StarSling's CI Harness v2 reworked the agent loop and the surrounding orchestration to turn a bottlenecked pipeline into a high-throughput system that cut end-to-end time by 5.4×. This workshop breaks down the concrete design and instrumentation choices that delivered that improvement so you can apply the same principles to your own CI and agent-run systems.

In this workshop, we use StarSling's CI Harness v2 as a case study. You'll see the agent-loop redesign, the profiling and metrics that revealed the real bottlenecks, and the practical changes—batching, scheduling, caching, and backpressure—that produced the 5.4× speedup. We'll walk through the code and the benchmark scripts so you can reproduce the results.

We'll cover:

• Agent-loop architecture and control flow: how work is fetched, scheduled, and acknowledged.

• Parallelism and batching strategies that raise throughput without increasing flakiness.

• Instrumentation and profiling: the metrics that identified where to invest effort.

• Caching, state management, and avoiding redundant computation.

• Backpressure and failure handling for predictable scaling.

• Reproducing and measuring the 5.4× speedup: benchmarks, test harnesses, and pitfalls to watch for.

Join us for a live walkthrough of the implementation, the performance data, and a short demo of the benchmark harness. Bring questions—you'll be able to ask the engineers who built the system about tradeoffs and how to adapt the approach to your pipeline.