#tokenjam
5 posts
-
Watching Claude Code with OTel: what Cursor and /cost won't show you
Claude Code ships a real OpenTelemetry pipe. Cursor doesn't. /cost is per-session and read-only. Here's what you can do with the wire, what each surface actually emits, and the failure modes none of the built-in views catch.
-
LangSmith costs $39/seat. And 10.7x that in real TCO. What self-hosted alternatives actually cost in 2026.
A pricing teardown of LangSmith (and the new SmithDB / LangSmith Engine launch), Langfuse self-host, and a local-first DuckDB alternative. Real numbers, real config, real cost-of-running.
-
The taxonomy of agent failure: 13 named alerts beat 'anomaly detected' at 2am
Every AI observability vendor ships 'anomaly detected.' That's the wrong abstraction for autonomous agents. Here's the typed vocabulary we ship instead. 13 named failure modes, each with its own trigger, payload, and prescribed response.
-
How to monitor Claude Code: a practical guide for indie devs running it unsupervised
A step-by-step guide to monitoring Claude Code on your own laptop. Turn on Anthropic's OTel telemetry, route the spans somewhere useful, and wire up alerts that fire while the agent is still running.
-
Behavioral drift detection for AI agents
A technical deep-dive on detecting when an agent's behavior wanders off its baseline. Using Z-scores on token / duration / tool-count distributions and Jaccard similarity on tool sequences, run locally over your own session history.