How is AI agent pricing shifting from per-seat to consumption and outcome-based models in 2027?
Published Jun 14, 2026 · Updated Jun 14, 2026
Direct Answer
In 2027, AI agent pricing is shifting away from the predictable per-seat model toward consumption and outcome-based models — per-conversation, per-action credits, and bundled "digital workforce" seats — and Salesforce Agentforce is the clearest case study, now offering roughly six ways to pay across three primary models. The shift matters because an AI agent does work whether or not a human is logged in, so the old per-seat license stops mapping to value.
Salesforce Agentforce started with a Conversations model at $2 per conversation (any interaction in a 24-hour window), then added Flex Credits — prepaid units consumed by agent actions at about $0.10 per action, sold in blocks of 100,000 credits for $1,000 — and, in late 2025, brought back per-user pricing at $125+ per month through an Agentic Enterprise License Agreement where the seat now bundles a "digital workforce." Each model trades predictability against alignment-to-value: per-conversation pricing converts a predictable seat cost into a variable line that scales with how often the agent runs, and the original conversation unit proved hard to budget because conversations branch and linger without reflecting real business value.
For operators, the AI-agent pricing shift is a clean lesson in why pricing must follow value when the unit of work changes — and why consumption models trade budget predictability for value alignment.
1. Why Per-Seat Stopped Working
Agents decouple work from headcount
The per-seat model priced software by the number of human logins, which worked when value scaled with users. An AI agent breaks that link: it does work continuously, without a human seat, so charging per seat either undercharges (one seat runs thousands of actions) or overcharges (seats sit idle).
When the unit of work decouples from headcount, the per-seat unit stops mapping to value.
The search for a new unit
That mismatch pushed vendors to look for a unit that tracks the work the agent actually does — a conversation, an action, an outcome. Salesforce Agentforce ended up with roughly six ways to pay across three primary models, a sign the industry has not settled on one unit and is testing several at once.
2. The Conversations Model and Its Limits
$2 per conversation
Salesforce Agentforce launched with a Conversations model charging $2 per conversation, where a conversation is any interaction within a 24-hour period. It was the first attempt to price by the agent's work rather than by seats — a step toward consumption pricing.
Why it proved hard to budget
The conversation unit proved problematic. Conversations could branch, linger, and fail to reflect meaningful business value, so the bill did not track outcomes. That made spend hard to budget and inhibited large-scale adoption: buyers could not predict the cost, and the cost did not clearly map to value delivered.
The shift from per-user to per-conversation converted a predictable seat cost into a variable line that scaled with usage — the central trade-off of consumption pricing.
3. Flex Credits and Outcome Alignment
Paying per action
To tighten the link to value, Salesforce introduced Flex Credits — prepaid units consumed by agent actions at roughly $0.10 per action, sold in blocks of 100,000 credits for $1,000. Instead of paying for a fuzzy "conversation," buyers pay for the exact actions the agent performs, which aligns price more closely with work done.
Closer to value, still variable
Flex Credits move pricing closer to outcomes and value delivered rather than flat interactions, but they remain a consumption model: the bill still scales with usage, so budget predictability depends on forecasting action volume. Credits make the unit cleaner and more value-aligned without removing the variability that makes consumption pricing harder to plan than a flat seat.
4. The Return of the Seat
Per-user pricing comes back bundled
In late 2025, Salesforce brought back per-user pricing at $125+ per month through an Agentic Enterprise License Agreement, where the seat becomes the primary wrapper again — but the seat now includes a "digital workforce" rather than just human access. The predictable seat returns, repackaged to bundle agent capacity.
Why predictability still sells
The return of the seat shows that budget predictability has real value: many buyers prefer a known monthly number over a variable consumption bill, even if the flat price is a looser fit to usage. The market is converging on a menu — consumption for buyers who want value alignment, bundled seats for buyers who want predictability — rather than one winning model.
5. The RevOps and Pricing Lessons
Price the unit of value, not the seat
The clearest lesson is that pricing must follow the unit of value, and when AI agents decouple work from headcount, the seat stops being that unit. Operators pricing or buying AI should ask what the customer actually values — conversations, actions, or outcomes — and price that, because a unit that does not track value (like the early conversation) is hard to budget and slows adoption.
Consumption trades predictability for alignment
Every model here trades predictability against value alignment: per-conversation and per-action scale with usage but are harder to budget; bundled seats are predictable but a looser fit. Operators choosing a consumption model must invest in usage forecasting and guardrails, because the variable bill that aligns to value also removes the budgeting comfort of a flat seat.
Offer a menu when the market is unsettled
Agentforce's roughly six ways to pay are not indecision — they let different buyers pick the trade-off they want. Operators in an unsettled category should consider offering a menu rather than forcing one model, because predictability-seekers and value-aligners are different buyers, and a menu captures both while the market figures out the unit.
FAQ
How is AI agent pricing changing in 2027? It is shifting from per-seat to consumption and outcome-based models — per-conversation, per-action credits, and bundled "digital workforce" seats — because an AI agent does work without a human login, so the seat no longer maps to value.
What are Agentforce's pricing models? Salesforce Agentforce offers roughly six ways to pay across three primary models: Conversations at $2 per conversation, Flex Credits at about $0.10 per action (100,000 credits for $1,000), and per-user licensing at $125+ per month via an Agentic Enterprise License Agreement that bundles a digital workforce.
Why was the per-conversation model hard to budget? Because conversations could branch, linger, and not reflect meaningful business value, so the bill did not track outcomes. It converted a predictable seat cost into a variable line that scaled with usage, making spend unpredictable and inhibiting adoption.
Why did Salesforce bring back per-user pricing? Because budget predictability still sells. The late-2025 per-user model at $125+ per month gives buyers a known monthly number while bundling agent capacity into the seat — predictability over precise value alignment.
What can operators learn from the AI pricing shift? Price the unit of value, not the seat; understand that consumption trades predictability for alignment; and consider offering a menu of models when the market has not settled on one unit.
Bottom Line
In 2027 AI agent pricing is moving from per-seat to consumption and outcome-based models, and Salesforce Agentforce is the case study with roughly six ways to pay: $2-per-conversation, Flex Credits at about $0.10 per action, and a bundled per-user seat at $125+ per month.
Each trades predictability against value alignment, and the return of the seat shows predictability still sells. For operators, the lessons are exact: price the unit of value, know that consumption trades predictability for alignment, and offer a menu when the market is unsettled.
Sources
- Saksoft — The evolution of Salesforce Agentforce pricing: what changed and what it means
- eesel AI — Salesforce Agentforce setup cost: complete 2026 pricing breakdown
- SaaStr — Salesforce now has 3 pricing models for Agentforce
- Monetizely — The doomed evolution of Salesforce's Agentforce pricing
- eesel AI — Is Salesforce Agentforce worth the cost? A 2026 pricing breakdown
- ekfrazo — Salesforce Agentforce pricing 2026: tiers, Flex Credits and real costs
*AI agent pricing review — Agentforce pricing reviews, rating, AI agent pricing review 2027, and a review of per-conversation, Flex Credits, and per-seat consumption models for RevOps operators.*