A sales agent that makes 500 calls on Tuesday and 5,000 on Wednesday costs the same under a per-seat contract. I would rather the buyer decide in the contract how that gap gets priced.
What the spending data says
Menlo Ventures’ 2026 State of Enterprise AI report found that enterprises reported AI spending at 4.7x last year’s level, against a planned 1.9x. Menlo’s explanation is that agents “work independently, run in parallel, and keep going even after people step away, so spending scales with work rather than headcount.” Menlo also found that 55% of incremental AI dollars came from net-new budgets, so many of these purchases are new lines with no spending history behind them.
Sales software is mostly priced on seats because the seat matched the worker. An agent is a different unit. Salesforce announced on November 4 that Agentforce Coworker has passed 185,000 users and costs $50 per user per month, a price it calls “simple and predictable.” Nooks announced the same day that its agents work toward goals “like pipeline created and deals advanced.” One vendor meters the person who asks for the work. The other describes the work as the product.
Advertisement
My position
Every contract for an agent that acts on its own should carry two terms that a seat license never needed. The first is a ceiling on activity, written as calls, emails or actions per month, with a named person who approves any increase. The second is a measured result with a holdout group: a slice of accounts or reps that the agent does not touch, compared with the slice it does, over a period the contract names.
The holdout is the part that most buyers skip. Lorikeet, which sells customer service agents, says its new Dreamweaver product uses a holdout group to measure impact, and that every message is checked against contact hours, limits and consent. A holdout takes a few weeks to run and gives a revenue leader a number to show the CFO. Nooks says customers typically source 3x more pipeline with its platform than with their previous tools. I would want to see that claim reproduced on my own accounts, with a group of reps left out, before the contract renews.
Sales leaders also need an owner for the number. We argued in If the Vendor Sells the Outcome, RevOps Still Owns the Context that the team holding the data decides whether an outcome claim is real. The same team should hold the holdout design.
The strongest objection
The best case for seats is predictability. A CRO has a budget, a finance partner and a board, and a fixed monthly price per user is easy to plan around. Outcome terms invite arguments about attribution, because a meeting can trace back to the agent’s email or to the rep’s follow-up. A vendor that is paid on pipeline created will count every touch as its own, and a buyer that is paid on closed revenue will count none of them.
That objection is right about outcome-only pricing, and I am not proposing it. A ceiling on activity keeps the bill predictable, which is what Salesforce’s seat price offers. The holdout settles attribution before the invoice arrives, because both sides agreed how to measure it. What I am proposing is a seat price or a usage price with a measurement clause attached, so that the renewal conversation starts from a result.
A second objection is that holdouts cost revenue, since the reps left out of the agent group may lose deals the agent would have helped win. That cost is real and small for most teams. A holdout of 10% of accounts for one quarter is a rounding error next to the price of a three-year commitment to a tool whose benefit was never measured.
What I would do this quarter
Ask every agent vendor for three answers in writing: the monthly ceiling on actions and who raises it, how the bill changes if the agent does ten times the work next quarter, and whether the vendor will support a holdout in the first 90 days, with the result that ends the contract. A vendor with a good product will answer. One that cannot is asking for a seat license on work that has no seat.
Spend limits are arriving in the platforms themselves, as we reported when Salesforce and HubSpot let buyers cap what AI agents can spend. Use them. A cap in the software and a clause in the contract do different jobs, and a sales team that has only one of them is exposed on the other.
Steve Hind, Lorikeet’s co-founder and CEO, said: “Our customers set the goals and the guardrails.” Buyers who write the goal and the guardrail into the contract are doing the same thing from the other side of the table.
