Turn on a goal, and a Team Agent works on it in the background: watching deploys, catching regressions, and paying down reliability debt. Goals surface findings, and every claim in a finding links to the query, trace, or deploy behind it. You set the scope and how much the agent can do on its own, and anything it can't do alone arrives as a drafted next step for a person. Team Agents are available in early access.

Canvas is where engineers and agents work the same investigation. Queries, evidence, and the agent's reasoning stay visible, so a conclusion is something your team can check. Each teammate works with their own agent, Canvas Connectors bring in source code, support tickets, and internal docs, and custom skills teach agents to follow the same runbooks as your best SREs.

Anomaly Detection learns what normal looks like for each service with enough steady traffic to model, then flags real changes: more errors than usual, or a service that stops reporting. There are no thresholds to set. You get a notification in Slack, and Honeycomb starts an investigation in Canvas, so the "why" is already in progress by the time you open it.