Observe It
Logs, metrics, traces, analytics — visibility without PII storms — matched to the quality regime.
What it is
Section titled “What it is”Make production (or staging) speak so you are not guessing. Instrumentation shape follows the system: service golden signals, web RUM/vitals, or LLM traces and quality scores.
Why it works
Section titled “Why it works”Google SRE’s four golden signals (latency, traffic, errors, saturation) are the minimum for user-facing services. Interactive products also need field Web Vitals and UX error paths. Generative systems need hierarchical traces of model/tool/retrieval steps plus scores (online evals, user feedback) — traditional APM alone under-describes quality (Quality regimes). OpenTelemetry keeps instrumentation portable; Langfuse is this house’s default LLM observability/eval plane on top of that idea. Watch cardinality and PII in every regime.
When to use it
Section titled “When to use it”When prod/staging is silent, quality is drifting, or readiness needs live evidence — after naming the regime.
- Pick signals for regime A (incl. data freshness/correctness), B (vitals + errors), or C (traces + scores)
- Invoke the skill; follow its safety rules
- Keep one write owner; collect evidence before claiming done
- Hand off to the next practice instead of boiling the ocean
- Equate “no 5xx” with “good generations”
- Flood logs with prompts/PII
- Skip orientation when you’re lost
- Spawn overlapping agents to “go faster”
Related concepts
Section titled “Related concepts”04-evidence-over-vibes, 11-quality-regimes, Operate
Further reading
Section titled “Further reading”- Quality regimes (handbook)
- Google SRE — Four golden signals
- web.dev — Web Vitals
- OpenTelemetry documentation
- Langfuse docs
See also the handbook Sources & grounding bibliography.
Agent skill
Section titled “Agent skill”observe-it · companion langfuse when operating LLM traces/evals