Extreme Networks ships Agent ONE Coworker to all Platform ONE users

Extreme Networks built Agent ONE Coworker after shipping Platform ONE with first-generation generative AI and watching it fall short. CTO Nabil Bukhari said that first wave was basically a chatbot on unconsolidated data: it demoed well but broke in production. A concrete example he gives is that four Extreme teams each have a field called client ID that means something different. Extreme's response is a custom context layer plus a knowledge graph, which it argues any competitor would struggle to replicate.

The design choices flow from that diagnosis. Coworker is ambient, so it runs continuously and nudges engineers rather than waiting for them to open a chat box. The Nudge mechanism sits in a command bar and blinks after a pre-investigation, but informational alerts can be snoozed while severity-one alerts cannot. Extreme also loaded its own validated designs and best practices into the agent, on the argument that a general model asked the same question seven times gives seven different answers. Coworker requires a human in the loop; Operator, the autonomous tier, arrives early next year.

The stakes hinge on whether that measured rollout actually changes how engineers work. Extreme's fiscal 2026 revenue was $1.28 billion, up 13%, and it counts 187 customers spending over $1 million annually, so it can ill afford a misstep. If Agent ONE reduces mean time to context as promised, it may help Extreme stand out against larger vendors without a spending war; if the agent bluffs or interrupts too often, it risks the same trust problem its predecessor had.