Two weeks ago this publication argued that seat-based pricing was winning the AI agent war, because enterprise buyers wanted predictability and vendors like Salesforce were building unlimited-seat license agreements to give it to them. That argument holds for agents that assist a rep who still occupies a seat. It stops holding the moment the agent occupies the seat itself, and this week’s launch of Fireflies Voice Agents is the clearest evidence yet that the two cases need two different pricing models.
What Just Shipped
Fireflies.ai launched Voice Agents, an AI that runs entire sales discovery, recruiting screening and customer-support calls on its own rather than assisting a human who runs them. The company priced it on a usage-based credit model, with no separate contract and no per-seat fee, available on every existing Fireflies plan. “For years, we’ve helped people capture and make sense of their calls,” said Krish Ramineni, CEO and co-founder of Fireflies.ai. “Voice Agents is the next step. Fireflies can now have the conversation for you, then hand your team back exactly what mattered.” Since launch, the company says Voice Agents have already run more than 40,000 conversations across 2,100-plus organizations.
The Argument
Conversation intelligence has been priced per seat since the category existed: one license per rep, because the product’s job was to observe and improve a human who did the actual work. That logic breaks the instant the AI is the one doing the work. A Voice Agent does not free up a rep’s time the way a better dashboard does, it replaces the call the rep would have made. Billing that by the seat charges a company for headcount it is actively trying to avoid needing, while a usage-based model charges for the thing that is actually growing, call volume, which scales independently of how many reps a company employs. Fireflies choosing credits over seats for this specific feature, while presumably keeping its core notetaking product on existing plans, is not an accident. It is an admission that seat pricing cannot describe a product where the seat is optional.
Advertisement
The Counter-Argument, Stated Fully
The strongest objection is the one this publication made in August about Salesforce’s own agent pricing: enterprise buyers hate unpredictable bills. A per-seat or unlimited-license model lets a CFO forecast next year’s software line item to the dollar. A usage-based credit model ties cost to call volume, which is exactly the variable that is supposed to go up when the product works, meaning success and cost rise together with no ceiling the buyer chose in advance. For a category already fighting tokenomics anxiety across the enterprise, adding another metered line item is a real objection, not a rounding error, and it is why large vendors keep defaulting back to flat, capacity-based agreements whenever they can.
Why the Objection Does Not Hold Here
That objection is real for agents that augment a seat a company was already paying for, because the buyer can compare the agent’s cost to the cost of the human it assists. It falls apart for an agent that replaces a call nobody would have made a human available for in the first place, an after-hours screening interview, a discovery call the moment a lead arrives, a testimonial interview scheduled around a customer’s calendar rather than a rep’s. There is no seat to compare that cost against because no seat was going to do that work. Fireflies’ own numbers make the point: 800-plus hours saved on recruiting screening and, per Laura Passmore, Talent Acquisition at WebFX, “nearly 300 screenings later, about a third of our new hires started with a Voice Agent.” That is capacity the company did not have before, not capacity it is redirecting from an existing rep, and capacity that did not exist before has no per-seat price to be predictable about. The CFO’s forecasting problem is real, but it is a problem for pricing an assistant, not for pricing a replacement.
What It Means for Revenue Leaders
The practical read for a RevOps or enablement leader evaluating any AI-agent vendor this quarter: ask which pricing model the vendor uses for which feature, and treat the answer as a signal, not a footnote. A vendor still charging per seat for a feature that fully executes a task without a human on the line is either overpricing the feature or has not yet admitted the seat is gone. A vendor that has moved a specific feature to usage-based credits, the way Fireflies just did with Voice Agents, is telling the market where it believes the work is actually heading. Watch whether ZoomInfo, Gong and the rest of the conversation-intelligence field follow Fireflies to credits for their own agentic features in the next two quarters. If they do, seat-based pricing was never winning the whole AI agent war, it was only winning the assistant half of it.
Source: Fireflies.ai / GlobeNewswire
