Salesforce and HubSpot each published controls this month that stop AI agent spending at a limit the customer sets, which moves a question sales teams used to settle in a contract into an admin setting.
What each vendor shipped
On October 27, Salesforce announced Agent Evals, API Evals and advanced cost controls for Agent Fabric, the control plane it sells for managing agents across platforms. Two of the cost pieces are generally available now. The FinOps Dashboard gives budget owners one view of AI spending, and the Model Wallet sets hard limits that stop requests before costs exceed budget. Salesforce also says its Cost Optimization Policies cut token usage by up to 75% in test scenarios, using semantic caching, payload compression and response trimming. That figure comes from Salesforce’s own tests.
The quality side arrives later. Agent Evals and API Evals enter beta in November 2026 and reach general availability in January 2027. The Agent Kill Switch is available now, and the version that triggers automatically from an evaluation result is scheduled for January 2027. Salesforce says that when evaluation results fall short, predefined interventions run, with approval where the team’s policy requires it.
Advertisement
HubSpot’s announcement, updated on October 15, covers the billing side. The account-level spend cap is on by default and limits credit usage to as much as 50% above a customer’s monthly credits. Customers can also set credit limits on specific features and actions, swap seats for credits without waiting for renewal, buy credit packs upfront at a flat monthly rate, or pay as they go for usage beyond their monthly credits. The page gives no effective dates for the controls.
Why the two releases belong together
The two vendors sell different things. Salesforce is selling oversight of agents built on many platforms, and HubSpot is selling a bill that has a ceiling. Both answer the same buyer worry, which Jon Dick, Chief Customer Officer at HubSpot, put this way: “Nobody should have to miss out on getting the most out of AI because they don’t know what the bill will be”.
When an agent bills per action or per token, a bad prompt, a looping workflow or a model swap changes the invoice without anyone approving it. We wrote about that shift earlier in AI Spend Risk Moves From the Contract to the Control Panel. The new releases put a hard limit inside the product on both platforms.
Andrew Comstock, SVP and GM of MuleSoft and Agent Fabric at Salesforce, framed the pairing of cost and accuracy in one sentence: “Enterprises can only scale agents as fast as they can keep them accurate and on budget.” Salesforce says fewer than half of organizations with agents evaluate them on live traffic, and that 96% of enterprises run agents on more than one platform. A sales team that runs a Salesforce agent, a HubSpot agent and an outbound tool from a third vendor therefore faces three separate caps and no shared view unless it builds one.
What the cap does and does not cover
A spend limit stops a bill. It says nothing about whether the agent was right. Salesforce’s own example shows the gap. In the scenario it describes, a retailer switches an order-support agent to a lower-cost model, and Agent Evals later detects that the agent is giving inaccurate information about delayed orders. A cost control would have approved that change as a saving. The evaluation catches the damage. For a sales team, the equivalent is an outbound agent that costs less per email after a model swap and starts quoting the wrong plan to prospects.
That timing matters this quarter. The Model Wallet and the HubSpot cap exist today, and the evaluation features that test agent answers are in beta from November. Sales teams that turn on the caps now have budget protection today, while Salesforce’s evaluation features arrive in beta in November and reach general availability in January 2027. HubSpot’s page does not describe an equivalent evaluation feature.
What it means for the sales leader
In our reading, the cap belongs to whoever owns the revenue forecast, because agent usage is now a line that moves with pipeline activity. Three actions follow from the releases.
First, find out what your current default is. HubSpot’s cap is on by default at 50% above monthly credits, which is a real overage allowance for a team on a tight budget, and customers can adjust it at any time. Set the number deliberately rather than inheriting it.
Second, set limits by feature and action where the vendor allows it. HubSpot lets customers limit credit spend for specific features, and a prospecting agent that sends fifty emails a day does not need the same ceiling as a research assistant.
Third, pair every cap with a named person who can switch the agent off. We argued in Name the Owner of Every Sales Agent Before You Write Its Policy that an agent without an owner has nobody to answer when it misbehaves, and a kill switch has the same problem. Salesforce’s Agent Kill Switch is available now, and someone has to be allowed to use it without opening a ticket.
Finally, ask each agent vendor on your list what the product does when the limit is reached. Salesforce says its Model Wallet stops requests before costs exceed budget. A team running a live outbound sequence should know whether a stopped request means a paused campaign, a queue that resumes later or a prospect who never gets the reply.
The caps will not tell a RevOps leader whether an agent is worth its spend. They set a ceiling, and the measurement of pipeline per credit still has to come from the team’s own CRM data.
