LangChain Managed Deep Agents v0.10 adds browsers and asynchronous subagents
Managed Deep Agents v0.10 combines browser control, human sign-in handoff and asynchronous subagents. The release makes long-running web workflows easier to build, while raising the stakes for session and task control.
LangChain announced browser tools, human sign-in handoff and asynchronous subagents on October 29. Managed Deep Agents v0.10 is a public beta; the new capabilities do not establish security or task-success benchmarks.
Claims and evidence
- Original source
Official project or dated release documentation supports the described features. View source ↗
- Not independently verified
No independently established reliability, security or benchmark performance is asserted.
What's new in v0.10
On October 29, 2026, LangChain announced Managed Deep Agents v0.10, with built-in browser use and asynchronous subagents. The product is described as available in public beta.
It is a managed runtime for running and coordinating agents, not a newly trained foundation model.
A managed browser for websites without APIs
Many interactive websites do not offer APIs for every task. Agents need a rendered browser to click, fill forms and navigate authenticated pages. Managed Deep Agents now provides browser_snapshot, browser_run, browser_screenshot and browser_handoff.
The browser runs in a separate LangSmith sandbox. These documented tools make web interaction possible, but do not establish a verified success rate for bookings, forms or account changes.
Human sign-in with browser_handoff
With browser_handoff, an agent can give control of the remote browser to a person when it encounters a login. The user authenticates directly and gives control back, keeping passwords out of model messages and traces.
According to LangChain's official description, handoff links expire. The resumed agent can nevertheless access the signed-in session, so permissions and approval gates still matter.
Async subagents continue in the background
Subagents previously ran inline in this managed runtime, blocking the main agent until they returned. Version 0.10 supports asynchronous tasks so a delegated agent can work in its own thread while the main conversation continues.
The orchestrator can check, update or cancel background work. Results arrive after completion. Because subagents can share the same sandbox, teams need to manage file conflicts and duplicated operations.
Browser scopes, Slack origins and stopping runs
The default browser is scoped to a conversation; a user-scoped browser can share cookies and sign-ins across conversations. The second option trades additional convenience for longer-lived access.
The release also adds runtime.thread origins, replies back to the originating Slack thread and a Stop button for running tasks. Stopping a run cannot undo an external action that has already completed.
Parallel execution needs verification
Parallel execution does not guarantee faster or better results. It can produce duplicate tool calls, conflicting edits and higher total cost.
A sound pilot measures correct task completion, model and tool spending, authorization enforcement and recoverability. AgentenCode has not carried out an independent performance or security audit.
Conclusion
Agents increasingly need durable runtime infrastructure to coordinate browsers, people and concurrent tasks. Managed Deep Agents is one example of this architectural direction.
The meaningful test is whether work is correct, permissioned, traceable and affordable—not whether a demonstration looks autonomous.
Original sources and editorial boundaries
- LangChain: Managed Deep Agents v0.10 (29.10.2026)
- LangChain: Browser Use in Managed Deep Agents (29.10.2026)
- LangChain: Deep Agents – Dokumentation
Editorial status: October 31, 2026. The claims refer to vendor material. No hands-on benchmark, independent security test or formal certification is claimed.
Frequently asked questions
What changed in Managed Deep Agents v0.10?
Browser use with human sign-in handoff and asynchronous subagents.
Does the model need my password?
browser_handoff allows the user to sign in directly; the resulting authenticated session still needs access controls.
Can subagents work in parallel?
Yes. Asynchronous tasks can run while the main agent continues, but this does not guarantee better results.