Writing on agent observability
Deep dives into AI agent monitoring — alerting, cost-per-product analytics, execution traces, and how AgentPulse works under the hood.
Why We Built AgentPulse
AI agents fail quietly and bill loudly. The story behind AgentPulse — from a runaway token bill discovered after the fact to an alerting-first observability tool that tells you the moment a run goes wrong.
Read post →Alerting Beats Dashboards for Agent Reliability
Nobody watches a dashboard at 2am. Why agent observability has to be alerting-first — and what thresholds actually catch the failures that matter before they cost you.
Tracking LLM Cost per Product, Not per Model
Your model invoice tells you what you spent, not where it went. How attributing token cost to products and agents turns a surprise bill into a budget you can actually manage.
Instrumenting an Agent with One HTTP Call
No SDK, no library, no agent code changes. How to report a run to AgentPulse with a single POST /v1/runs — and why a webhook beats a vendor SDK for portability.
How to Read an Agent Execution Trace
LLM calls, tool calls, retrievals, and reasoning steps with per-step timing and tokens. A walkthrough of what to look for when a run is slow, expensive, or just wrong.