Nobody Agrees How to Charge for AI Agents

AI agents now book travel, write code, and answer customer questions around the clock. But ask who should pay for that work and how, and the industry cannot agree how to Charge for AI Agents. On October 22, 2026, a half-dozen software vendors rolled out their own answers to that question in a single day, according to a roundup of the announcements published by the Agile Brand Guide.

The split matters because agents are moving from demos into daily work. Companies want predictable bills, but nobody agrees on the meter. Some vendors count tokens, some count seats, some count finished tasks, and one startup is paying shoppers instead of billing them. The result is a live experiment in what machine labor is actually worth.

Pay only when the agent finishes the job

ThoughtSpot launched AgentSpot, a workspace for building and sharing AI agents, and said it will charge neither for tokens nor for seats, according to reporting by TechTarget. Customers pay when an agent completes a conversation, a workflow run, or an app session, while building and testing cost nothing. The free tier allows unlimited users and agents with a modest weekly credit allowance, and a paid Fleet tier raises that weekly allowance for heavier use.

An executive quoted in the coverage called unpredictable token billing a nightmare that is slowing adoption, and said the new model is meant to be pricing everyone can understand because it is not linked to any particular AI model. Analysts quoted in the same coverage said outcome-based pricing is gaining momentum among vendors, though they noted open questions about what exactly counts as a completed outcome.

Budgets, cheap models, and the token police

Kochava took the opposite approach with StationOne Enterprise, its AI hub for marketers. The platform routes each AI request to the lowest-cost model judged adequate for the task, lets administrators set token or dollar budgets by user or team, and reports what each user's tasks cost today and on a projected basis, according to the announcement roundup. An audience-management plugin runs deduplication and segmentation in deterministic code so audience rows never pass through a model at all, and an embedded-model option runs compatible models on the user's own machine.

Founder and chief executive Charles Manning said the company is giving enterprises the same financial discipline over AI spending that they already apply to media buying. Kochava published no figures on how much the routing or local execution actually reduces a bill, leaving buyers to measure the savings themselves.

The day's announcements also exposed a gap between marketing teams and engineering teams. Peec AI launched its agent on a monthly credit allowance shared across each customer company, a model aimed at teams that want a simple budget line rather than a meter that ticks with every prompt. Quantum Metric's move to make its model-context-protocol server generally available is about plumbing instead: standardized data connections that let agents reach into more business software without custom wiring, which could decide how fast workplace agents spread.

The contrast captures the industry's identity crisis. Seat-based software pricing assumed humans do the work. Token pricing assumed every model call costs roughly the same. Outcome pricing assumes the agent's finished task is the real product. For shoppers, the most visible experiment came from Disco, which launched Pre-Purchase, a Shopify checkout unit in which an advertiser funds cash back of up to thirty dollars while the merchant earns roughly three dollars and fifty cents per eligible checkout. Sprinklr put its voice-of-customer agent suite into Google's enterprise agent marketplace, and Quantum Metric made its model-context-protocol server generally available, extending the plumbing that lets agents act across apps.

Not everyone buys the outcome story. Critics argue that defining a finished task is itself slippery: does a half-written draft count, and who decides when a workflow truly ran? The fight over how to Charge for AI Agents is really a fight about what an agent's work is worth, and until the industry settles on a meter, buyers will keep doing what Kochava's customers now do — watch the bill in real time and steer work to the cheapest model that gets the job done.

The pricing scramble lands as AI investment keeps climbing across the sector. Related coverage: DeepSeek's fifteen-billion-dollar raise days after a US federal ban, OpenAI's record funding round racing Anthropic's IPO clock, and how Gen Z now powers small businesses in Canada.