Catch silent semantic failures
Learn why 200 OK can still mean a broken user experience, and how to validate intent vs. tool execution.
Free Content · Inspired by the book
First-person field notes from building and debugging AI agents in production—free stories inspired by the book to help you see what happens between a prompt and a 200 OK.
What you need to know about observability when you live among agents
Free Content · Field Notes
Each post is a first-person experience from shipping AI agents—the silent failures, the traces that saved me, and the guardrails that stuck.
Every dashboard was green. Zero exceptions in Sentry, 180ms latency, and a happy 200 OK. Yet the agent confidently confirmed a flight to Texas instead of Paris.
Changing a single sentence in a system prompt felt like blindfolded surgery—until I borrowed distributed tracing from microservices and saw every step in one timeline.
An autonomous retry loop + a silent schema mismatch burned through a week's API budget in twenty minutes. Here is how I set guardrails I could actually see.
Core pillars
Observability is not another dashboard—it is the habit of making every agent decision inspectable.
Learn why 200 OK can still mean a broken user experience, and how to validate intent vs. tool execution.
Connect prompts, RAG chunks, tool calls, and model responses into one OpenTelemetry timeline.
Spot autonomous retry loops, repeated tool signatures, and TTFT bottlenecks before the bill arrives.
What you need to know about observability when you live among agents
The Book
The complete field guide by Dany Paredes—from OpenTelemetry semantic conventions to production evaluations and cost guardrails.
Explore the book Free stories above are inspired by chapters in the book.