Is SaaS dead? No. "SaaS is dead" is the wrong diagnosis of a real change. The software half of software as a service is doing fine. What is ending is the habit of selling seats to people who operate the product, and leaving the actual work to them. The companies in trouble sell access. The ones that last will sell the result of operating their product, delivered for each customer.
Most of the writing on this question splits into two camps. One says AI agents will absorb applications and the market will shrink. The other says AI is just better software, so SaaS gets bigger. Both camps argue about the software. Neither looks closely at the second word in the acronym.
SaaS has always promised a service and mostly shipped a tool. The customer bought access, then did the work: setting it up, keeping data clean, onboarding new staff, chasing the things that slipped. That service half was always left to the customer. AI can now do it inside the product, as the customer, and keep doing it. That is the shift, and the companies best placed to make it are the ones that already sell the software.
Why people say SaaS is dead
The phrase is old, but 2026 gave it a market event. On January 30, Anthropic released 11 plugins for Claude Cowork that bundle skills, connectors, commands and sub-agents into role-specific assistants. In the days that followed, Outlook Business reported that a Goldman Sachs basket of US software stocks fell 6% in a single session, that the broader selloff wiped out nearly $285 billion in market value across software, financial services and asset management stocks, and that the S&P North American Software Index had fallen around 15% in January, its steepest monthly fall since October 2008. A Jefferies equity trader, Jeffrey Favuzza, named it the "SaaSpocalypse."
The intellectual version of the argument came earlier. In an interview reported by Outlook Business in January 2025, Microsoft CEO Satya Nadella described business applications as "essentially CRUD (create, read, update and delete) databases with business logic," and said that "in future, this logic will migrate to AI agents." Agents working across many apps would form what he called an "AI tier," and the idea that business applications exist could collapse.
Pricing is the third leg. At HumanX in September, Zendesk CEO Tom Eggemeier told the audience, "Unequivocally: seats are dead," according to The Next Web. He added that companies still on seats in two or three years will struggle. Zendesk moved its AI agents to outcome-based pricing in August 2024.
Put together, the case sounds strong: general agents can operate any product, the logic moves out of the app, and the seat model loses its anchor.
Why the "SaaS is fine" camp is also right
The counterargument has real numbers behind it. Christoph Janz of Point Nine points out that the market cap of public SaaS companies grew from about $690 billion in 2018 to over $1.7 trillion, despite the 2022 correction, and that Marc Benioff's own "End of Software" campaign in 2000 preceded the rise of SaaS rather than ending anything. His line is hard to argue with: "AI is not going to kill software, it is software."
Ethan Ruby makes a practical version of the point in Is SaaS Dead?!?. Even when AI reads the contracts, humans still have to review the exceptions, and that takes an interface, permissions and assignment rules. "AI is changing SaaS, but it's not killing it," he writes.
Both camps are describing something true. Software is not going away. But the argument over whether software survives misses where the value is moving.
The half of SaaS nobody shipped
Consider what a customer actually wants from an operations product, a learning platform, or a compliance tool. They do not want to learn the settings page. They want every location compliant, every new hire onboarded, every certificate renewed before it lapses. The software is how that happens. It is not the thing they wanted.
For twenty years, SaaS sold the means and left the end to the customer. The vendor shipped features. The customer hired people to operate them, or did it themselves between other jobs, or asked support to do it for them. The seat model fit that arrangement exactly: you paid per person because people did the operating.
That is the arrangement under pressure. When an agent can do the operating, the number of people operating the product stops tracking the value the product delivers. Bain's Technology Report 2025 puts it directly: "If an agent replaces a human task, customers will expect to pay based on outcomes, not log-ons." Bain expects routine, rules-based digital tasks to shift from "human plus app" to "AI agent plus application programming interface (API)" within three years, and sums up the change as: "The fundamental shift is to stop charging for access and start charging for work done."
So the honest version of "SaaS is dead" is narrower: selling access to people who do the work is dying. Delivering the work is not.
Who is exposed, and who is not
Not every product faces the same pressure. Bain sorts workflows by how automatable they are and how far AI has already reached them. The most exposed, which it calls battlegrounds, are tasks that are easy to automate and easy for others to copy, such as Tier 1 support, invoice processing and time-entry approvals. Products whose workflows are exposed through APIs can see third-party agents capture the value. Products that hold exclusive data and rules are better placed.
Read that through the service lens and a simpler test falls out:
- Exposed: a product whose value is a screen where a person does a task. If an outside agent can do the task through the API, the screen and the seat both lose their reason to exist.
- Resilient: a product whose value is a result the customer needs to stay true, over time, for every account. Holding that result takes the product, the customer's data and the customer relationship at once.
The first kind sells access. The second kind can sell the job itself.
Why the software company is best placed to deliver the work
The "service as software" idea is usually told as a story about AI-native startups replacing labor. Foundation Capital's essay on the shift sizes it at $4.6 trillion, built from salaries for automatable work and spending on outsourced services, and describes a world where "responsibility for achieving the desired outcome sits with the company selling the service." Its examples are mostly young companies.
The overlooked case is the company that already sells the software. It has three things a newcomer has to build or borrow:
- The product. It knows every action the product can take, including the ones it will never expose through a public API or MCP server.
- The data. It already holds each customer's configuration, history and usage, so it can see when the result slips.
- The customer. It already has the account, the billing relationship and the trust to act on the customer's behalf.
A general assistant reaching in from outside can call the tools a vendor chooses to expose. It does not hold the account context, and it does not carry the responsibility for getting the result right. The vendor does. That is why the "AI tier" Nadella described does not have to sit outside the application. For a lot of work, the most capable place to run it is inside the product, as the customer, operated by the company that built it.
What delivering the work looks like
Pricing changes are the visible part. Sierra describes its model as "Sierra gets paid only when we complete a task for you." Zendesk prices its AI agents on outcomes too. These are outcome prices for a single event, like a conversation resolved.
Most of what a software company's customers need is not a single event. It is a state that has to be held. That is why we describe a service as a job the AI owns, not a task. We wrote about that distinction in Outcome as a Service: A Service Is a Job, Not a Task. A job like "keep every location compliant" has a state you can check, such as 11 of 12 locations compliant, and the AI keeps working it whenever that state slips: sending reminders, enrolling people, looping in a manager, fixing what it can through the product. The reminders and escalations are things it does in service of the job. They are not what the customer buys.
A job is also a cleaner thing to sell than a seat. The customer pays for the result being held, usually as a subscription for the job, and the vendor sets the price. The hard part is not the price list. It is delivering and proving the outcome inside the product, which we covered in Outcome-Based Pricing: Delivering the Outcome Is the Hard Half.
What to do if your product sells seats today
You do not need to tear up your pricing to respond. A practical order of work:
- List the work your customers do by hand in your product. Setup, data cleanup, onboarding new users, renewals, compliance checks. Support tickets that say "can you just do this for me" are the best source.
- Find the jobs, not the tasks. Group the work by the state the customer wants to hold. "Every new hire onboarded" is a job. "Send onboarding email" is a task inside it.
- Check what your product can already do through an API or MCP. The job runs on your product's own actions, as the customer, with their permissions.
- Deliver one job for a few customers before you price it. The price is easier to set once you can show the state being held.
- Keep seats where people still operate the product. Seats are not wrong everywhere. They are wrong where nobody needs to sit in the seat.
The short answer
SaaS is not dead. The tool-only version of it is under real pressure, and seat pricing is losing its anchor wherever agents do the operating. The service half that SaaS always promised is finally deliverable, and the company that owns the product, the data and the customer is the one best placed to deliver it.
FAQ
Is SaaS dead in 2026?
No. Public markets punished software stocks in early 2026, but the products still run the work. What is losing ground is the model of selling seats to people who operate the product, especially for tasks an agent can do through an API.
Will AI agents replace SaaS?
AI agents will replace a lot of the clicking, not the product underneath it. An agent still needs the product's actions, data and permissions to do the work. The open question is who runs the agent: an outside assistant, or the company that built the product.
What replaces per-seat pricing?
Pricing tied to the work done. For single events that can mean per resolution or per task. For ongoing work, a subscription for the job, such as keeping every location compliant, fits better because the customer is paying for a result that stays true.
Your customers do not want to learn your product. They want the work done. Aimdoc connects securely to your product and your systems, then delivers the work your customers need: metered, monitored and cost controlled. See how services work and how to sell them through your own Stripe account, or start free.