Per-seat pricing is breaking — the 2026 evidence on hybrid, usage, and outcome billing

The seat is the wrong unit when an agent does the work

Per-seat pricing was built on one assumption: one human, one seat, predictable consumption. AI agents break that link. A single employee running three agents can produce the output of ten, and a vendor still billing per seat watches revenue shrink at the exact moment the product is delivering the most value.

This is not a hypothetical repricing debate. It is the mechanism behind the 2026 software selloff, and it is forcing a structural change in how B2B software is packaged and billed. The question for every founder and CPO is no longer whether to move off pure seats — it is how far, and how fast, without trading predictable revenue for an unforecastable bill.

What the 2026 adoption data actually shows

Gartner projects that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5% in 2025. That is the supply side. On the demand side, ICONIQ Growth's 2026 State of Go-to-Market report finds 48% of B2B SaaS companies now run hybrid pricing as their primary model — a base subscription plus a metered layer.

The shift is already visible in the products people buy daily. Atlassian announced usage-based pricing effective December 3, 2026: Rovo AI credits at $0.01 each, automation steps at $0.50 per 1,000, and — its first outcome-based meter — $1.00 per fully autonomous customer-service resolution. Intercom bills Fin at $0.99 per resolved ticket. The unit of software value is moving from 'a person who could use it' to 'work that got done.'

• 40% of enterprise apps will embed task-specific agents by end-2026, up from <5% in 2025 (Gartner).
• 48% of B2B SaaS run hybrid pricing as their primary model (ICONIQ Growth, 2026).
• Atlassian moves to usage billing Dec 3, 2026; Intercom Fin charges $0.99 per resolved ticket.

The buyer side is bleeding, and it is measurable

The repricing is not painless for customers, and the data on the buyer side is blunt. Zylo's 2026 SaaS Management Index finds the average organization runs 305 SaaS applications and leaves 46% of licenses unused. AI-native application spend grew 393% year-over-year at large enterprises, and ChatGPT is now the single most-expensed application inside the average organization.

The consequence is governance pressure. Zylo reports 78% of IT leaders hit unexpected charges tied to consumption-based or AI pricing in the past 12 months, and 61% were forced to cut projects because of unplanned SaaS cost increases. Business units now control 81% of SaaS spend while IT directly manages just 15% — visibility is falling exactly as pricing gets more volatile. Vendors who ship an unforecastable bill are handing procurement a reason to cut them.

• 305 SaaS apps per organization on average; 46% of licenses unused (Zylo, 2026).
• AI-native app spend +393% YoY at large enterprises (Zylo).
• 78% of IT leaders hit unexpected consumption/AI charges; 61% cut projects over unplanned costs (Zylo).

The four pricing models, ranked by evidence not novelty

There are four models in play, and they are not equally mature. Per-seat remains the simplest to forecast and still fits products whose value genuinely scales with human users. Hybrid — a base subscription plus a metered layer — is the 2026 default because it keeps revenue predictable while letting value scale with consumption.

Usage-based billing (per token, per credit, per API call, per workflow unit) matches cost to consumption but makes the customer's bill variable, which is precisely what triggers the governance backlash above. Outcome-based pricing (per resolved ticket, per closed deal, per document processed) aligns price most tightly to value, but it requires a clean, auditable definition of 'done' and a way to attribute the result to your software rather than the customer's process. Most teams are not ready for pure outcome billing — and pretending otherwise is how you build a metering dispute instead of a product.

• Per-seat: simplest to forecast; breaks when agents replace seats.
• Hybrid: the 2026 default — predictable base plus metered value.
• Usage: matches cost to consumption; creates an unforecastable customer bill.
• Outcome: tightest value alignment; needs an auditable definition of 'done.'

The honest counter-evidence: do not over-rotate

The hype says agents are replacing software and per-seat is dead this quarter. The data says slower and messier. Gartner separately projects that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. S&P Global and McKinsey put the share of enterprises running an agent in production at roughly 31%.

McKinsey's August 25, 2026 State of AI survey of 1,719 respondents is the sharpest reality check: 80% report individual productivity gains, but only 37% report any enterprise-level EBIT impact — flat versus the prior year. About 6% qualify as true AI high performers. The gap between personal leverage and P&L attribution is the whole story, and it is the reason to treat outcome-based pricing as a destination you earn your way into, not a switch you flip.

• >40% of agentic AI projects at risk of cancellation by end-2027 (Gartner).
• Only ~31% of enterprises run an agent in production (S&P Global / McKinsey).
• 80% see productivity gains; just 37% see EBIT impact, flat YoY (McKinsey, Aug 2026).

The build-vs-buy signal hiding in the pricing shift

One McKinsey finding deserves its own section because it changes what you are actually competing against: 32% of respondents decided not to buy a software product because agentic coding tools let their team build a comparable capability in-house. When a harness plus a model can ship a thin internal tool in days, the SaaS SKU has to earn its margin on something the builder cannot replicate cheaply — proprietary data, distribution, compliance, integration depth, or a workflow the buyer would not want to own.

This is the real pressure behind seat compression. It is not only that agents replace seats; it is that agents lower the cost of the alternative to buying you at all. Pricing that reflects durable, hard-to-replicate value survives. Pricing that reflects 'a feature an agent could rebuild' does not.

A decision framework for your next pricing change

Sequence the work by evidence, not by what a competitor announced. First, meter usage independently of how you charge for it — token counts, seats, and outcomes are different lenses on the same event stream, and you need the instrumentation before you pick a model. Second, price in the unit your customer perceives as value, which for an agent-driven product is usually work completed, not access granted.

Third, make a new tier a configuration change rather than a deploy — teams that treat pricing as infrastructure ship changes monthly instead of annually. Fourth, publish the meter definition and the cap. The buyer-side data above is a warning: an unforecastable bill is a churn event waiting to happen, and the vendors who win 2026 are the ones whose customers can predict next month's invoice within a narrow band.

• Meter usage before you choose how to charge for it.
• Price in the unit the customer perceives as value — usually work done, not access.
• Make tiers a config change, not a deploy.
• Publish the meter definition and a spend cap; forecastability is a feature.

What this means for a B2B software team in 2026

Per-seat is not dead, but it is no longer sufficient on its own for any product where an agent can do the work. The durable position is hybrid: a predictable base that covers platform value, plus a metered layer that scales with consumption, moving toward outcome billing only where you can define and audit the outcome cleanly.

Gartner's longer-range projection — that by 2028, 90% of B2B purchases will be intermediated by AI agents, routing more than $15 trillion in spend through automated exchanges — is the reason to start now. When the buyer is an agent optimizing across price, availability, and terms, your pricing has to be machine-readable and defensible, not buried in a sales call. The teams that reprice deliberately, with instrumentation and a forecastability guarantee, capture margin their competitors do not realize they are leaking.