A quick PromptLayer update.

When I launched PromptLayer a few weeks ago, I described it as AI observability.

The more time I've spent talking to developers and watching real AI systems run, the more I've realised that observability is only part of the problem.

The first model call is easy. The twentieth isn't.

Once requests start passing through prompts, tools, retries, fallbacks, caches, multiple models, and branching workflows, it becomes surprisingly difficult to answer simple questions:

• Why did this request cost so much?

• Why did this workflow suddenly get slower?

• Which path was actually executed?

• What changed between successful and failed runs?

• Where are we spending tokens?

Over the last few weeks I've rebuilt large parts of PromptLayer around answering those questions.

The product is becoming less about collecting traces and more about helping developers understand how their AI systems actually behave.

Still a lot to build, but the direction feels much clearer now.

If you've tried PromptLayer before, I'd love for you to take another look and let me know what you think.