How Sentry Built End-to-End Bug Fixing with Claude Managed Agents

Sentry monitors software at scale, processing billions of events daily to surface the exact context developers need when production breaks: stack traces, profiling data, spans, logs, and metrics. The company serves engineering teams at companies of every size, positioning itself as the definitive debugging platform. As AI capabilities matured, Sentry saw an opportunity to go beyond diagnosis and automate the fix itself.

Before Claude Managed Agents, Sentry’s AI debugging agent Seer already used Claude to analyze telemetry and identify root causes with high accuracy. But that left a gap: developers still had to context-switch into their codebase, plan the implementation, write the code, and open a pull request. That handoff from diagnosis to resolution was where time and momentum drained away. Building a coding agent to close this gap would have required Sentry to construct sandboxing, lifecycle management, and an agent runtime from scratch — a significant detour for a team focused on debugging.

Sentry had already selected Claude for Seer after evaluating multiple models, partly because running Claude through Vertex AI let the company keep data within Google Cloud and avoid adding a new subprocessor. When Claude Managed Agents became available, it provided the secure runtime and lifecycle management Sentry would otherwise have spent months building. A single engineer integrated Managed Agents and shipped the initial version in weeks. The workflow runs automatically: Seer analyzes telemetry to produce a root cause, hands it off to a Claude agent running on Managed Agents, and that agent plans a solution, implements the code change, and opens a pull request ready for developer review.

The impact was immediate. Seer now processes over 1 million root cause analyses per year efficiently, and reviews are provided on more than 600,000 pull requests each month. The developer experience shifted fundamentally — from receiving a diagnosis to receiving a finished PR to review. Junior developers who once needed deep system knowledge to navigate complex issues now get complete proposed fixes. Senior engineers skip hours of context review and validate rather than investigate.

Sentry is building toward a workflow where the most actionable bugs are detected, diagnosed, and fixed automatically, with developers reviewing proposed changes rather than writing them. With Claude handling the agent infrastructure, Sentry’s team can focus on expanding what Seer covers and pushing the boundary of automated debugging rather than maintaining a custom runtime.