We’re partnering with @Plaid to bring more seamless financial experiences to Decagon customers. Customers can securely connect their bank accounts within a conversation, allowing agents to help resolve failed payments, disputed charges, account issues, and more without sending them elsewhere to authenticate. For longer workflows like claims and applications, Decagon agents can carry context forward over days or weeks and pick up where the customer left off.

We put nine leading models through 1,373 battles to find the best balance of quality, speed, and cost on complex enterprise tasks. In our latest results, GLM-5.3 broke the Pareto frontier by achieving frontier-level quality with incredible efficiency, finishing tasks in less than a third of the time than Opus 5. More in the thread. ↓

Decagon is growing across Belgium, the Netherlands, and Luxembourg. 🇧🇪🇳🇱🇱🇺 We've been working with brands like Rituals, which has 41M loyalty members and generated 2.4B in revenue last year. We launched its AI concierge, Ray, across 19 countries and 15 languages in just two months. Excited to serve more customers in the region!

Generative Analytics are now available in Decagon. Ask Duet questions about your conversation data, generate custom charts, and identify patterns across thousands of conversations.

Decagon is expanding into Brazil with a new office in São Paulo. 🇧🇷 We’re already working with some of Latin America’s most ambitious companies, including @MercadoLibre. With a local presence, we can work even more closely with customers in the region.

Qasar Younis, CEO of @AppliedInt, has been a mentor and friend to @AshwinSreenivas since the earliest days of Decagon. He came to our SF office to share what he’s learned across two startups, Google, YC, and 8+ years building Applied Intuition. Thanks for joining us, @qasar!

Autopilot has also started applying its improvement loop to itself. When its simulation runner repeatedly got stuck, Autopilot diagnosed the issue, repaired its own instructions, and reduced recurrences from roughly 55 per week to nearly zero.

We then tracked what happened to accepted Autopilot changes over time. More than 85% were still fully or partially reflected 30 days later, showing that teams continued using and building on the work Autopilot started.

DuetBench-2 introduced harder tasks for building and revising Agent Operating Procedures. Autopilot passed 81.9% of them. The quality of its generated simulations also increased from 60% to 96% in a single month.