Personalized AI infrastructure
AI Agent Pricing Models: Seat, Task or Managed Retainer Compared
Seat, task and managed retainer pricing for company AI agents, compared on vendor incentives, ownership of memory and permissions, and pre-signing checks.

AI agent pricing usually takes one of three shapes: a fee per person who can use the agent, a fee per completed unit of work, or a managed retainer that pays for ongoing operation, oversight and controlled improvement of a defined role. Each model rewards the vendor for a different behavior and leaves the company owning a different part of the work. For a business owner, the right choice depends less on the headline number than on who will be accountable when the agent's memory, permissions and workflows need to change.
Key takeaways
- Seat pricing scales with people, so it favors broad access but says little about whether the agent completes useful work.
- Task pricing scales with volume, which makes cost traceable, but it can reward more activity rather than better results.
- A managed retainer prices continuing responsibility, so the vendor's incentive is to keep the agent accurate, narrowly permissioned and observable.
- Ownership of memory, tool credentials and logs matters as much as the pricing unit, and it should be written into the agreement either way.
- Start with one role and one success criterion, then choose the pricing unit that matches how that role creates value.
What each pricing model actually buys
Seat pricing charges a recurring fee for each named user or team member who has access to the agent. It is easy to budget because cost moves with headcount, not with activity. The company buys access for people, and whether the agent improves the work itself is a separate question that the contract usually does not answer.
Task pricing charges for each unit of work the agent completes, such as a routed inquiry, a reconciled record or a drafted reply that a person then approves. The unit must be defined with care. A task that the agent retries three times may count once, three times or not at all, and the contract should say which. Task pricing ties cost to output, which finance teams can follow, but it only works when the unit is stable and completed work can be counted without argument.
Managed retainer pricing pays a recurring fee for an operated role. The fee covers monitoring, scheduled evaluations, integration upkeep, permission review, incident response and controlled changes to memory and tools. The company is not buying seats or tasks. It is buying continued responsibility for a role that is expected to behave correctly as products, policies and systems change.
How each model changes the vendor's incentives
Incentives follow the revenue line. Under seat pricing, the vendor earns more by adding users, so the natural pressure is toward wide rollout, even before the first role has been proven. An agent that many people open but that changes little in the workflow still produces revenue, which is why seat-based deployments often look busy while the underlying process stays the same.
Under task pricing, the vendor earns more when the agent does more work. That is healthy when the tasks are valuable and reviewed. It becomes a problem when volume rises for reasons that do not improve results, such as repeated retries, looping tool calls or tasks that a person must correct afterward. The buyer then pays for activity that looks like progress on a dashboard but does not reduce the work waiting for a person.
Under a managed retainer, the vendor's revenue depends on keeping the role stable, accurate and inside its approval boundaries. The incentive is to reduce incidents, make evaluations meaningful and keep permissions narrow. The risk moves elsewhere: a retainer can drift into vague scope, where the buyer pays for attention without a clear success criterion. The agreement has to name what is maintained, what is measured and how often the review happens.
A quick test works for any proposal. Ask what the vendor earns if the agent does nothing harmful and nothing new in a quarter. If the honest answer is that revenue is the same either way, the model is a platform fee, and the company must measure the value itself.
Who owns the work when the contract ends?
Pricing is only half of the decision. The other half is ownership: who holds the memory, the instructions, the tool connections, the permission records and the evaluation results. A company that pays per seat may find its agent's operating context stored inside a vendor's account, which makes switching slow even when the monthly cost is low. A retainer can create the same dependency if the scope never says where artifacts live or in what format they can be exported.
Before signing, list the assets the agent depends on and assign each one an owner. Memory should be exportable in a documented format. Tool credentials should sit in the company's own accounts, with the vendor holding only the access the role requires, which is the principle of least privilege described in the OWASP authorization guidance. Logs should be retained by the company for as long as its own audit and review needs demand. The AI agent pricing models explained guide covers the commercial terms in more depth, and the AI agent statement of work shows how to write these assets into scope.
Which model fits which kind of work?
No single model fits every role. The better question is how the role creates value and how that value can be checked. Use the following decision guide as a starting point.
- Choose seat pricing when the value comes from people asking questions, looking things up or drafting content, and no single action changes a system of record.
- Choose task pricing when each unit of work has a clear, countable outcome that a person can verify quickly without redoing the job.
- Choose a managed retainer when the agent touches memory, several tools and approval steps, and it must stay correct as products, prices or policies change.
- Avoid task pricing for open-ended advisory work, where counting outputs rewards volume over judgment.
- Avoid seat pricing for any role whose success depends on actions in operational systems, because access alone says nothing about outcomes.
Illustrative scenario: a small equipment distributor wants an agent to prepare quote follow-ups for its sales team. Seat pricing looks simple because the team is small and everyone would use the tool. The valuable outcome, however, is a follow-up that matches current stock levels and approved discount rules. Paying for access would leave the real risk in the stock and pricing tools unmanaged. A managed retainer scoped to that one role, with a measure of follow-ups accepted without correction, keeps the vendor's attention on accuracy. This scenario is illustrative and does not describe a real customer.
A step-by-step plan to compare the three models
Run these steps before any pricing conversation. They keep the discussion anchored to the role instead of the vendor's packaging.
1. Name one role and write its purpose in a single sentence. 2. Define the success criterion as something observable, such as how many drafts are accepted without edits, and name the person who checks it. 3. List the memory, tools and permissions the role needs, and mark each action that requires human approval. 4. Estimate the volume of work and the cost of a failed task, so you know whether errors or throughput matter more. The FinOps Foundation's material on unit economics is a useful reference for tying cost to a unit of output. 5. Ask each vendor which unit they would bill, how retries and failed runs are counted, and what happens to the bill when the agent is paused. 6. Ask who owns memory exports, credentials, logs and evaluation results at the end of the term. 7. Choose the model whose incentives match the success criterion, then write the scope, the review cadence and the change process into the agreement.
How Be Human handles this
Be Human operates AI agents as a managed service rather than a per-seat tool. A first engagement defines one useful role, the memory and tools it needs, its approval boundaries and an observable success criterion. Ongoing work can include monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements, all inside the scope the company has approved. The company keeps visibility into the memory, tool connections and permission records, and changes to the agent follow a defined request and approval path. Commercial terms are set in the engagement itself, not in this article.
If you are not sure which task should go first, take the free Be Human AI diagnostic to find the first task worth handing to an agent.
Frequently asked questions
Is seat pricing ever the right choice for a company agent? Yes, when the agent works as a shared assistant whose value comes from people using it for questions, drafting and lookup, and when no single action changes a system of record. Seat pricing is simple to budget and easy to start with. Still, measure whether the agent changes any work, because access alone does not prove value.
Does task pricing reduce risk for a small business? It can, because cost follows output, but only if the unit of work is defined precisely and retried or failed runs are excluded or clearly counted. Without that definition, the bill can rise while results stay flat. Ask for the counting rule in writing before comparing proposals.
What should a managed retainer include to be worth paying for? It should name the role, the monitoring performed, the evaluation cadence, the permission review schedule and the process for controlled changes. It should also state who owns memory exports, credentials and logs. If a retainer cannot describe what is maintained and what is measured, it is closer to an open-ended availability fee.
Which pricing model should I ask about first? Ask about the one that fits your first role, not the one a vendor leads with. Write the role, the success criterion and the approval boundary first, then ask each vendor how they would charge for that exact scope. The answer usually reveals whether the incentives point toward accuracy or toward volume.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


