A procurement manager at a 400-person company opened a software renewal notice last quarter and noticed something off: headcount had been flat for 18 months, but the team had shipped twice as much. Two of the seats she was renewing were being used almost entirely by agents - no human was logging in.
She paid for them anyway. There was no other pricing option.
That scene is playing out everywhere right now. The per-seat SaaS model built a $315 billion industry on one clean assumption: more employees means more seats means more revenue. The logic was elegant - as companies grew, SaaS vendors grew with them, generating predictable recurring revenue that Wall Street loved. That assumption held for two decades. It is now wrong in a specific, measurable way, and the break is showing up in renewal negotiations before it fully shows up in headlines.
Why Per-Seat Pricing Is Breaking, Not Just Bending
The unit economics of per-seat pricing rest on a one-to-one relationship between a human user and value delivered. Traditional SaaS pricing assumed more users equals more value equals more money - the pricing formula is simply cost per seat times number of seats. That formula is brittle the moment you have software that can do work autonomously.
AI agents do not log in. Per-seat pricing was built for a world of human users. AI agents don't consume named-user licences and don't map to headcount - so the model is structurally broken in an AI agent economy. This is not a future scenario. Pure per-seat pricing fell from 21% to 15% of SaaS companies between 2025 and 2026, per Bessemer's tracking, with most vendors moving to hybrid - per-seat plus usage overage - rather than killing per-seat entirely.
The shift is also showing up in how enterprises behave, not just how analysts predict. Multiple SaaS companies reported slowing growth in Q4 2025 earnings - not because AI failed to boost productivity, but precisely because it succeeded too well. Customers are reducing software seats rather than adding them, as AI-enhanced workers accomplish more with fewer licences. The signal that made markets pay attention: in February 2026, $285 billion in market capitalisation evaporated from SaaS stocks in what analysts dubbed the SaaSpocalypse, with per-seat companies hit hardest.
The break is not that agents are replacing SaaS applications. It is that agents are replacing the human users those applications were priced around.
The inference economics behind this are worth understanding, because they explain why the break is structural rather than cyclical. LLM inference costs declined 10x annually - faster than PC compute or dotcom bandwidth - with GPT-4 equivalent performance falling from $20 per million tokens in late 2022 to $0.40 by December 2025. As inference gets cheaper, deploying agents into workflows that used to require human operators becomes economically rational at lower and lower stakes. The shift happened because enterprises moved from experimental chatbots to production-scale agentic deployments - and agentic AI consumes tokens in ways that no traditional budget model anticipated.
The Steelman for Incumbent SaaS
Before writing off traditional software vendors, it is worth taking the strongest version of their case seriously.
When you look at how enterprises are actually deploying AI agents in 2025 and 2026, they are not replacing their systems of record - they are building orchestration layers on top of them. As one former Microsoft manager put it, a "reality check" is occurring among CIOs who realise LLMs lack the deterministic consistency required for critical industries. For use cases such as underwriting, a system that provides a correct answer "six out of ten times" is insufficient - these processes demand 100% consistency.
This is real. Agents hallucinate, loop, and drift. Just 6% of companies fully trust agents to autonomously execute core business processes. The compliance, audit trail, and governance infrastructure inside mature SaaS products took years to build and cannot be recreated in a weekend with an LLM wrapper.
Disruption does not mean destruction for the entire industry. The most likely outcome is selective unbundling, where commoditised point solutions face replacement while differentiated platforms with deep data moats, network effects, and regulatory compliance emerge stronger. A CRM that holds ten years of customer interaction data has a moat. A form-routing tool does not.
So the steelman holds - for some of the stack. The problem is that "some" is doing a lot of work. The SaaS categories with genuine moats are probably a third of the market. The rest are in a more uncomfortable position.
What Is Actually Replacing Per-Seat
Three models are gaining ground, each with different implications for buyers and vendors.
Outcome-based pricing is the most aligned with buyer incentives and the hardest for vendors to run. Outcome-based pricing charges only when the AI delivers a specific business outcome. Intercom charges $0.99 per resolved conversation; HubSpot dropped its Customer Agent pricing to $0.50 per resolved conversation in April 2026. The buyer pays nothing for AI attempts that fail or get escalated to a human - which aligns vendor and buyer incentives because the vendor only gets paid when the AI actually solves the problem.
Consumption or action pricing is what most large incumbents are actually shipping. Salesforce Agentforce charges $2 per conversation or $0.10 per standard action under the Flex Credits model rolled out in late 2025. Microsoft Copilot Studio runs on the same structure - $0.01 per credit or $200 per 25,000 credits pay-as-you-go via Azure. Consumption pricing meters activity, not business value - the vendor gets paid whether the agent worked or not.
Hybrid: base subscription plus variable overage is where most enterprise renewals are actually landing. Hybrid pricing - a base subscription combined with usage and outcome components - is now the industry standard, adopted by 41% of AI vendors per Bessemer Venture Partners' 2026 AI Pricing Playbook, up from 27% in 2025.
Companies using hybrid pricing report 38% higher revenue growth and 38% higher net revenue retention compared to pure subscription firms, with 43% of SaaS companies now using hybrid models, projected to hit 61% by end of 2026.
The pattern emerging from watching real enterprise renewals: Microsoft is collecting seat revenue from customers who don't know the transition is happening, and consumption revenue from customers who do. That is the transition window. Vendors are bridging their old model and their new one simultaneously, and buyers who are not paying attention will fund both sides of it.
What a Procurement Team Should Do Before the Next Renewal
The practical question is not whether per-seat pricing is dying - it clearly is, for most of the stack. It is whether your organisation shapes its next contract or inherits one written against your interests.
Vendors in transition are often looking for design partners - buyers willing to commit on favourable terms in exchange for early access. The framing that works: "The analyst consensus says seat-based pricing is declining. We'd like a contract that reflects that trajectory, with transition terms written in now."
Three questions worth putting to every major SaaS vendor at the next renewal: What happens to my per-seat bill when an agent is the primary user of a seat? How does your pricing change if our headcount stays flat but our agent-driven output doubles? And - specifically for outcome-based offers - how do you define a resolved outcome contractually?
The risk for buyers is that 'resolution' definitions vary between vendors and can be gamed if not contractually specified. Outcome pricing is appealing in principle. It needs a written outcome measurement agreement before you sign, not after.
The broader shift is real and it is faster than most teams have priced into their vendor strategy. Gartner says that "by 2030, at least 40% of enterprise SaaS spend will shift toward usage-, agent-, or outcome-based pricing." Most organisations will not wait until 2030 to feel it. Some are already feeling it at renewal, right now, sitting on contracts written for a world where every user was a person.
That procurement manager who paid for two agent-occupied seats last quarter will probably get a better deal next time. Not because vendors have become generous - but because she will know what question to ask.