Systems | Development | Analytics | API | Testing

Testing a FastAPI app that calls OpenAI, Anthropic, or Gemini gets expensive fast. The problem is not just the API bill in production. It is all the repeated traffic in development: prompt tweaks, CI runs, regression checks, and the load tests you keep putting off because every run burns tokens. Hand-written mocks do not help much once the app is doing multi-step LLM work.

Most teams know they are spending money on AI in production. Far fewer realize how much they are spending outside production. It’s easy to get lost as you evaluate which model has the best responses, is fast enough, and cheap enough to run in production. That is because the AI bill usually shows up as a giant blob. It is easy to see the total.

Recent outages tied to AI-assisted code changes have pushed companies into a corner. After several incidents with massive “blast radius” impacts, organizations like Amazon introduced stricter controls—mandating that senior engineers manually review all AI-generated code before it hits production. That response makes sense on paper, but it exposes a fatal flaw in the modern development pipeline.

Your tests are not flaky. Your test data is. That 401 Unauthorized that fails every Monday morning? The OAuth token in your test fixture expired 72 hours ago. The order_id that works in staging but not in CI? It was hardcoded six months ago and the format changed from integer to UUID in January. The timestamp assertion that passes at 2pm and fails at midnight? You are comparing a hardcoded 2026-01-15T14:30:00Z against Date.now(). These are not test infrastructure problems. Retrying them will not help.

The pitch was simple: let AI write your code so you can focus on the hard problems. Three years into the AI coding revolution, and developers are focused on hard problems alright, just not the ones anyone expected. Instead of designing systems and solving business logic, engineers in 2026 spend a startling amount of their day managing the AI itself. Should you use Fast Mode or Deep Thinking? Haiku or Opus? Cursor or Claude Code or Windsurf? Should you write a SKILL.md file or a custom system prompt?

Your WireMock stubs are lying to you. They were accurate when someone wrote them six months ago, but the payment API added a metadata field in January, the inventory service switched from REST to gRPC in February, and nobody updated the stubs because the tests still pass. Meanwhile, production is breaking in ways your mocks will never catch. This is not a WireMock problem. It is a hand-written mock problem.

Production bugs that only reproduce in actual traffic can be some of the most frustrating bugs in software development. You can stare at your logs, add traces to your code, add instrumentation – and still not be able to see the actual requests that went over the wire. And that gets even harder when the requests are encrypted and the system is a black box. You can use tools like Wireshark or Kubeshark to capture the requests.

Enterprise Spring Boot APIs should be tested at three levels: unit tests for business logic, integration tests for external service behavior, and traffic replay for production edge cases. Most teams only do the first. This guide shows all three using a real Spring Boot application that calls external APIs (SpaceX, US Treasury) with JWT authentication. The kind of service that looks simple in development and breaks in production.

We’re excited to announce that Speedscale’s proxymock is now available as an OpenClaw skill on ClawHub, bringing powerful API traffic replay capabilities directly into your Claude workflows.