Thesis
Trajectories make the session’s behavioral path a usable unit for debugging and review, while leaving the underlying trace as the execution-detail record.
Observed facts
LangChain describes Trajectories as aggregating messages from humans, AI, and tools across a main agent and its subagents, in the order they first appeared, with each message shown once.
1The company says trajectories can be scored by online evaluators and routed to annotation queues or datasets.
1LangChain says the feature is available on all LangSmith plans in the US.
1
Mechanism
A trajectory is a projection over traces in a thread: it removes nested run structure while retaining the messages and actions that explain behavior. This makes a long session easier to scan for the point of deviation, after which an engineer can inspect the corresponding full trace for inputs, outputs, timing, retries, and nested execution details.
Why now
LangChain is presenting session-level behavior as an input to production evaluation and agent improvement, not merely as a more readable debugging display.
Implications
Reviewers who are not runtime engineers may be able to assess behavior through a readable session view.
Teams may also be able to connect production sessions to evaluator scoring, human feedback, datasets, and supervised fine-tuning examples.
Evidence boundary
These capabilities and benefits are described by LangChain in its product announcement; the supplied source does not provide independent measurements of debugging speed, evaluator quality, or downstream model improvement.
Watch conditions
Look for evidence of how teams use trajectories in production, including whether online evaluators and human review produce reliable signals and whether exported sessions improve later testing or fine-tuning.