How CoreWeave Transformed Customer Support with Cohere North in 90 Days

CoreWeave operates some of the most performance-sensitive infrastructure in the AI industry, serving the world’s top AI labs and global enterprises. Its “Direct to Expert” support philosophy reflects a non-negotiable premise: customers’ time cannot be wasted. An hour of infrastructure downtime for the organizations relying on CoreWeave can carry enormous costs. That expectation puts extraordinary pressure on the customer support engineering (CSE) team to be fast, accurate, and fully informed from the moment a ticket arrives.

The reality before deploying AI was slower. When a new support issue arrived via Slack, CSEs had to manually piece together customer context—region, cluster, system history—before they could even begin routing the ticket. The workflow required jumping between Slack, FreshDesk, and Jira, adding manual steps that created friction and delay. “There were too many barriers,” said Ivy DeWitt, Senior Manager of Customer Support Engineering. “It was clunky and slow—the opposite of what we wanted.” The team needed automation that met them where they already worked, not a new system that demanded behavioral change.

CoreWeave engaged Cohere and selected North, an agentic AI platform deployable via API into existing business tools. Two factors made it the right fit: North could be embedded directly into Slack rather than requiring a separate interface, and it could be privately deployed inside CoreWeave’s own data centers, keeping sensitive customer data within their security perimeter. A two-day on-site workshop with Cohere’s solutions architects mapped the end-to-end support workflow, identified where agents could intervene, and established a 90-day implementation timeline. Change management and internal training were completed in a single business week.

Once deployed, the workflow changed significantly. When a customer submits an issue via Slack, a CSE triggers the North triage bot, which automatically collects key contextual data—region, cluster, and system-specific information—from CoreWeave’s backend records. The CSE reviews and confirms the ticket, and a downstream automation creates a Jira issue, opens a swarming channel, and pulls in the relevant engineering teams based on North’s routing recommendations. A separate North resolution agent then surfaces relevant documentation and historical post-mortems to suggest resolution steps before engineers even begin investigating. The result: CSEs arrive at every issue fully contextualized and ready to act, without ever leaving Slack.

CoreWeave measured concrete improvements within months of go-live. Mean time to resolution dropped from a range of 4–8 days to 2–5 days. Customer satisfaction scores—already a point of pride—held between 4.9 and 5.0. Routing accuracy improved, and the minor customer complaints about support quality that occasionally surfaced before the deployment have not recurred since. For the engineering team, the shift freed bandwidth previously consumed by context-gathering, allowing CSEs to focus on actual problem-solving. CoreWeave is now planning to extend North across critical incident response, signaling that the 90-day customer support transformation is a foundation rather than a finish line.