Sierra’s agents now get paid per originated loan, not per token

Signals Inbox·July 20, 2026·AI Agents

Sierra has launched Horizon, a new class of AI agents that can chase goals like originating a loan or securing medical prior authorization over days, weeks, or even months. The bigger shift is how Sierra gets paid: not for tokens, conversations, or employee seats, but when the agent produces the agreed result.

Q1What did Sierra officially launch?

Sierra officially introduced Horizon on July 16. Horizon lets companies build agents that pursue goals over days, weeks, or months. Instead of answering one question and disappearing, an agent can follow up with a customer, collect missing information, handle several steps, and keep working toward a finished result.

Q2What is actually new here?

Sierra already charged for completed customer-service resolutions. Horizon pushes that model into much bigger workflows. The billable result could now be an originated loan, an approved medical procedure, a completed renewal, or a closed sale. Sierra is moving from getting paid for solving a conversation to getting paid for completing a business process.

Q3How is that different from normal AI pricing?

Most AI providers charge for tokens, API calls, seats, or monthly access. The customer pays even when the model produces nothing useful. Sierra is taking the opposite bet. It absorbs the model usage, retries, follow-ups, and failed attempts, then charges when the agreed outcome happens. That makes the agent look less like software and more like a worker earning a commission.

Q4Why do loans and prior authorizations matter?

Because these are not simple chatbot tasks. A loan can require documents, identity checks, reminders, decisions, and several conversations. Medical prior authorization can involve clinical records, billing codes, insurer rules, missing information, and appeals. These workflows can stretch across weeks and carry financial or medical consequences. If Horizon handles them reliably, Sierra is competing for operational budgets, not just chatbot budgets.

Q5What is the catch?

Everyone must agree on what counts as success. Did the agent truly originate the loan, or did a human adviser close it? Does a submitted authorization count, or only an approved one? What happens when several systems and employees contributed? Outcome pricing sounds clean, but large workflows create messy questions about attribution, quality, compliance, and who carries the cost when the agent fails.

Q6Why does this matter now?

AI models are becoming cheaper and easier to access, so selling raw tokens is getting harder to defend. Sierra is trying to move the value layer upward, from generating words to owning measurable work. If the model holds, AI companies may stop asking how many users opened the software and start asking how many loans, renewals, approvals, and sales their agents completed.