Microsoft automates end-to-end bug triage with two AI agents in Azure DevOps, reducing time from identification to resolution
Microsoft built and published an Auto Triage AI solution using Copilot Studio and Azure DevOps, using two autonomous agents to extract bug details from customer emails, generate reproduction steps, create bug records, and post follow-up updates automatically.
Background
Enterprise software support teams face escalating bug report volumes. Manual triaging — reading emails, extracting technical details, cross-referencing documentation, creating DevOps records, and sending follow-up communications — is slow, inconsistent, and scales poorly. Microsoft designed the Auto Triage AI solution to automate this end-to-end process using Copilot Studio agents that interact with Azure DevOps as the system of record.
What Was Implemented
- Agent 1: Autonomous bug report creation — triggered by customer email, uses generative AI to extract issue details, cross-reference documentation, generate reproduction steps, create Azure DevOps bug record, and email tracking number to submitter
- Agent 2: Autonomous bug update and follow-up — triggered by reply email with tracking ID, retrieves bug from Azure DevOps, updates record with new information, sends status update email
- Built on: Microsoft Copilot Studio, Azure DevOps, Power Automate, AI Builder
- Knowledge sources: product specifications, business and technical process flows, installation guides, error code lists, knowledge bases
Results
The Auto Triage AI solution automates the full bug reporting and update cycle, reducing manual burden on support teams and improving consistency and speed of bug record creation. Microsoft's published architecture documentation does not include specific quantified performance metrics for internal deployment; the book's claim of "reduced time from identification to resolution" is consistent with the solution's design but unquantified in the source.
Lessons
- Two-agent architecture (create vs. update) cleanly separates the bug lifecycle stages, enabling each agent to be optimized for its specific task
- Cross-referencing product documentation at bug creation time enriches records automatically, reducing the back-and-forth typically required for reproduction
- Tracking-number-based follow-up removes the "black hole" experience for bug submitters and creates a closed-loop communication pattern
- Published reference architectures (as Microsoft Learn solution ideas) enable enterprises to implement proven patterns without building from scratch