Bret Taylor and Clay Bavor lead Sierra
Bret Taylor and Clay Bavor lead Sierra, an AI startup founded in 2023. Taylor, the OpenAI board chair and former Salesforce co-CEO, founded Quip and served as Facebook’s CTO. He also held a board position at Twitter. Bavor, a veteran of 18 years at Google, managed Google Workspace and Google Lens. Sierra reached a $4.5 billion valuation in 2024 and hit $10 billion in September 2025. A recent $950 million funding round led by Greenoaks Capital and Google’s GV pushed the valuation to $15.8 billion. This capital brings Sierra’s total funding to $285 million. The company reported annual recurring revenue exceeding $20 million in November 2025. Current clients include SiriusXM, Sonos, WeightWatchers, and Rocket Mortgage. These brands use the platform to manage customer engagement. The expansion of Sierra into the enterprise market follows Taylor’s experience in building customer engagement technology at Salesforce. Bavor brings expertise in large-scale product development from his years at Google. The rise of agentic organizations involves using AI agents to automate complex knowledge work like researching, reasoning, and transacting.
Agent OS manages the lifecycle of AI agents
Sierra uses Agent OS to manage the lifecycle of an AI agent. The platform uses an adaptive routing client to send API requests to a constellation of models from providers like OpenAI, Anthropic, and Meta. Sierra’s adaptive routing client dynamically sends API requests to a constellation of models from providers like OpenAI, Anthropic, and Meta to reduce latency by 70% during provider outages. Developers use the Agent SDK to define specific goals and deterministic guardrails. For example, an agent might receive a goal to help a customer return an order while following a guardrail that restricts returns to 30 days. Sierra’s agents integrate with enterprise systems to perform actions like processing returns, managing insurance claims, or handling mortgage origination. These agents communicate in 59 languages via voice, chat, WhatsApp, Apple Business Chat, SMS, email, and a ChatGPT app. The technology allows agents to see and interpret a company’s internal business data to address customer issues. Sierra employs a multi-model approach to maintain reliability. Because no single large language model guarantees 100% accuracy, Sierra uses secondary models to monitor the primary model’s performance and detect errors. This architecture aims to mitigate hallucinations where an AI might invent facts or citations. Sierra holds FedRAMP High and PCI DSS Level 1 certifications. This setup allows brands to create agents with specific tones, such as the youthful and sarcastic "Duncan Smothers" used by Chubbies Inc.
Pricing and the competitive landscape
Sierra uses outcome-based pricing where the customer and the company agree on what a completed task means. It does not publish a standard price list. I find the lack of a free trial or self-serve signup a massive hurdle for teams wanting to test capabilities quickly. You should demand a clear definition of a completed task before signing. Companies like Macha and Robylon offer different structures. Macha, an independent layer for helpdesks, charges $299 for 750 tickets, which is about $0.40 per ticket. Robylon uses an AI credit model starting at $25 for 500 credits. Intercom’s Fin offers a $0.99 per resolution rate. Decagon, another enterprise player, reached a $4.5 billion valuation in January 2026. Forethought, which was acquired by Zendesk in March 2026, provides options for teams already using Zendesk. The market organizes into four groups: independent agent layers, enterprise agent platforms, helpdesk-native agents, and ecommerce-specialist tools. The industry distinguishes between rule-based bots, generative AI chatbots, and AI support agents. Rule-based bots follow scripted decision trees, while generative chatbots use language models to interpret natural language. AI support agents go further by retrieving knowledge, maintaining context, and executing configured actions like updating a record or issuing a refund.
| Vendor | Pricing Model | Lowest Published Price |
|---|---|---|
| Macha | Per ticket | $299 (750 tickets) |
| Robylon | Per AI credit | $25 (500 credits) |
| Intercom Fin | Per resolution | $0.99 |
| Sierra | Per outcome | Not published |
Which company will ultimately own the rails of the agentic economy?
