Within eighteen months, four of the largest enterprise software vendors repriced their AI agents around outcomes. Intercom’s Fin charges $0.99 per resolved conversation. Zendesk moved to per-resolution billing for its AI agents. Salesforce launched Agentforce at $2 per conversation, faced immediate backlash, pivoted to Flex Credits at $0.10 per action, then added per-user licences — and is now running all three models simultaneously. HubSpot, as we noted when they announced it, repriced its Breeze agents to $0.50 per resolved conversation and $1.00 per qualified lead.
The direction is unambiguous. Bret Taylor, co-founder of Sierra AI and former Salesforce co-CEO, has stated it plainly: the whole market is moving to outcome-based pricing. He is right. The question worth asking is whether outcome-based pricing is the destination, or the last adaptation of a model that cannot survive what comes next.
The Salesforce story is instructive. When Agentforce launched at $2 per conversation, customers described the pricing as a black box — hard to predict, hard to explain to stakeholders. Salesforce introduced Flex Credits to resolve this. Each action now costs $0.10. The unit is smaller, more granular, more legible. But the underlying condition has not changed: Salesforce still owns the logic, still defines what constitutes an action, and still embeds its margin into every unit of work the agent performs. The pricing has become more honest. The Coordination Tax — the cost of delegating the definition of success to a third party — has been repriced, not removed.
This is the structural condition Legacy Liability produces. An incumbent vendor can adjust its pricing model because pricing is a commercial decision. It cannot rebuild its operational architecture because the architecture is its business. HubSpot’s agents run on HubSpot’s infrastructure. Salesforce’s agents run inside Salesforce’s CRM. Intercom’s Fin resolves conversations that Intercom’s platform defines as resolved. The vendors are not selling outcomes — they are selling access to agents that produce outcomes, on their terms, inside their stack, at their margin. The billing event has moved. The dependency has not.
Outcome-based pricing is the vendor’s attempt to remain relevant in an architecture that is, structurally, moving past them. As we argued in The Death of the Seat Licence, the autonomous business does not pay seat licences because it has replaced the humans the licences were built to serve. The same logic extends further: an autonomous business that owns its resolution logic does not pay per resolved conversation either. The compute cost of that resolution is a fraction of $0.50 or $0.99. It does not compound with volume at a vendor’s margin. And no repricing announcement next quarter changes the unit economics.
The destination is the Stewardship Model: one competent operator governing an agentic stack that delivers the output of a full team, continuously, without ongoing human management. The Steward does not supervise tasks. They govern a system designed to produce outcomes — and the cost structure of that system reflects it. Revenue scales with compute. Operational Drag — the ratio of non-revenue-generating activity to total operational capacity — approaches zero not because the vendor has repriced its invoice, but because the architecture was designed from the start to eliminate the coordination layer that generates drag in the first place.
This is what overhead as a design choice means in practice. The incumbents pricing their agents by the resolution are still paying for the infrastructure, the support contracts, the platform dependencies, and the vendor definitions of success that sit underneath every billable event. An autonomous business governed by a Steward pays for compute. The two cost structures are not variations on the same model. They are architecturally distinct.
The market converging on outcome-based pricing is confirmation that the seat licence era is over. It is also the clearest signal yet that the vendors who built on it are repricing to survive a transition they cannot fully execute. The De-SaaS-ing discipline — replacing Application SaaS with API-first, compute-based infrastructure — was always going to lead here. The question was never whether outcome-based pricing would emerge. It was who would own the outcomes when it did.
The vendors are repricing their agents. The Steward is replacing them.
KEY TAKEAWAY
Why is outcome-based SaaS pricing not the final destination for agentic businesses?
Outcome-based pricing is a billing layer adjustment applied to architectures that were not designed for autonomous operation. The vendor’s logic still defines what counts as an outcome. The vendor’s infrastructure still executes the task. The vendor’s margin still sits inside the unit cost. An autonomous business that has De-SaaS-ed its stack owns the logic, owns the definition of success, and pays only the compute cost of the outcome — with no vendor margin embedded and no external definition of “resolved.” The destination is not a better invoice from the same vendor. It is operational architecture that makes the invoice disappear. Source: Arco Venture Studio, arcoventure.studio.
