Overview
LangSmith is the agent development and observability platform from LangChain. On 17 November 2026 the company announced it had been rebuilt around the agent as the core object, as the LangSmith Agent Platform, published alongside LangChain's Series C.
The change is in the object model. Before the rebuild, LangSmith was organised around runs and traces, which answer what happened inside a single call. Now the agent itself is a first-class object with its own agent record, and the vendor describes its scope as build, run, improve and govern, with registry and governance added on top. For a team maintaining a dozen agents, that difference is concrete: versions, permissions and evaluation verdicts can hang off one object instead of being tracked in a spreadsheet.
Why this matters has to do with where agent engineering stands in 2026. Tracing tools answer how to investigate a failure after it happens. Once agent count grows, the more common pain is not remembering which version is live, who approved the rollout, or when the last evaluation ran. Treating an agent as an object turns those things from tribal knowledge into a row on the platform.
Two caveats. The announcement does not state the size of the Series C and does not touch pricing, so the free allowance and paid tiers remain as listed on the site. Third-party hands-on reports and migration write-ups were still scarce as of 17 November, so teams already on the previous LangSmith should check the official migration documentation rather than infer the path from a blog headline.
Key Features
- The agent as the core object: Organisation moves from runs and traces to agents, with one record tying together an agent's versions, runs and evaluations.
- Agent record: A new record type holding an individual agent's identity and history, replacing the lookup table teams used to maintain themselves.
- Build, run, improve, govern: The vendor's lifecycle split, covering construction, operation, iteration and governance.
- Registry and governance: Agent registration and governance controls aimed at admission and audit when many agents run in parallel.
- Tracing and evaluation continue: Existing call tracing, datasets and evaluation remain in the same platform; the rebuild changes organisation rather than removing prior capability.
- Native to the LangChain stack: LangChain and LangGraph projects connect directly and stay under the same account and project structure after the rebuild.
Use Cases
- Running several agents with a single place for versions and admission control
- Evaluating before release and tracing after release in one tool
- Answering which version is live and who signed off, as a team
- Building LangGraph workflows that need matching observability and regression records
Pros
- The object model follows the agent lifecycle instead of scattering call records
- Agent records give versions and evaluation verdicts a home
- Registry and governance fill the admission gap for parallel agents
- Tracing, datasets and evaluation stay in one platform, so switching is bounded
- Direct integration with LangChain and LangGraph avoids a second access layer
Pricing
The rebuild announcement does not change pricing. LangSmith continues to bill on the tiers and free allowance listed on its site; quotas, seats and overage terms should be read from the official page. The Series C amount is not stated in this post and needs the funding announcement.
Summary
The value sits on the governance side: an agent stops being a pile of traces and becomes something you manage. Teams already running multiple agents in parallel with several people involved map onto this structure well. For single-pipeline debugging the previous tracing was already enough, and there is no need to migrate for governance features alone. Hands-on reports are still thin, so evaluate against the migration documentation and a trial.
Version History
- Rebuilt as the LangSmith Agent Platform (2026-11-17): LangChain announced the rebuilt LangSmith Agent Platform, organised around the agent as the core object and covering build, run, improve and govern, with agent records, registry and governance. Published alongside the Series C. No pricing change was announced and the funding amount is not in the post.