LangSmith launches Trajectories: a readable timeline view of full agent sessions
LangChain · x · 2026-09-30
LangChain launched Trajectories in LangSmith, a chronological, human-readable view of agent sessions.
Design
- Existing traces nest runs into an execution tree across multi-turn threads; Trajectories flatten this into ordered messages from humans, AI, tools, and subagents
Uses
- Debugging: pinpoint where behavior deviates before diving into full traces
- Review: SMEs can audit agent behavior without parsing nested traces
- Improvement: score trajectories with online evaluators, route to annotation queues or datasets, and export production sessions as SFT fine-tuning data
Related event: LangSmith Launches Trajectories for Readable Agent Runs(2 posts)→
More from coding & agent
- Open source maintainers push back as agents spam PRs to farm GitHub stats — wightmanr · 2026-09-30
- Claude Code 2.1.285 removes a dozen legacy models, prompt tokens jump 51% — ClaudeCodeLog · 2026-09-30
- Claude Code 2.1.285 ships 136 CLI changes, adds WebFetch kill switch — ClaudeCodeLog · 2026-09-30
- XBOW's AI agent finds and exploits a Linux kernel 0-day, achieving local root privilege escalation — moyix · 2026-09-30
- With Login with OpenAI, harness token efficiency becomes the new price war — NathanWilbanks_ · 2026-09-30
- Cerebras runs Qwen 3.8 27B at ~1,500 tokens/sec, making an AI assistant 19x faster at dinner reservations — Sethwinterroth · 2026-09-30