
Modern Observability, OTel, and Spans

Stop being skeptical about AI for development with Charity Majors
Topic 13 of 17
Published Aug 12, 2026 · Clip 55:12–1:01:56 (6:44)
Charity explains that auto-instrumentation with OpenTelemetry is now so good that it's literally faster and easier to build with instrumentation than without. She recounts how old-school instrumentation choices (metric vs. log vs. trace, cardinality, dashboards) used to multiply development time, whereas now telemetry arrives in the development loop—Honeycomb even surfaces suggestions inline. She defines spans and traces, argues the transaction no longer works as the default building block for AI-era workloads, and describes Honeycomb's new timeline product for zooming out across multi-agent, multi-hour traces. On agents querying observability, she notes most telemetry is trash data and the relationships between events matter most; some AI SRE startups bypass traditional telemetry for richer upstream data.