Sierra

Sierra builds branded, autonomous customer-service agents for large companies — AI that handles support, billing and account questions across chat, voice, SMS, WhatsApp, email and third-party surfaces, and that can take real actions on connected backend systems rather than only answering questions. Issuing a refund, changing a subscription, rebooking a service: those are the transactions the platform is designed around, and they are what separate it from a retrieval chatbot bolted onto a help centre.

It is enterprise software in every sense — no self-serve signup, no free tier, and a deployment measured in quarters rather than afternoons. The company was co-founded by Bret Taylor (former Salesforce co-CEO, chair of OpenAI's board) and Clay Bavor (previously Google AR/VR and Labs), and it publicly names Rocket Mortgage, Vanguard, SiriusXM, Brex, Wayfair and Sutter Health as customers, with reported penetration of more than 40% of the Fortune 50 and a May 2026 round valuing it around $15.8B on a reported $150M+ ARR. Those funding and ARR figures come from press reporting, not from Sierra's own published documentation.

Two things are worth understanding before you evaluate it. First, Sierra is deliberately model-agnostic: it routes across frontier models from multiple providers and does not commit customers to a named model version, so you get provider-swap resilience but not the ability to pin a specific model release for a specific agent. Second, the company authored the τ-bench and τ²-bench agent benchmarks, and that evaluation-first culture shows up in the product as simulation-based testing and the Agent Data Platform's monitoring — the strongest engineering argument for Sierra over a rival that ships the same feature list. The commercial model follows the same logic: you pay per resolved interaction, which aligns cost with results but makes spend impossible to forecast, and makes the contractual definition of "resolved" the single most important line in the agreement.