Staff+ Software Engineer, Observability

About the role

Anthropic is seeking Software Engineers to join our Observability team within the Infrastructure organization. The Observability team owns the monitoring and telemetry infrastructure that every engineer and researcher at Anthropic depends on, from metrics and logging pipelines to distributed tracing, profiling, error analytics, alerting, and the dashboards and query interfaces that make it all actionable. By joining this team, you'll have a direct impact on the reliability and operational excellence of Anthropic's research and product systems.

As Anthropic scales its infrastructure across large GPU, TPU, and Trainium clusters, the volume and complexity of operational data is growing by orders of magnitude, and an increasing share of the most challenging problems live below the application layer. We're building next-generation observability systems, including high-throughput telemetry pipelines, fleet-wide continuous profiling, eBPF-based tracing and network visibility, and agentic diagnostic tools, so engineers can detect, diagnose, and resolve issues in minutes rather than hours, even when the answer lies in the kernel, the network stack, or an accelerator rather than on a dashboard.

Key responsibilities

• Design and build scalable telemetry ingest and storage pipelines for metrics, logs, traces, and error data across Anthropic's multi-cluster infrastructure
• Build observability solutions that give engineers deep, low-overhead visibility into system behavior across the fleet
• Own and evolve core observability platforms, leading migrations and architectural improvements that improve reliability, reduce cost, and scale with organizational growth
• Build instrumentation libraries, SDKs, and eBPF-based auto-instrumentation to emit high-quality telemetry, with and without code changes
• Reduce mean time to detection and resolution by building cross-signal correlation (from kernel-level events up to application traces), unified query interfaces, and AI-assisted diagnostic tooling
• Improve fleet-wide efficiency by turning continuous profiling and utilization telemetry into actionable optimization insights across CPU, memory, and accelerator fleets
• Partner with Research, Inference, Product, and Infrastructure teams to ensure observability solutions meet the unique needs of each organization

Minimum qualifications

• Experience building and operating large-scale observability or monitoring infrastructure
• Hands-on experience with observability signals end to end, from instrumentation through ingest to query and analysis
• Understanding of high-throughput telemetry pipelines and the tradeoffs involved in collecting, storing, and querying operational data at scale
• Comfort debugging below the application layer, in the kernel, the network stack, or the hardware
• Excellent communication skills and the ability to partner with internal teams to improve their operational visibility and incident response capabilities
• Ability to work independently on ambiguous, high-impact technical challenges

Preferred qualifications

• Experience building or operating eBPF-based observability in production, such as tracing, profiling, or network visibility
• Experience running continuous profiling at fleet scale, including managing overhead budgets and symbolization
• Kernel- and syscall-level debugging experience and performance engineering skills
• Experience profiling or instrumenting accelerator workloads
• Experience operating metrics systems at very high cardinality, or large-scale telemetry storage backends
• Experience with log storage migrations or operating columnar databases (e.g., ClickHouse, BigQuery) for analytics workloads
• Experience with OpenTelemetry instrumentation, collector pipelines, and tail-based sampling strategies
• Interest in applying AI/LLMs to operational workflows such as automated root cause analysis, anomaly detection, or intelligent alerting
• Enthusiasm for building foundational infrastructure