The agent pricing reckoning

The four contenders

Per-resolution pricing, which Intercom has pushed hardest, charges for a successful outcome. It aligns vendor and customer incentives more honestly than anything else on the table, and it makes budgeting genuinely difficult. A product incident that triples conversation volume triples the bill in the same month the team is least able to absorb it.

Per-execution pricing, n8n's model, charges once per workflow run regardless of depth. It is the customer-friendly choice and the reason n8n has taken so much share from per-task competitors, but it caps vendor revenue exactly where value is highest — the complex, deep workflows customers would pay far more for.

Credit systems, used by Clay, Relevance AI, YourGPT and most no-code agent builders, are the most common and the least loved. They price genuine underlying variability, which is defensible. They also make it nearly impossible for a buyer to forecast a bill before committing, which is why every review of every credit-priced tool contains the same complaint.

Consumption pricing on tokens or compute is the most honest and the least commercial. It works for infrastructure, where buyers are engineers who can model it. It fails for business software, where the buyer wants a number to put in a budget.

The winning model will not be the fairest. It will be the one a finance team can forecast.

What buyers should do now

Model your expected volume against each shape before comparing headline prices. The same workflow can differ tenfold between a per-task and a per-execution vendor, and the direction of that difference depends entirely on how many steps your workflows have.

Negotiate caps rather than rates. On outcome and consumption pricing, a ceiling that converts overage into a lower marginal rate is worth more than a discount on the base price, because it removes the tail risk that makes these models frightening.

Instrument your own usage independently of the vendor's dashboard. Every disputed bill in this category comes down to a definition — what counts as a resolution, an operation, a credit — and the customer who measured it themselves has a materially better conversation.