Bret Taylor on Sierra's 400-customer surge and publishing AI agents directly to ChatGPT

Summary

Bret Taylor's Sierra hit 400 enterprise customers, a milestone he marked at the company's first-ever customer conference, Sierra Summit, held November 5th. The growth rate is striking enough that Taylor, a two-time founder and current OpenAI chairman, describes it as the "everything is on fire" variety of startup stress rather than the existential kind.

Publishing Agents Directly to ChatGPT

Sierra's most consequential product announcement is a one-click publishing feature that lets enterprise customers deploy their existing AI agents as native ChatGPT apps, built on the ChatGPT Apps extensibility model OpenAI unveiled at its recent Developer Day. The strategic logic is distribution parity: brands have already invested in first-party agents for their websites and mobile apps, and this eliminates the need to rebuild separately for third-party AI surfaces.

Taylor frames ChatGPT, and eventually Gemini, as the new App Store or search portal, drawing an explicit parallel to how mobile and the early web created entirely new distribution channels for consumer brands. The discovery mechanics inside ChatGPT are still unresolved, and Taylor acknowledges no one yet knows how intent-matching or app-surfacing will mature. His advice to customers is not to wait: "You can't wait until your competitors figure it out first."

The brand tension is real. Companies must decide what customer experience to protect on owned properties versus what to expose on third-party AI platforms, a calculus that mirrors how retailers in the 2010s weighed building their own mobile apps against dependence on Instagram for traffic.

Customer Results and New Products

On stage at Sierra Summit, Rocket Mortgage reported that its Rocket Assist agent, built on Sierra's platform, improved refinancing conversion rates by 4x. Wayfair's CTO and SiriusXM's Chief Product Officer also presented, speaking to impact on returns, warranty claims, and retention.

Sierra launched two additional products at the summit. Live Assist places the consumer-facing agent on call-center agent desktops, creating a unified intelligence layer across human and AI interactions. The Agent Data Platform is designed to push agents beyond transactional customer service toward long-term customer relationships with persistent memory, with an explicit goal of driving sales rather than just deflecting support volume.

The Applied AI vs. Consumer Agent Divide

Taylor draws a sharp distinction between enterprise agents and general-purpose consumer agents. Structured, domain-specific agents, handling mortgage refinancing at Rocket, revenue cycle at R1, legal work at Harvey, or healthcare queries at Cigna, are engineering problems today. General consumer agents that automate individual lives are still science problems, requiring a breadth of generalization that approaches AGI.

He endorses Andrej Karpathy's "decade of agents" framing from a recent interview, but predicts the progress will be highly uneven. His internal term for companies running AI pilots without measurable outcomes is "AI tourism," and Sierra explicitly positions against it.

Workforce and Operational Shifts

At several Sierra customers, contact center managers have already retitled themselves AI architects, shifting from managing human teams to overseeing hybrid human-AI systems. Taylor sees the deeper transformation as the digitization of previously analog channels: A/B testing, real-time analytics, and self-improvement loops are now applicable to phone calls in the same way they have been applied to web pages for two decades.

Model Dependency and the 1997 Analogy

Taylor says current models are "pretty great" for Sierra's use cases, though he flags two weak points: native voice-to-voice models remain prone to hallucination and poor rules adherence, pushing many deployments back to speech-to-text and text-to-speech pipelines. He also describes the agent-building toolchain itself as roughly equivalent to the state of web development in 1997, citing a Wired-era article about banks spending $23 million to add login functionality to websites. The ambition is a future where a seven-person team can build a multi-billion-dollar business on agents, the same way Kylie Jenner built a cosmetics company with seven full-time staff on social commerce infrastructure.