Delivering high-performance customer support

OpenAI's latest customer story puts Decagon front and center, framing the startup as a live example of what happens when large language models get pointed at the unglamorous grind of customer support. Decagon builds AI agents that handle support tickets end to end, not just chatbots that hand things off to a human the moment things get tricky.

The pitch here isn't a demo or a pilot program. It's about production-scale deployment, the kind where a support agent has to resolve real complaints, refunds, and account issues without a person quietly cleaning up after it. OpenAI frames this as a performance story: response times drop, resolution rates climb, and businesses can absorb support volume that would have required hiring sprees a few years ago.

What makes this notable isn't the concept, which has been the promise of AI support for a while now. It's the claim that Decagon has actually gotten the reliability and consistency needed to run without constant human oversight. That's the harder problem. Support work is full of edge cases, weird phrasing, and company-specific policy quirks, and models that sound confident but hallucinate details have burned plenty of brands already.

OpenAI clearly wants this positioned as a flagship use case for its models in enterprise settings, and Decagon becomes the proof point rather than the protagonist. It's a familiar move: let a customer's results do the marketing. Expect more of these paired announcements as OpenAI leans harder into showing, not just telling, that its models can run unsupervised in business-critical roles.