The Death of the Seat License: Will AI Agents End SaaS?

Short answer: AI agents probably won't end SaaS, but they are breaking the seat license, the default way software has been priced for two decades. Gartner puts up to $234 billion of enterprise application spending at risk by 2030, and I think a large minority of today's SaaS vendors, and a bigger share of search-driven content sites, will effectively vanish within about five years. That last part is my estimate, not measured data, and I'll keep it labeled that way.

What happens to SaaS when agents do the clicking?

Pull up your last list of software renewals. How many lines are priced per seat? Now ask how many of those seats would exist if an agent did the clicking and a person only reviewed the result.

That's the question I keep coming back to. A seat license works when more people using the product means more value. Agents break that. One person supervising an agent can do the work of several, and the agent never needs a login in the old sense.

Gartner says the same thing in its own words. Agentic systems "deliver outcomes directly, bypassing traditional user experience (UX)-heavy applications," which "breaks the link between user growth and revenue growth for many enterprise software vendors," according to George Brocklehurst, a managing vice president there. Gartner's wording for vendors "defending legacy dashboards and seat-based models" is "existential threat."

Here is where I have to be careful, because the numbers around this topic get sloppy fast.

What the sources say. Gartner says up to $234 billion of enterprise application spending is exposed to what it calls "agentic arbitrage" between now and 2030, roughly 20% of enterprise application SaaS spending by 2030. "Exposed" means at risk of moving. It does not mean the vendors die. Gartner also calls the shift "less an apocalypse and more of a metamorphosis."

In a separate Gartner forecast quoted by Deloitte, "by 2030, 35% of point-product SaaS tools will be replaced by AI agents or absorbed within larger agent ecosystems of major SaaS providers." That counts tools, not companies, and Deloitte itself says full replacement of some enterprise applications will likely take "at least five years or more."

On consolidation, AlixPartners predicted software M&A deal volume would rise 30% to 40% year over year in 2026, and said mid-market laggards "face distressed M&A or shutdown." Note again: that is deal volume, not the share of companies that disappear. And Citizens analyst Pat Walravens told Business Insider he expects two-thirds of today's top SaaS companies not to survive the AI era. His own history lesson is useful, though: when cloud software displaced older vendors, "no one went bankrupt." They got acquired or rolled into something else.

What I couldn't find. I looked for a credible source for the claim that 30% to 40% of traditional SaaS, or up to 50% of content-driven sites, will vanish over five years. I didn't find one. The closest figures are the ones above, and they measure different things.

So here is my own number, clearly as judgment: over roughly five years, I expect something like 30% to 40% of traditional SaaS products to effectively vanish, meaning shut down, get absorbed into a bigger platform, or become irrelevant because an agent does the job. For content-driven sites that live on search traffic, I'd say up to half. This is a thesis I'm arguing, not a statistic anyone measured. Argue with it.

Which software is most at risk?

I think three groups are most exposed.

Point-solution wrappers. Single-purpose tools such as PDF converters, background removers, thin email sequencing and single-metric dashboards. If a general agent can do the job inside a chat window, the standalone product has little left to sell. Walravens offers a blunt test: if your product can be easily vibe-coded with tools from Claude, OpenAI or Cursor, "you're in a really bad spot."

Seat-licensed basic record-keeping apps. Ernesto Spruyt of Tunga describes one kind of SaaS company as "essentially a database with an interface on top", naming simple CRMs, form builders and basic project management. His point is that agents can store, retrieve and present data without a separate product. Spruyt's firm serves SaaS clients, so read it as one informed view, not a measurement.

Search-bait content sites. Here there is real data, though it comes from news publishers, not recipe blogs or affiliate sites, so treat it as a proxy. The Reuters Institute's 2026 trends report found that publishers expect search traffic to fall by more than 40% over three years. Chartbeat data in the same report shows Google organic search traffic to over 2,500 news sites down 33% globally and 38% in the US between November 2024 and November 2025, though the report says "it is not clear how much of this is down to AI overviews." Lifestyle and utility content looked more exposed than hard news.

Pew Research tracked 900 US adults in March 2025 and found they clicked a regular result link in 8% of visits when an AI summary appeared, versus 15% when it didn't. Similarweb data, as reported by Search Engine Roundtable, put the zero-click share of searches at 69% in May 2025, up from 56% a year earlier, while organic traffic to news sites fell from over 2.3 billion visits at its mid-2024 peak to under 1.7 billion.

The other side of the trade is thin. Reuters says Google delivers 500 times as many referrals as ChatGPT from search alone. And Cloudflare's crawl-to-referral numbers swing hard: Business Insider reported about 2,800 crawls per referral for Anthropic in the first week of July 2026, versus 24,700 in the first week of May, and noted Anthropic has disputed the method. I'd read the direction, not the decimals.

What survives, and why?

I see four things that hold up.

1. Pricing tied to usage or outcomes. Vendors are already moving. GitHub announced in April 2026 that Copilot plans would move to usage-based billing on June 1, with credits tied to token use, while keeping base seat prices. Zendesk bills per automated resolution, meaning a request an AI agent resolved without escalation. Salesforce's Agentforce page lists consumption-based Flex Credits ($500 per 100,000 credits) next to a $125 per user per month add-on. That's a hybrid, and I expect hybrids to be common. Gartner, again via Deloitte, forecasts that "by 2030, at least 40% of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing."
2. Data and compliance that can't be copied. Lutz Finger, writing in Forbes (I read the syndicated copy), argues that "SaaS is not dead. The interface is," and that building a tool is not the same as running a platform: someone still has to handle security, compliance and scale.
3. Deterministic systems. Payroll, ledgers and financial reporting can't be approximately right. Spruyt makes this point directly: an AI that is right six times out of ten isn't usable for payroll. AlixPartners likewise expects regulated industries and core infrastructure to hold up better.
4. Products agents can call. Finger notes HubSpot, Salesforce and DocuSign have opened MCP servers. MCP, the Model Context Protocol, is a shared standard for letting AI tools connect to other systems. I've written a practical guide to MCP if you want the mechanics. Finger also warns that an endpoint widens your attack surface.

Here is how I'd compare the two layers. This is my framework, not a measured split.

Sprawl is the other half of this. Stat roundups disagree on how many apps companies run: Vena cites an average of 220 in 2024, down from 371 in 2023, while Technource cites Zylo data of 90 to 120 on average and 473 at 10,000-plus employees. Treat all of it as rough. BetterCloud (which sells SaaS management, so it has a stake) says consolidation and fragmentation coexist. My read: the center of gravity moves from many tools people operate to fewer platforms agents operate.

What should you do about it?

If you're a founder

• Map your revenue by pricing unit. Which lines depend on seat counts, and what happens if an agent does that work?
• Make your product callable. An API or MCP server is cheaper to ship before someone wraps your product without you.
• Own something an agent can't copy. Customer context, regulated data, a system of record. Gartner's advice is to "capture and retain customer-specific knowledge, not just data."
• Price a unit tied to value, with caps. Buyers will fear surprise invoices. GitHub added budget controls for exactly this reason.
• Run the vibe-code test on yourself. Be honest about it.

If you're buying

• Inventory by pricing metric, not just by vendor.
• Ask every vendor two questions: how do agents access this product, and how will you charge me when they do?
• Negotiate caps and alerts on any consumption pricing, and track it like cloud spend.
• Don't rebuild everything. Finger's point holds: building a tool is not running a platform. Keep money, compliance and anything deterministic with vendors who carry that responsibility.
• Look hard at the long tail of single-purpose tools. That's where I'd expect consolidation first.

What I see building this at Soft Pyramid

Soft Pyramid is a software and AI automation company in Frisco, Texas, and we've been in business since 2013. We build software and AI automation, so I have an interest in how this plays out. Weigh my view with that in mind.

These are judgments, not data. I'd expect careful buyers to ask which of their tools exist mainly to move data from one screen to another, and whether an agent with the right access could do that work. Our usual advice is to start where the work is repetitive, the rules are clear and a human can review the result. We'd also say to leave payroll, ledgers and compliance logic alone unless you can test them hard.

FAQ

Will AI replace SaaS? Not the category. Gartner calls this "less an apocalypse and more of a metamorphosis," and says SaaS "will emerge in a different form." My judgment is that many individual products and some vendors won't make the trip.

Which SaaS products are most at risk? Single-purpose tools an agent can do inside a chat, and basic record-keeping apps sold per seat. Gartner's 35% figure for point-product tools by 2030 covers the first group. I'd add search-driven content sites to the list, though the data I found covers news publishers only.

What is seat-based pricing and why is it under pressure? You pay a fixed amount for each named user, usually monthly. It works when more users means more value. Agents deliver outcomes without those users, which, per Gartner, breaks the link between user growth and revenue growth.

What makes software AI-defensible? In my view: data you own that others can't copy, responsibility for money or compliance, deep integrations, and being easy for agents to call. If a competent team could rebuild your product in a weekend with an AI coding tool, you aren't defensible yet.