Honeycomb switches to RWX for agent-ready CI/CD

About

Honeycomb is the observability platform for the new shape of software. Built on a decade of distributed tracing leadership and deep roots in the OpenTelemetry community, Honeycomb gives engineering teams—and the AI agents now operating alongside them—real-time, high-cardinality answers about any production system, with no pre-aggregation and no cardinality limits.

Challenge

Honeycomb's CI feedback loop was straining under AI-assisted development

Like many companies operating in the increasingly competitive observability vertical, Honeycomb leans on agentic solutions to support its growing engineering team. However, as internal adoption of these solutions significantly accelerated development velocity, Honeycomb's CI pipeline became strained under a new level of validation demand.

Honeycomb engineering had long used its own SLO product to ensure builds completed in under 10 minutes, but influxes of AI code had pushed P95 build times to over 13 minutes. Those extra minutes slowed pull request (PR) turnaround, delayed preview workflows, and made it harder for developers to remain productive while waiting for CI to validate changes.

Speed was far from the only blocker. Reliability had also become a concern, with a sharp increase in flakes and failures. "Builds were failing for reasons that were entirely out of our control," Dean Strelau, Staff Engineer at Honeycomb, shares. As engineers had to investigate platform-level failures, valuable time was siphoned away from validating product work and moving changes toward merge.

With years of experience optimizing their CI configuration, the team at Honeycomb recognized that they would need more than incremental improvement to address both speed and reliability constraints. The team needed a CI platform that would keep pace with Honeycomb's development velocity.

That's when Honeycomb first discovered RWX. After witnessing firsthand how RWX had built-in OpenTelemetry tracing that aligned precisely with Honeycomb's measurement-based CI/CD approach, without the speed and flakiness shortcomings, Dean had all the conviction needed to move forward.

Solution

An agent-native CI/CD platform built for fast, observable validation

RWX gave Honeycomb a shorter, clearer path from PR to deployment.

During the proof of concept, the efficiency gains showed up immediately. The team pointed Claude at RWX's documentation, gave it access to the CLI, and used that loop to test, adjust, and validate pipeline changes without waiting on a slow manual cycle. "Letting Claude go with the CLI and pointing it at the documentation was fantastic," shares Dean.

The first major unlock was RWX's content-based caching. Instead of manually tuning around redundant work, RWX uses prior task outputs when the underlying files and definitions have not changed. Graph-based execution added another layer of efficiency. Rather than treating the pipeline as one large unit of work, Honeycomb now builds around discrete tasks. Then, RWX schedules those tasks independently, runs them in parallel when dependencies allow, and reuses them through caching.

Those improvements have a direct impact on day-to-day development loops. PRs move faster, preview deployments complete minutes sooner, and engineers spend less time waiting for CI to catch up before reviewing, validating, or merging changes.

The team also has a clearer path for handling the reliability issues that felt outside developers' control. RWX's built-in OpenTelemetry tracing and observable runs align with Honeycomb's existing approach to understanding what happened inside a pipeline and why. And through a shared Slack channel, the RWX team helps Honeycomb resolve configuration questions, performance issues, and tracing-related feature requests quickly.

That combination makes RWX feel deliberately engineered for real developer workflows. It does not rely on flashy AI features or vague automation claims. RWX provides Honeycomb with the CLI, caching model, task primitives, and observability hooks needed for fast, reliable feedback loops.

Results

Faster validation, reduced failures, and agent-ready CI feedback loops

With RWX, Honeycomb transformed its CI feedback loop from a perpetual blocker to a competitive advantage. P95 build times are now 7 minutes, and failures have been reduced by 50%, enabling Honeycomb to preserve engineering bandwidth for deeper platform development.

• 45% faster CI feedback loops with RWX
• 50% reduction in build failures
• 400+ discrete tasks orchestrated per main build

Looking ahead, RWX will remain Honeycomb's core foundation for validating both human and agentic code. As the company moves into fully autonomous development, the team plans to give agents direct access to the same RWX pipeline that engineers rely on today, enabling fast feedback and turnaround within the development loop.