FTE Pricing for AI Agents: When to Price Your Agent Like an Employee
TL;DR
When an AI agent can do the job of a human employee end to end, pricing it per token or per API call is misaligned with the value it delivers. FTE pricing charges a monthly or annual seat fee, anchored to what the buyer would otherwise pay a human. Companies including Cognition (Devin), Sierra, and Intercom have adopted variations of this model for autonomous agents. It works when the agent replaces a defined job function, the scope is bounded, and the buyer can audit the displacement. It breaks down for assistive tools, variable-load tasks, and use cases where the buyer cannot measure the replacement value.
The AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.
Why Token Pricing Breaks for Autonomous Agents
Token-based pricing was designed for a world where models assist humans. You pay for the AI's contribution to a task; the human does the rest. When the agent does the whole task, token pricing creates three problems.
Value disconnection
An agent that closes a customer support ticket autonomously delivers the same value whether it uses 500 tokens or 5,000. Charging per token aligns cost to model efficiency, not to business outcome. Buyers who understand this will optimize for fewer tokens, not for more valuable outcomes.
Budget unpredictability
Enterprise buyers cannot forecast token consumption for autonomous agents. Human labor costs are fixed and plannable; token costs are variable and dependent on task complexity. Finance teams resist per-token pricing for headcount-replacement decisions for the same reason they would not accept a human consultant charging by the word.
Value capture failure
If your agent saves a buyer $120,000 per year by replacing a support representative, pricing at $5,000 per year in token costs leaves $115,000 of value on the table. The buyer's willingness to pay is bounded by the replacement value, not by inference cost. Token pricing systematically undercharges when the agent is genuinely valuable.
The Three Pricing Models: Token, Outcome, and FTE
There are now three distinct pricing architectures for AI agents, each suited to a different capability and buyer relationship. Understanding where your agent sits determines which model fits.
Token pricing
Anchor: Inference cost
Best for
Assistive tools, copilot features, variable-load use cases
Breaks when
When the agent replaces a job function and the buyer measures value as labor displacement
Examples
Claude API, OpenAI API, most LLM inference providers
Outcome pricing
Anchor: Completed task (resolved ticket, booked meeting, closed deal)
Best for
High-volume, well-defined tasks with clear completion criteria
Breaks when
When outcomes are hard to attribute, tasks are complex and interdependent, or the agent creates value through accumulated context rather than individual completions
Examples
Sierra (per conversation handled), Intercom Fin (per conversation resolved)
FTE pricing
Anchor: Monthly or annual seat fee, anchored to labor replacement cost
Best for
Fully autonomous workflows where the agent replaces a defined role
Breaks when
When the agent is assistive, the scope varies unpredictably, or the buyer cannot measure the replaced labor cost
Examples
Cognition Devin ($500/month per developer-equivalent), emerging enterprise agent pricing in 2026
How to Structure FTE Pricing: The Four Decisions
FTE pricing is not simply picking a monthly price and calling it a seat. The model requires four decisions that most token-priced products have never had to make.
1. Define the role you are replacing
The FTE price must be anchored to something. The clearest anchor is the loaded cost of the human role your agent replaces. A junior customer support representative in the US runs $55,000 to $70,000 per year fully loaded. An outbound SDR runs $80,000 to $100,000. A junior software engineer runs $110,000 to $150,000. Your agent price should be a fraction of this — typically 20% to 50% of the replaced role's cost — to give the buyer clear ROI without leaving too much value uncaptured.
If you cannot clearly name the role your agent replaces, FTE pricing is probably not the right model. The buyer needs to run the same calculation you did.
2. Define what one seat means
In human hiring, one seat is one person working at their full capacity. For agents, one seat should represent the equivalent throughput. If your support agent handles 200 conversations per day, one seat covers up to 200 conversations per day. Overage is priced separately or additional seats are purchased. This creates a clear unit of capacity the buyer can plan around.
Avoid defining a seat as 'unlimited usage.' Buyers will overprovision relative to actual throughput, which damages the relationship when you need to revisit pricing.
3. Set the measurement period
Monthly pricing is more common than annual for new agent categories. Buyers are still evaluating whether the agent delivers on its capability claims and are reluctant to commit annually until they have three to six months of production data. Annual contracts with monthly commits are a reasonable middle ground: annual commitment discount (10 to 20%), monthly invoice.
Annual upfront deals are possible in year two for enterprise buyers who have already seen production results. Do not push for them in the first contract.
4. Decide what is included vs. metered
Pure FTE pricing bundles all usage into the seat fee. This is the cleanest model and easiest for buyers to plan. The alternative is FTE pricing plus metered overage. Overage adds complexity and billing disputes. The cleaner model is to define a seat's throughput cap clearly and add seats when the buyer exceeds it.
If your inference cost varies dramatically by task type, build the variance into your pricing model at contract time rather than metering it after the fact.
Build Your AI Pricing Strategy in the Masterclass
The AI PM Masterclass covers pricing, monetization, and go-to-market strategy for AI products. Taught live by a Salesforce Sr. Director PM who has priced enterprise AI products at scale.
When FTE Pricing Works and When It Does Not
The fit criteria for FTE pricing are specific. The model is not universally applicable just because your product uses agents. Use this checklist before committing to the model.
FTE pricing works when...
- •The agent can handle a job function end to end without human review of every output
- •There is a human role you can clearly name and whose cost you can benchmark
- •The buyer's budget for this came from headcount, not software
- •The scope of the agent's work is reasonably bounded (e.g., Tier 1 support, outbound prospecting, code review)
- •The buyer can measure whether the agent is performing the job, not just being used
FTE pricing breaks when...
- •The agent is an assistant that augments a human rather than replacing their job
- •Task scope varies dramatically (the agent sometimes handles 10 tasks per day, sometimes 1,000)
- •The buyer has no existing headcount in this function to benchmark against
- •Value is delivered through accumulated learning or context, not individual task completions
- •The buyer's stakeholder is in IT (software budget) not HR or operations (headcount budget)
Enterprise Considerations: What Buyers Will Push On
Enterprise procurement teams are not familiar with FTE pricing for software. They understand per-seat SaaS pricing and they understand headcount. FTE-priced agents sit between these categories, and procurement will probe several questions before signing.
"What happens to our data?"
Buyers treating agents as employees will apply employment-grade data handling expectations: confidentiality, data retention limits, and strict access controls. Your data processing agreement and security documentation need to match this expectation. If your model learns from buyer data, that is a significant concern for most enterprise buyers.
"Who is liable for mistakes?"
A human employee's errors are managed through HR processes. An agent's errors are managed through contract and SLA. Be specific in your contract about what constitutes a failure, what the remediation path is, and what your liability limits are. Vague SLAs will slow procurement or block the deal.
"Can we audit what the agent did?"
Enterprise buyers expect audit logs for headcount-equivalent decisions. Build audit trails into the product: what tasks the agent ran, what decisions it made, what outputs it produced. This is not just compliance; it is the evidence the buyer needs to justify the ROI calculation internally.
"How do we turn it off?"
Unlike a human employee, an agent should be pausable or terminable immediately. Enterprise buyers will want a contractual and technical mechanism to suspend the agent's access without a lengthy offboarding process. Build this into the product and document it in the contract.
"How do we scale up or down?"
FTE pricing implies predictable headcount planning. Buyers expect to be able to add seats before a peak period (e.g., holiday support season) and reduce seats after. Build this flexibility into your subscription model or expect pushback from operations and finance buyers.
Setting the Price: The ROI Anchoring Calculation
The FTE price should land in a range that gives the buyer clear ROI while capturing a reasonable share of the value delivered. Too cheap and you undercharge; too expensive and procurement compares to hiring a human and decides to hire. The calculation is straightforward.
The anchoring formula
1. Find the loaded cost of the replaced role
Junior support rep in US: $60,000 salary + $18,000 benefits/overhead = $78,000 fully loaded annually
2. Set your capture rate
Target 20-40% of replacement value. At 30%: $78,000 × 0.30 = $23,400/year per seat
3. Convert to monthly
$23,400 / 12 = $1,950/month per agent seat
4. Validate against your margin
At $1,950/month, what is your inference cost? If inference runs $200/month per seat, gross margin is ~90%. Healthy. If inference runs $1,500/month, gross margin is ~23%. Unsustainable. Adjust capture rate or reexamine which model you are using.
5. Set anchor against the buyer's alternatives
Compare to: (a) hiring a human at $78,000, (b) the nearest competitive agent product, (c) the buyer's existing software spend in this function. Your price should undercut (a) significantly, match or beat (b), and add value over (c).
The 20-40% capture rate is not arbitrary. Buyers need to show internal ROI to justify the purchase. At 50% capture, the ROI math is too thin for most procurement processes. At 10% capture, you are leaving substantial value on the table and potentially signaling that your product is not confident in its value. The 20-40% range tends to produce deals that close and renew.
Learn to Price and Position AI Products
The AI PM Masterclass covers AI pricing strategy, go-to-market, and enterprise sales in live instructor-led sessions. Taught by a Salesforce Sr. Director PM with direct experience pricing AI products at scale.
Related Articles
Before you go: get the AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.