Skip to main content

Running NLP flows alongside AI agents

What shipped​

An automation migrated from the NLP platform to the AI agent platform can now run both at once: it keeps its existing intents and NLU-triggered flows, and you can add AI agents on top. Previously moving to the agent platform meant the intent-triggered flows stopped driving the conversation, so a migration had to be all-or-nothing.

Routing is shared between the two. Natural language understanding (NLU) stays in charge of the flows it already knows; the Context Expert only decides the turns NLU could not match.

Who it's for​

Admins and developers moving an existing NLP automation onto the AI agent platform who want to migrate gradually — adding agents for new journeys while the trained intents keep serving the ones already live.

How a turn is routed​

On each new turn, with no flow already in progress:

  1. NLU runs first. The trained intents are matched against what the customer said. A confident intent → flow match enters that flow directly, and no Context Expert call is made.
  2. Otherwise the Context Expert decides. When NLU matches nothing, the Context Expert chooses among both the AI agents and the eligible flows, and hands the turn to whichever fits.
  3. Inside a flow, NLU stays on. Entity slot-filling and NLU conditions keep working mid-flow, exactly as they did before the migration.

The intent confidence threshold is the one already configured on the automation — nothing new is introduced, so flows that routed before the migration keep routing the same way.

Enabling it​

Set the following on the automation's llmAgentConfig:

agentAI.enabled  = true
agentAI.version  = "v3"
agentAI.v2Hybrid = true

All three are required. v2Hybrid is the switch; enabled and version are preconditions, so setting v2Hybrid on an automation that is not on the agent platform has no effect. One flag covers everything — NLU-first routing and Context Expert access to flows are enabled together.

What changes in Studio​

Because the automation still routes on its intents, Studio keeps the NLU authoring surfaces that the agent experience normally hides — Train (intents, entities, FAQs), the full flows list, and the classic flow nodes — while also showing the agent surfaces. See the linked PR below for the detail.

Flow descriptions matter now​

Worth knowing before you migrate, because it is the most common surprise:

When the Context Expert picks between agents and flows, it sees only each flow's name and description — not its contents, nodes, or the intents it was trained on. A flow with no usable description is invisible to the Context Expert and can only be reached by an NLU intent match.

Flows authored for intent triggering often have no description, so review them after migrating and write each one as routing intent — "when the customer wants to check an existing claim" — rather than as an internal label.