Introducing Decagon Labs. The research team behind the models and infrastructure powering 80% of Decagon’s model traffic. More in the thread. ↓ https://t.co/mAhX8B508y
The Hook
“When we started Decagon, we used the same frontier models as everyone else. We found that they're incredibly powerful, but they're also general purpose. They're not really built specifically for customer service experiences.”
Problem Agitation
“And customer-facing AI really has specific requirements. It needs us to be fast, accurate, consistent, and conversational across every interaction. So we made a decision pretty early on that we should train our own specialized models to complement the frontier models. One model can detect when a user finishes speaking, one for picking the right agent operating procedure, one for catching hallucinations, and so on. Over time, that effort grew into something bigger.”
Product Reveal
“And we're calling that Decagon Labs.”
Feature Teaser
“Decagon Labs is the research and agent orchestration arm of our engineering team. The simple version of what we do is make the platform better. We do that through post-training, new architectures, evaluation methods, all focused on building the best AI agents to power concierge customer experiences. At this point, over 80% of Decagon's traffic runs on our in-house models that achieve state-of-the-art performance on our use cases. We invested deeply in the research stack to get there, including human annotation pipelines, end-to-end evaluation infrastructure, and training platforms. It's still early days, but it's already paid off.”
Call to Action
“What makes Decagon Labs different is the kind of research we ship. The foundation labs are solving broad problems, our focus is narrower and more applied. We're building the best models in the world for giving everyone the concierge customer experience that they deserve. That specialization lets us move faster and dig deeper. Our agents now power millions of customer interactions for some of the largest businesses in the world. And we're sharing some of our work. From post-training techniques to agent orchestration to advanced retrieval systems, everything we publish reflects what's already improving our agents in production. There's a lot of room to push this space forward, and that's what Decagon Labs is here for.”
Synthesizable in Remotion · key components of the applied-ai-lab-announcement template.
Clean sans-serif two-line title fade-and-slide up at lower left with soft opacity ramp.
Use interpolate() on frame to translate translateY from 12px to 0px and opacity from 0 to 1 over 15 frames.
Centered translucent hexagonal border card on deep charcoal gradient with iridescent violet/cyan edge glow.
Render container div with linear-gradient border and CSS box-shadow glow, applying spring() scale from 0.96 to 1.0.
Vector geometric icon and wordmark gently fading in over dark ambient backdrop.
Sequence logo opacity ramp using interpolate() over 20 frames alongside soft radial gradient pulse.
Purpose. Explain why off-the-shelf foundation models fall short in production support.
Execution. Lead researcher interview intercut with whiteboard algorithm design.
Purpose. Formally name and frame the dedicated internal R&D organization.
Execution. Minimalist graphic title card with neon edge refraction.
Purpose. Provide credibility through metric proof points (80% production volume).
Execution. Secondary researcher interview delivering direct, measured delivery.
Purpose. Articulate open research stance and resolve to brand identity.
Execution. Tertiary researcher interview, whiteboard collaboration b-roll, and final brand slate.