Atlassian says redesigned Jira agent tools reduce failures and token use

Atlassian says an internal October evaluation found that its redesigned Jira tools for AI agents completed more tasks and reduced failed tool calls from 0.20 to 0.09 per run. On 62 cases both versions could attempt, the newer version passed 57 and the earlier version passed 54. The server coordinates multiple Jira APIs around jobs such as starting sprints, rather than exposing each endpoint separately. It loads six tools initially and defers another 46; separate internal benchmarking found savings of up to 25% in tokens for the same Jira work. Agents retain the signed-in user's permissions, while deletion and administration require separate scopes. Five shared evaluation cases still failed with the newer tools.