LangChain founder: agent observability is 'Databricks for agents', not Datadog
hwchase17 · x · 2026-10-31
LangChain founder Harrison Chase and co-founder Ankush Gola break down why agent observability differs fundamentally from traditional observability.
Traditional patterns: leading-edge queries on recent minutes/hours; fast breakage detection and paging; digging into latency and errors.
Agent observability adds two new patterns: mining traces across longer time horizons to surface failure modes, edge cases, and high-quality examples; and turning those findings into better evals / training data so agents keep improving.
Chase's framing: it's not "Datadog for agents" but "Databricks for agents" — the value lies in accumulated data assets. He says LangChain has done some work here with more coming soon.
Related event: LangChain Founder: Agent Observability Is Databricks, Not Datadog(2 posts)→
More from coding & agent
- Three.js merges Nanite-style meshlet pipeline, rendering 1B triangles at 150FPS — anselm · 2026-10-31
- Steve Yegge: Beads hits 4 transactions/sec worldwide, 500k downloads this month — Steve_Yegge · 2026-10-31
- agent-stat 0.3.0 tracks your Codex and Claude Code usage and costs locally — talkaboutdesign · 2026-10-31
- Northflank hacked, all env vars stolen; key-hoarding platforms are supply-chain magnets — andersonbcdefg · 2026-10-31
- Game dev shares 'character gym' workflow: ask AI to make every movement parameter tunable — chongdashu · 2026-10-31
- Google XR Labs lead demos on-device spatial AI with XR Blocks, LiteRT.js and Gemini — DuRuofei · 2026-10-31