Importing Langflow & LangGraph agents into watsonx Orchestrate

One of the quieter but most consequential pieces of IBM's Think 2026 watsonx Orchestrate update is multi-framework agent import. Orchestrate now accepts agents built in Langflow, LangGraph, and the open A2A protocol — in addition to IBM native agents — and registers them in the same Control Plane. For teams that have been pushed to standardize on a single framework before going to production, this is the announcement that changes the conversation. The standardization debate was always more about platform fear than engineering need, and now there's a credible answer.

The Langflow-to-production gap, closed

Here's the practical pattern we see in the field: a data-science team prototypes a workflow in Langflow because the visual canvas is fast to iterate in and easy to demo to a business sponsor. A platform team's been waiting for it to be "rewritten for production" before it can go live — usually for six to nine months, sometimes longer if the original developer has rotated to another project. With multi-framework import, that rewrite stops being necessary. The Langflow flow exports to JSON, imports into Orchestrate as a tool, and inherits the same governance, observability, and identity that any native agent would get.

What the platform wraps around the agent

The same applies to LangGraph. The orchestration logic stays where the developer wrote it; what changes is what wraps it: identity (via your existing IdP), audit logging (per-turn, with the same retention policy as the rest of the agent estate), policy enforcement (against the same policy surface other agents use), and — through the native Langfuse integration — traces and spans flowing into your observability stack alongside everything else without bespoke instrumentation. The agent code itself doesn't need to know any of this is happening, which is the right separation of concerns and the reason this architecture will age well.

You keep your team's framework choices; you give your security team the audit surface they need; nobody has to negotiate.

There's an architectural lesson buried in this: the right boundary between "the agent" and "the platform that runs the agent" turns out to be different from what most early adopters assumed. The agent is the reasoning logic and the tool list — the pieces the business actually cares about. The platform is identity, governance, observability, and lifecycle — the pieces the security team and the SREs care about. Orchestrate's import support draws the line at that boundary cleanly. You keep your team's framework choices; you give your security team the audit surface they need; nobody has to negotiate.

What to do with your stuck prototypes

If you're sitting on Langflow or LangGraph prototypes that haven't shipped because they were stuck in the "productionize it first" queue, this is the moment to revisit. They can ship into the same Control Plane your IBM native agents are running in — without a rewrite, without losing the velocity that got them built in the first place, and without forcing your platform team to learn yet another runtime. The architectural conversation moves from "which framework do we standardize on" to "what's the right framework for this specific workload," which is a much healthier place to be — and it frees engineering leaders from a debate that was never really about engineering in the first place.