- The target is the services and salaries budget, which is several times larger than software spending.
- Pricing moves from seats to outcomes, so the definition of an outcome becomes the most important line in the contract.
- It starts with work companies already outsource, and it needs clear accountability, approval rules and proof.
What is service as software?
Service as software is a business model in which AI systems perform work that was previously done by people, either in-house or through outsourced service providers, and the vendor sells that work as a product. The customer buys a result: a resolved ticket, a drafted filing, a qualified lead, a completed reconciliation.
Foundation Capital, whose 2024 essay made the phrase widely known, drew the line by who carries responsibility. In software, the firm wrote, "customers are still responsible for using that tool to achieve the desired outcome." In services, "responsibility for achieving the desired outcome sits with the company selling the service." Service as software moves that responsibility to the vendor while delivering it through software.
The model does not remove people. Someone still sets the goals, approves what goes out and handles the cases the system cannot. What changes is who does the routine middle of the work, and what the buyer pays for.
Where did the idea of service as software come from?
Several investors described the shift within about a year of each other.
- August 2023. Benchmark partner Sarah Tavel published AI startups: Sell work, not software, arguing that rather than sell software to improve an end-user's productivity, "founders should consider what it would look like to sell the work itself." Selling tools, she wrote, means "selling a 10% productivity improvement instead of 95%." Her example was EvenUp, which produces legal demand packages for plaintiffs rather than software for lawyers to write them.
- April 2024. The venture firm Foundation Capital published AI leads a service as software paradigm shift, describing "a $4.6 trillion opportunity, since the global services market dwarfs the software market in size." They counted $2.3 trillion in salaries across sales and marketing, software engineering, security and HR, plus spending on outsourced IT and business process services.
- October 2024. Sequoia's Sonya Huang and Pat Grady wrote in Generative AI's Act o1 that "Software companies turn labor into software. That means the addressable market is not the software market," but the far larger services market.
- March 2026. Sequoia's Julien Bek returned to the theme in Services: The New Software, with the subtitle "The next $1T company will be a software company masquerading as a services firm."
Different words, same claim: the money is in the work, not the tool.
Why is service as software happening now?
Three things changed at once. Models became capable of multi-step work such as reading a request, gathering facts, drafting a response and checking it against rules. Costs fell enough that doing a task with AI is cheaper than doing it by hand, even after human review. And the budgets for work are much larger than budgets for tools. Bek's Sequoia essay puts the ratio plainly: "For every dollar spent on software, six are spent on services." It gives the example of a company that might spend $10,000 a year on accounting software and $120,000 on an accountant.
That ratio is the whole argument. A vendor that sells a tool competes for the $10,000. A vendor that sells the finished books competes for the $120,000, as long as it can deliver the result reliably. The evolution of enterprise software concept page places this in the longer history of business software.
SaaS vs service as software
| Dimension | SaaS | Service as software |
|---|---|---|
| What is sold | Access to a tool | Completed work |
| Who does the work | The customer's people | AI, with people approving and handling exceptions |
| Budget it competes for | Software and IT | Salaries and outsourced services |
| Typical pricing | Per seat, per month | Per outcome, per task or a share of savings |
| Who carries the result | The customer | Shared, with more on the vendor |
| What the buyer must define | Who gets a licence | What counts as done, and how it is proved |
| Main risk | Shelfware: paid for, unused | Disputed outcomes and hidden quality problems |
Bek summarises the distinction in one line: "A copilot sells the tool. An autopilot sells the work." Copilots keep a professional responsible for the output. Autopilots sell the result to the buyer directly. Our concept page on from copilots to coworkers describes the same move from the user's side.
How does pricing change? From seats to outcomes
If a vendor sells work, charging per seat stops making sense. Fewer seats may be needed precisely because the product works. Outcome pricing follows, and Foundation Capital's essay argued that "An outcome-based approach aligns the software's cost with the value it actually brings to a business."
Customer support shows how this works in practice. Sequoia's Act o1 essay noted that "Sierra gets paid per resolution." Intercom lists its Fin AI agent at $0.99 per outcome, and defines an outcome as counted when a customer confirms their issue is resolved, when they do not ask for more help after Fin responds, or when Fin completes a workflow, including handoffs, charged once per conversation.
That definition is worth reading slowly, because it shows the central design question. "Did not ask for more help" is a reasonable proxy for resolution in support. It is also a proxy, and a customer who gives up looks the same as one who is satisfied. Every outcome-priced contract rests on a definition like this. The outcome-based pricing article covers the trade-offs for buyers and sellers in depth.
Where does service as software start?
Bek's essay argues that autopilots should start with work companies already outsource. Outsourcing signals three things: the company accepts that the work can be done externally, a budget line already exists, and the buyer already purchases outcomes. Swapping one outside provider for another is far simpler than restructuring an internal team.
MIT NANDA's 2025 study of enterprise AI found something similar from the buyer's side. The organisations that got value from AI acted "like BPO clients, not SaaS customers," demanding customisation and holding vendors accountable to business metrics. Buyers who already know how to manage an outsourcing contract are well prepared for service as software.
What can go wrong with service as software?
- Outcome definitions drift. If "resolved" means "the customer stopped replying," the vendor is paid for silence. Write definitions that reflect the business result, with proof attached.
- Accountability blurs. When the AI sends a wrong price, who is responsible: the vendor, the person who approved it, or the company? Agree this before launch. See who is accountable when an AI agent gets it wrong.
- Quality hides behind volume. Thousands of outcomes a month can conceal a steady rate of small errors. Sample and review regularly.
- Context is missing. A vendor doing one slice of work does not know your goals, your other customers or what was promised elsewhere. Its work can be correct locally and wrong for the company.
- Lock-in grows quietly. When a vendor holds the process, the knowledge of how the work is done can leave with it. Keep records of decisions and outcomes in your own systems.
How to buy service as software: a step-by-step approach
- List the work you pay for today.
Outsourced services and internal tasks with a clear, repeatable output. Note what each costs per unit.
- Pick work with a provable result.
A filed document, a sent quote, a reconciled account. If you cannot say what done looks like, you cannot price it.
- Write the outcome definition yourself.
Include the proof: which record, email or system entry shows the work was completed correctly.
- Set approval rules.
What the vendor's AI may send alone, what a person must approve, and what it must never do.
- Run a measured trial.
Compare cost per completed outcome and error rate with your current baseline. The guide to why enterprise AI pilots stall explains how to make the trial end in a decision.
- Keep the record.
Make sure outcomes, edits and exceptions are stored where your company can see and learn from them.
How to measure service as software
| Measure | What it tells you |
|---|---|
| Cost per completed outcome, vs the old cost | Whether the economics work |
| Outcomes verified by proof | Whether "done" is real |
| Outcomes reopened or disputed | Whether the definition matches the business result |
| Share needing human edits before approval | Whether quality allows lighter review |
| Exceptions handed back to your team | How much work actually moved |
How GeniOS approaches service as software
Service as software needs something above any single vendor: a view of what the company is trying to achieve and what work is pending. GeniOS provides that view. It reads the places where a company's work happens, such as email, calendar, chat and the business tools you already use, read-only to start. It forms the company's goals from its own emails, decks and meetings, and checks every team's pending work against them.
What reaches a person is one situation, what to do, why, one owner and a date. The owner answers Delegate, Done, Later or Not relevant. Delegate hands work to a person or an AI agent, and nothing is sent until a human approves.
An item closes only on proof, such as the email that went out, and GeniOS learns from each outcome. That proof step is what makes delegated work measurable, whoever or whatever does it. It is also the basis of the autonomous company described in the GeniOS manifesto.
Worked example: buying audit paperwork as an outcome
The names, dates and numbers in this example are made up to explain the idea. Vaultex makes safety footwear in Greater Noida. Adidas has scheduled a vendor audit and needs IS 15298 test reports and supporting documents. Bata wants 12,000 pairs of S3 safety shoes at $12 a pair, about $144,000, on 60-day payment terms, with a vendor review on Friday. Reebok samples are running on Line 2, and the steel toe cap supplier has raised prices 8%.
Vaultex currently pays an outside consultant about $1,500 per audit to assemble compliance files, and Anushka in production spends two days per audit chasing documents. The management team compares three options:
| Option | What Vaultex pays for | Who carries the result |
|---|---|---|
| Document software, per seat | A tool Anushka uses to organise files | Anushka |
| Consultant, per audit | A finished file, assembled by a person | The consultant |
| Service as software, per outcome | A complete audit file, assembled by AI, approved by Anushka | Shared: vendor for completeness, Anushka for approval |
They write the outcome definition themselves: an audit file counts as done when every required IS 15298 report is attached, matches the current product list, and the email to Adidas has been sent after Anushka's approval. Silence from Adidas does not count.
The audit is not the only pending work, though. The same week, the toe cap rise means Harsh in sales must re-check the Bata price before Friday, and Line 2's Reebok samples mean the Bata volume needs a slot. A vendor selling audit files sees none of that. A layer that reads across the company raises all three situations, each with one owner, and the audit file becomes one delegated item among them, closing on the sent email. Vaultex pays for the outcome and still keeps the whole picture.
Conclusion
Service as software is the business model that follows from AI doing real work. Vendors sell results instead of tools, compete for salary and services budgets instead of software budgets, and price by outcome instead of seat. For buyers it can mean paying only for work that is done. The catch is that "done" must be defined precisely, proved with evidence and seen in the context of everything else the company is trying to achieve. Start with work you already outsource, write the outcome definition yourself and keep the record. For how the people side changes, read AI employees and the future of AI at work. To see GeniOS plans, visit pricing.
Sources
- Sarah Tavel, AI startups: Sell work, not software, 21 August 2023
- Foundation Capital, AI leads a service as software paradigm shift, April 2024
- Sequoia Capital, Generative AI's Act o1: The Reasoning Era Begins, 9 October 2024
- Sequoia Capital, Services: The New Software, 5 March 2026
- Intercom, Pricing (Fin AI agent), accessed 10 November 2026
- MIT NANDA, The GenAI Divide: State of AI in Business 2025, July 2025