AWS Makes Its Contact-Center AI Designer Generally Available With Rules Beside the Model

AWS has made agentic CX designer generally available in Amazon Connect Customer, giving contact-center teams a no-code canvas that places generative AI conversation steps alongside fixed business-rule steps. The design targets a central production problem: a system can converse naturally without being allowed to improvise identity verification, credit eligibility, compliance disclosures, or escalation decisions.

The split is more consequential than a visual-builder feature. Agentic steps can interpret intent, retrieve approved knowledge and formulate a response. Deterministic steps collect constrained data through validated fields, with configurable retries and explicit escalation routes. AWS is positioning the two as complementary jobs in one workflow, rather than asking a model to make every decision in a customer interaction.

Two systems of decision-making, one customer flow

The agentic side is intended for the ambiguous parts of a call: understanding what a customer means, asking follow-up questions, drawing from approved knowledge, and tailoring the wording of an answer. Builders can select models for individual agentic steps, matching a model to speed-sensitive conversation or more involved multi-turn work. AWS says Connect Customer does not charge by model or token.

The deterministic side is for decisions with a prescribed answer. Its slots can validate dates, emails, phone numbers and custom constrained values; builders can add regular-expression checks, retry limits and a defined path when validation fails. That gives policy owners a way to set the conditions around actions such as an account check or credit decision without making that judgment a language-model output.

A call can become a screen session without a handoff

AWS is also extending the interaction beyond speech. Live Sync connects a live voice conversation to an on-screen interface in real time, so a customer can view and select options without leaving the call. The company’s Touchpoint client SDK is designed to embed web chat, in-app voice and those multimodal capabilities in a website or mobile application.

That approach shifts a phone interaction from a purely spoken exchange into a shared session: the assistant can present a choice visually, and a tap or form submission returns to the same flow. Connect Customer also offers agentic voices across more than 50 languages and 100 voices, with interruption handling, adaptive tone and mid-conversation language switching, according to AWS.

Integrations and transfers remain part of the product boundary

A conversation still needs access to systems that hold account records, service status and transaction functions. At general availability, the designer supports REST requests to GET, POST, PUT and DELETE API endpoints, plus MCP egress calls to tools exposed through a remote Model Context Protocol endpoint. AWS says secrets can be stored in the workspace and referenced in request headers instead of being placed in individual workflow nodes.

The final distinction is between resolving a request and moving it elsewhere. When a conversation escalates to a human agent or another AI agent, collected fields, variables, invoked tools and their results can persist across the transfer. That could remove the familiar need for a customer to repeat information, but it also makes the accuracy of the preceding workflow and its integrations consequential to the next handler.