A performance engineer agent identifies and resolves throughput and latency issues across the stack. It profiles code, analyzes benchmarks, and produces concrete changes with measured before-and-after comparisons rather than speculative optimizations.
Performance work requires knowing what to measure and what good looks like for your service. A role-specific prompt captures your SLO thresholds, your profiling toolchain, and the parts of the stack where optimization effort is most valuable.
What this agent owns
- Profile application code to identify hot paths and unnecessary allocations
- Write benchmarks that isolate the performance of specific functions or queries
- Implement targeted optimizations and measure their impact before opening a PR
- Review PRs that touch performance-sensitive paths and flag regressions
- Maintain a performance baseline and alert when new changes exceed regression thresholds
Recommended model: Claude Opus
Performance root cause analysis requires reasoning about complex interactions between layers; Opus produces fewer incorrect diagnoses on ambiguous profiles.
Example tasks
- Profile an API endpoint that is over the 200ms SLO and identify the bottleneck
- Write a benchmark for a serialization function and measure two implementation variants
- Find and eliminate N+1 queries in a list endpoint that scans a join table
- Audit a high-traffic code path for unnecessary memory allocations
# create an agent from this template, then start it
$ fleet agent create --name performance-engineer--vendor claude-code --template <template-name>
$ fleet agent start performance-engineer
Find the exact template name with fleet template list.
