Modern Pricing & UBB
March 17, 2026 pricing

AI Agents Are Killing the Seat License (And Nobody Has a Plan B)

Here's the situation. Salesforce stock is down more than 27% this year. Not because the business is broken — revenue hit $11.2 billion last quarter and Agentforce closed 29,000 deals. The stock tanked because investors looked at agentic AI and realized the per-seat license model that prints money today might not survive the decade.

They're right to panic.

The Math That Broke the Model

The seat license works on two assumptions: every user consumes roughly the same resources, and adding a user costs nearly nothing. Both assumptions are now lies.

AI agents don't need seats. They need compute. An agent handling a sales workflow that used to require a human logging into Salesforce doesn't need a $300/month Enterprise license. It needs API calls and inference. OpenAI's Frontier platform — launched last month with Uber, State Farm, and Intuit as early customers — is designed to make agents first-class employees with their own identities, permissions, and performance reviews. One early deployment freed up 90% of salesperson time on admin tasks. That's not a productivity gain. That's a seat elimination event.

Meanwhile, IT services firms are getting hammered from the other direction. Agentic AI is crushing blended billing rates across contracts in both the US and India. Clients demanded productivity-linked discounts that cut total contract value by 20–35% in 2025. In 2026, that pressure is accelerating as tools like Claude Cowork and OpenAI's agents make developer and engineer roles the first to get repriced.

The Replacement Models Are Still in Beta

Everyone knows seats are dying. Nobody agrees on what comes next.

Salesforce introduced the "Agentic Enterprise License Agreement" — a fixed-price, all-you-can-eat model for Agentforce. Translation: please stop asking us about per-agent pricing, here's a buffet. ServiceNow moved to consumption-based pricing for AI agent offerings. Microsoft bundled agent governance into M365 E7 at $99/user/month, seeding it as the default control plane starting April 1st.

Three different companies. Three completely different pricing architectures. That's not a market converging on a solution. That's a market throwing spaghetti at the wall while the dining room is on fire.

The Numbers That Should Scare You

Zylo's data shows AI-native spend up 393% year over year, with 61% of IT leaders cancelling existing projects just to cover the AI overruns. That's not adoption. That's a budget reallocation panic. Companies are cannibalizing their own SaaS stack to fund AI experiments, and the seat-priced vendors are the ones getting eaten first.

By 2025, 85% of SaaS leaders had adopted usage-based or hybrid pricing models. The holdouts aren't principled — they're just slow. AI products run at roughly 52% gross margins versus the 75–85% that traditional SaaS enjoys. A single power user can consume 100x more resources than a light user while paying the same flat fee. That's not a pricing model. That's a charity.

So What Now?

If you're still running pure seat-based pricing in 2026, you're not being conservative. You're being negligent. The market is bifurcating into two camps: vendors who can meter, cap downside, and connect spend to outcomes — and vendors who can't. The first group is pulling away. The second group is becoming acquisition targets.

The seat license had a great run. Forty years of enterprise software built on counting humans in chairs. But AI agents don't sit in chairs. They don't need onboarding. They don't take PTO. And they definitely don't need a $300/month Salesforce seat.

Figure out your metering story. Yesterday.

Sources

← Pricing Strategy · All posts