Per-seat pricing for AI software made sense in 2023 and 2024 when most products were enhanced productivity tools. In 2026, with autonomous agents doing meaningful work without a logged-in human, per-seat is increasingly broken. The vendor with a 0 per user per month AI assistant has to explain why an enterprise should pay for 5000 seats when the agent does most of its work for the operations team automatically at 2am.

The pricing experiments in the enterprise AI market are now visible enough to talk about.

The four models in market

Per-seat. Still the default for assistant-style products. Holding up where the product is genuinely a tool a human uses. Eroding where the product crosses into autonomous work.

Usage-based. Increasingly common for AI agents. Often expressed as “per task completed” or “per workflow run” rather than raw token consumption. The maths is reasonable for buyers but the predictability is poor, which is creating procurement friction.

Outcome-based. Some startups are pricing on outcomes — per ticket resolved, per dollar collected, per lead qualified. This sounds clean and is operationally messy. Defining the outcome, measuring it, attributing it to the agent versus the rest of the stack — all of this is hard. The companies trying this are mostly in single-purpose domains where the outcome is easy to define.

Hybrid platform. A platform fee plus usage. This is where most enterprise AI vendors are landing in 2026. A floor fee for the platform, a meter for the usage, occasional outcome adjustments for specific high-value workflows.

What buyers are doing

Australian enterprise procurement teams are getting smarter about this fast. The 2026 procurement playbook for AI products includes a usage estimate at three traffic tiers, contractual price protection above a forecast envelope, and a clear definition of what counts as a “task” or “run.” Vendors who do not have answers to these questions are losing deals.

The point that gets missed

Pricing structure matters more than headline price in this market. A 0,000 contract that becomes 00,000 because of usage growth nobody modelled is a worse deal than a 00,000 contract with a clean cap. The enterprise teams who are getting AI procurement right are doing the usage forecasting work up front, not pretending it does not matter.

For vendors, the pricing model is the product now. The interface is increasingly commoditised. The pricing structure is where differentiation actually lives.