Why Seat-Based SaaS Pricing Fails For AI Agents

Why AI Agents Break the Traditional SaaS Seat Model

Seat-based SaaS pricing assumes that a human logs into software to perform work.

An AI agent breaks that assumption.

It does not simply access a dashboard, click through a workflow and log out at the end of the day. It can analyse information, make decisions, call tools and complete tasks with limited human involvement.

A human seat measures access.

An AI agent creates output.

That difference changes how AI products create value, generate costs and should be billed.

The commercial pressure is no longer theoretical. Gartner estimates that agentic AI could expose as much as $234 billion in enterprise application software spending to disruption by 2030, representing around 20% of enterprise application SaaS spending.

The reason is structural. AI agents can increase the amount of work completed without requiring a corresponding increase in human users.

An agent that takes over work previously divided among three employees may be undervalued if it is sold at the price of one software seat. At the same time, two customers paying for the same number of AI seats may create radically different levels of agent activity and delivery cost.

What Is Seat-Based Pricing?

Seat-based pricing is a SaaS model in which customers pay according to the number of users licensed to access a product.

If a platform charges $50 per user each month and a company purchases 20 seats, its monthly subscription is $1,000.

The model became a SaaS standard because it was simple for both sides.

Customers could estimate software costs based on team size. Vendors could forecast recurring revenue. Sales teams could expand accounts as customers added employees.

The equation was straightforward:

More employees = more seats = more recurring revenue

That worked because the human user was usually the unit of access, activity and value.

AI agents separate those three things.

Why Seat-Based Pricing Worked for Traditional SaaS

Traditional SaaS products were generally built to help people perform work.

A sales representative used a CRM to manage leads. A designer used creative software to produce assets. A support representative used a help desk to resolve customer issues.

The software improved the employee's ability to work, but the employee still carried out the task.

Human usage also had natural limits.

An employee could work only a certain number of hours, manage a limited number of processes and complete a finite number of actions at once. Some users were more active than others, but the variation was often manageable within a recurring subscription.

Seat count therefore created a reasonable connection between:

• The number of people using the product
• The value received by the customer
• The revenue earned by the vendor
• The cost of supporting the account

AI agents weaken that connection because the user buying access may no longer be the entity performing most of the work.

An AI Agent Does Not Log In. It Does the Work

An AI agent can gather information, make decisions and perform tasks autonomously toward a defined goal. ServiceNow describes AI agents as systems that interact with their environment, determine a course of action and carry out tasks.

An agent might:

• Research prospects
• Process documents
• Classify support requests
• Draft communications
• Update records
• Call external systems
• Generate reports
• Execute multi-step workflows
• Continue operating outside normal working hours

In traditional SaaS, a seat represents the person doing the work.

In agentic software, the person may define an objective and review the result while the agent performs much of the activity independently.

Bessemer Venture Partners describes this as a philosophical shift in AI pricing: AI is becoming more like a productive coworker than another software tool because it completes work rather than merely extending human capacity.

Once software begins performing the work, human headcount is no longer a reliable measure of software usage or value.

One AI Agent May Replace Three Human Seats

Consider an AI sales agent that performs work previously divided among three employees.

It researches prospects, prepares account summaries, drafts outreach and updates the CRM after each interaction.

If the vendor charges for that agent at the same rate as one human seat, the customer may receive the output of several workers while paying for one ordinary licence.

That creates a value-capture problem.

Tomasz Tunguz has proposed one possible response: if an AI agent delivers roughly three times the productivity of a human user, the vendor could charge three times the traditional seat price.

The logic is understandable. If the product creates substantially more value, the price should rise accordingly.

But a fixed multiplier does not automatically repair the model.

It assumes the agent's productivity can be represented by one stable number. It also assumes the agent will be used in a reasonably similar way across customers.

One company may deploy the agent for a narrow workflow. Another may connect it to several systems and run it continuously across thousands of records.

Both may purchase one "AI seat," even though the work performed is entirely different.

Tripling the price may capture more revenue. It does not necessarily make the seat an accurate billing unit.

Why Equal AI Seats Can Create 20x Different Agent Usage

The second problem sits on the cost side.

Imagine two customers that each purchase ten AI agent seats.

The first uses its agents to produce occasional summaries and answer a small number of internal questions.

The second uses the same number of agents to:

• Analyse large document libraries
• Run repeated reasoning steps
• Call external applications
• Retrieve information from multiple databases
• Process thousands of records
• Execute workflows throughout the day

In this illustrative scenario, the second customer could generate 20 times more agent usage while paying for the same number of seats.

For the customer, the licence count looks identical.

For the AI vendor, the cost structure is not.

Every model request, retrieval step, tool call and workflow execution can contribute to the cost of serving an account. The exact cost depends on the model, architecture and task, but the underlying mismatch remains:

Seat revenue is fixed while agent usage is variable.

That can turn adoption into a margin problem.

Gartner predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027, with escalating costs, unclear business value and inadequate risk controls among the reasons. Pricing does not cause every failed project, but a model that disconnects fixed revenue from variable agent costs makes the economics harder to control.

In traditional SaaS, heavier usage was usually a positive signal. It suggested stronger adoption and potentially lower churn.

For an AI product, higher usage remains valuable only when the AI billing model captures enough of that activity to support the cost of delivering it.

Otherwise, the most active customers may become the least profitable.

Seat Count No Longer Matches Agent Activity

AI agents also create a classification problem.

Suppose a company has:

• 20 employees
• 10 specialised agents
• 40 automated workflows
• One shared agent used across the organisation

How many seats should it buy?

There is no natural answer because an agent is not equivalent to a human user.

One person may control multiple agents. Several employees may use the same agent. One agent may activate other agents. A workflow may run thousands of times without anyone logging in.

Once activity is separated from individual users, seat count becomes an artificial measure of consumption.

Can AI Agent Pricing Still Work Per Seat?

The strongest counter-argument is that seat pricing does not need to disappear. Vendors can redefine what an AI seat represents and charge more for it.

Some companies are already extending familiar software structures through AI tiers, entitlements and usage allowances.

ServiceNow structures its products across Foundation, Advanced and Prime tiers, with progressively broader access to AI capabilities, agents and governance tools. Underneath those licences, agentic workflows are also measured according to activity. ServiceNow classifies workflows by the number of actions involved, with different Assist consumption levels for small, medium and large workflows.

This preserves the familiar enterprise contract while acknowledging that two agents can create very different workloads.

Salesforce has taken a similarly flexible approach with Agentforce. Its current options include Flex Credits charged according to agent actions, conversation-based pricing and per-user licences for selected employee-facing use cases.

That range is revealing. Even one of the companies most closely associated with seat-based SaaS no longer assumes that every AI workload should be priced through the same unit.

There is still an important distinction:

Repricing the seat is not the same as rethinking the billing model.

A higher-priced AI seat can work when the included usage is relatively predictable and the price reflects both expected value and delivery cost.

It becomes less reliable when:

• Agent activity differs sharply between accounts
• One agent performs several types of work
• Workflows operate without direct human involvement
• Customers increase output without adding users
• Model and infrastructure costs fluctuate
• Value has little connection to employee headcount

Calling something an "AI seat" does not eliminate those differences.

It may simply place variable agent usage inside a more expensive fixed subscription.

Intercom Shows That the Billing Unit Can Be Split

Intercom provides a clear alternative to the AI-seat approach.

Its human support platform continues to use per-seat plans, but Fin can also be purchased for an existing help desk from $0.99 per outcome, with no separate Fin seat cost. Minimum commitments may apply.

Intercom counts an outcome when Fin resolves a customer issue, completes a defined workflow or satisfies another listed outcome condition. Customers are generally charged once per conversation, even when Fin performs multiple actions within it.

Intercom has therefore separated two commercial units:

• Human access to support software
• Work completed by the AI agent

The employee still has a seat.

The agent is priced according to what it accomplishes.

The point is not that outcome pricing is the universal answer. It is that human access and autonomous work no longer have to share the same billing unit.

Most AI Seats Are Still Human Seats in Disguise

Many AI agent pricing models still anchor the product to human headcount.

A company buys a licence for each employee who can access an AI assistant, copilot or agent. AI functionality is added as a premium feature or higher subscription tier.

This can work when the AI remains closely tied to an individual user.

A writing assistant supports a writer. A coding copilot assists a developer. A research assistant helps an analyst find information.

The person remains the primary unit of use, so a per-user licence still has a logical connection to the product.

The model becomes weaker as the software moves from assistance to execution.

A copilot helps a person perform a task.

An agent may perform the task itself.

That shift is critical for SaaS pricing for AI agents because the economic unit changes from access to work.

The more autonomous the product becomes, the harder it is to justify measuring its value only through the number of employees who can open it.

Seat-Based Pricing vs Usage-Based Pricing for AI

No single model captures every part of AI value and cost.

Seat-based pricing hides usage variability.

Pure usage pricing may expose too much of it.

Outcome pricing can align the bill with value, but only when the outcome is clear, measurable and attributable to the agent.

That leaves a gap between:

• Access to the platform
• Agent usage
• Compute consumption
• Work completed
• Business value created

A human seat cannot represent all five layers at once.

Is Seat-Based Pricing Dead for AI Agents?

No.

Seat-based pricing can still make sense for product elements connected to human participation, including:

• User access
• Collaboration
• Permissions
• Administration
• Governance
• Security
• Shared workspaces

It may also remain appropriate for copilots where one AI assistant is clearly attached to one user.

The model fails when a vendor treats the human seat as a complete measure of autonomous agent activity.

Seat pricing can survive, but vendors need to be clear about what the seat represents.

• Is it payment for human access?
• Is it a licence for one deployed agent?
• Is it a subscription containing a defined amount of agent usage?
• Is it a premium based on estimated productivity?

These are different forms of agentic billing, even when each one appears on the pricing page as "per seat."

The Pricing Gap AI Agents Leave Behind

Traditional SaaS had one convenient commercial unit: the human user.

AI agent pricing does not have an equally convenient replacement.

An agent may create more value than one employee, consume more resources than another agent and continue operating independently of human headcount.

That leaves AI companies with a difficult gap.

1. They need predictable recurring revenue, but they also need to account for variable usage.
2. They need to charge for platform access without ignoring compute.
3. They need to capture the value of automated work without turning every agent action into a confusing line item.

Seat-based SaaS pricing cannot close that gap on its own.

Pure usage-based pricing does not close it cleanly either.

The Seat Is Not Dead. It Is Incomplete

Seat-based SaaS pricing assumes that software value begins with a human logging in.

AI agents break that assumption because they do not simply access the software. They perform the work.

An agent that takes over work previously divided among three employees may be undervalued at one traditional seat price. Two customers paying for the same number of AI seats may also generate sharply different levels of activity and delivery cost.

Some vendors are responding by charging more for AI seats or packaging AI into higher product tiers. Salesforce and ServiceNow now combine familiar licences with action, workflow or consumption-based units. Intercom separates human access from outcomes completed by its agent.

These approaches show that the market is already questioning what the billing unit should be.

Repricing the seat is not the same as creating a model that reflects access, consumption and value.

The seat is not dead.

It is incomplete.

The remaining question is how AI companies can fill that gap without giving customers unpredictable bills or absorbing unlimited agent costs themselves.

Frequently Asked Questions