The Pod Was Cheaper. The Service Wasn’t.
Use OpenCost and proxymock to prove Kubernetes rightsizing lowers cost per successful request without hiding behavior or throughput regressions in testing.
Speedscale Blog
Latest insights on Agentic AI workflows, cloud native architectures, and performance optimization best practices from the Speedscale team.
Featured Post
Recent Posts
Flamegraphs Find It. Replay Proves It.
Use Grafana Pyroscope and proxymock with an AI coding agent to find a Go CPU hotspot, preserve API behavior, and verify the performance fix.
eBPF: Preventing Garbage HTTP Payloads When Reading Kernel Scatter-Gather Buffers
How our nettap eBPF agent produced garbage HTTP bodies from kernel iov_iter scatter-gather buffers, and the CO-RE plus task-local storage fix that made it correct.
The Best Model Is a Routing Decision
The best model isn't the smartest — it's whichever should get the next unit of work. Notes from NVIDIA on routing, local models, and the AI factory.
Two confident fixes missed this production bug
A production bug survived two confident fixes and green tests. Replaying the captured request exposed the missing state and proved the real fix.
Imaginary Test Data. Real Token Bill.
Testing AI applications with invented traffic looks fine until real users arrive. Then come the retries, the fallback models, and the token bill.
Top 14 Postman Alternatives for API Testing in 2026
A rebuilt 2026 guide to Postman alternatives. Which API clients are still maintained, what they cost, and which one fits local-first, gRPC, or team workflows.
AI writes code in seconds, but delivery still takes days
AI pushed throughput up 59%, yet median delivery got worse. The bottleneck moved to validation. How replaying production traffic in CI closes the gap.
The API tests passed. The database didn't.
Our v2 release looked clean in HTTP tests until we diffed the SQL workload — an N+1 loop, a startup migration, and 70 ms of extra DB time hiding in plain sight. Here's how to compare two releases without database access.
Trace without traces
The trace was sampled out. I found the bug anyway — by filtering recorded traffic on the customer's email instead of a trace ID. Here's how to follow one request across four services with no trace IDs and no OpenTelemetry.