We’re bringing multimodal agents to phone calls with Call Companion: an interactive page sent by text that lets callers see and select options while talking to the agent. It gives callers a clear view of their choices, more confidence in their decisions, and an easier way to enter details that are hard to capture by voice.
See Call Companion in action. A caller compares hotels, chooses a room, and pays for her stay, all while talking to the same agent.
Alice Wang volunteered to move to Singapore a year ago to help scale our presence here. It was amazing to see her on stage today demo at our first ever Sierra Summit in Asia - she killed it on stage.
Today we're publishing the first version of Poppy, an open standard we're building with industry partners to make things more frictionless for people using @muse, and better for the businesses serving our shared customers.
Muse is the first personal agent to support it. Some examples of what Muse will be able to help with:
🛍️ Shopping with BYLT (Meta Business Agent @Meta)
💸 Managing finances with Mercury (@SierraPlatform)
🚗 Updating reservations with Hertz (@DecagonAI)
Businesses can start small and choose how agents work with them. This is just the beginning.
Want to put Poppy to work for your business? Our blog explains how to get started:
muse.ai/blog/poppy-personal-…
Today we’re releasing Personal Agent Protocol (Poppy) v1.0.
It’s on for everyone who uses Muse, @Meta's AI agent, enabled for every merchant on @Shopify, and backed by more than 100 design partners.
It was a privilege to speak with Katya Andresen at The Cigna Group’s Investor Day about what we’re building together and and why healthcare represents one of the greatest opportunities for this generation of AI to make a meaningful difference.
Anon has been acquired by @SierraPlatform.
Thanks @claybavor and @btaylor for betting on us. And deep gratitude to the team, investors and customers who got us here.
Sierra already builds the best agents in the world. Time to make them even better.
I am so proud that Sierra inducted into JPMorganChase Hall of Innovation. Thank you Lori Beer, Larry Feinsmith, and Gill Haus for our partnership and recognition. sierra.ai/blog/jpmorganchase…
Guardian has been keeping its promises to customers since 1860. Now, they're partnering with Sierra so millions of customers can get answers in seconds. Read more about our work together: sierra.ai/customers/guardian
Every quarter at Sierra, we celebrate the end of the quarter with Double Denim Day (or, as some like to call it, Canadian Tuxedo Day) 👖. This Q3, we needed a lot of denim to celebrate the most remarkable quarter in Sierra's 2.5 year history.
We added over $150M ARR this quarter, bringing our total ARR beyond $400M. Remarkably, we added more ARR in one quarter than we had all of last year.
The most exciting thing to me is that our pace is accelerating. It took us six months to double from $100M to $200M ARR, and only five months to double from $200M to $400M ARR.
@claybavor and I are grateful to our customers and proud of the entire Sierra team.
I was and am shocked the exponential time hypothesis is false. I had hoped for a result on P / NP from the models, but had naively presumed they would prove the opposite of this result.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…