Amazon Connect • Guide & Hub • 2026

ACXD: What It Is, How It Works & Everything to Know

ACXD is one of the most important additions to Amazon Connect in years — and one of the least explained. This is the complete guide to ACXD: what it is, what it does, how it fits together, and a hub linking every deep-dive on building, securing, shipping, and operating it.

ACXD stands for Amazon Connect Agentic CX Designer. It's a no-code service within Amazon Connect Customer for building agentic AI conversational applications — the kind that don't just answer a question, but guide a customer through a whole task and take action on their behalf.

If you've seen "ACXD" in AWS material or job descriptions and wanted a straight explanation plus a map of everything around it, this page is both: a concise guide to the term, and a hub into the detailed articles.

On this page
  1. What ACXD is, in plain terms
  2. What you can build with ACXD
  3. How ACXD works
  4. ACXD vs contact flows vs Lex
  5. The ACXD hub: every deep-dive
  6. ACXD FAQ

What ACXD Is, in Plain Terms

ACXD (Amazon Connect Agentic CX Designer) gives teams a single workspace to build AI-powered conversational applications that can guide structured workflows, answer questions from trusted knowledge, connect to APIs, reason through multi-step tasks, and escalate to human agents. AWS deliberately calls these conversational applications rather than "bots," because they do far more than respond to simple messages — they're agentic, task-completing AI.

The one-line version: ACXD is Amazon Connect's no-code environment for building agentic self-service — AI that completes tasks, not just an FAQ chatbot. For the full beginner's walkthrough, see our complete guide to what ACXD is.

What You Can Build With ACXD

ACXD targets the common, high-volume self-service use cases every contact centre wants to automate well:

The defining feature is that a single application can combine deterministic workflow logic with generative AI, integrations, knowledge bases, and guardrails — structured enough for business control, flexible enough for natural conversation.

How ACXD Works

ACXD organises everything into a clear hierarchy:

You build flows visually on a canvas, supported by data requests, knowledge bases, slots and variables, and guardrails. A Connect Customer flow invokes an ACXD application via an Agentic CX block, and the whole lifecycle — build, test, deploy (as immutable builds), and observe — happens in one workspace.

ACXD vs Contact Flows vs Amazon Lex

 What it isWhen to reach for it
Contact flowsDeterministic routing & entry-point logicRouting, queues, channel setup — the front door
Amazon LexIntent-and-slot bot serviceIntent classification and slot filling
ACXDAgentic conversational application design environmentAgentic, task-completing self-service combining logic, GenAI, knowledge & guardrails

They're complementary: contact flows own the entry point, ACXD runs the intelligent conversation, and Lex still has its place and can coexist.

The ACXD Hub: Every Deep-Dive

Once you're building ACXD for real, these are the detailed guides worth having:

What is ACXD? (Beginner's guide)

The full plain-English walkthrough of ACXD, its building blocks, and when to use it.

ACXD Guardrails

Input vs output guardrails, the detection methods, enforcement actions, and best practice.

ACXD CI/CD: SDK & MCP Tools

The ACXD SDK as infrastructure-as-code, MCP tooling, and the honest reality checks.

ACXD CI/CD Pipeline

Integrating agentic Amazon Connect experiences into a full CI/CD pipeline.

ACXD Infrastructure as Code

What's available for ACXD IaC today — SDK only; no CloudFormation, CDK, or Terraform for the app itself.

ACXD Runbooks

Operating ACXD in production — guardrail breaches, hallucinations, bad builds, and fallback-to-human.

Evals for ACXD

Using evaluations to deliver high-performing agentic self-service and catch regressions.

AI Agent Metrics

The new observability metrics for agentic self-service — hallucination rate, groundedness, and more.

Preview note: ACXD was introduced as a preview, so specific capabilities, availability, and pricing can change — always confirm current details in the official AWS documentation before building for production.

ACXD FAQ

What does ACXD stand for?

ACXD stands for Amazon Connect Agentic CX Designer. It is a service within Amazon Connect Customer for building AI-powered, agentic conversational applications — experiences that can guide workflows, answer questions from knowledge, call APIs, reason through multi-step tasks, and escalate to human agents.

What is ACXD used for?

ACXD is used to build agentic self-service applications for contact centres — customer service automation, appointment scheduling, order lookup, account support, knowledge base Q&A, troubleshooting, task completion, and human handoff. A single application combines deterministic workflow logic with generative AI, integrations, knowledge bases, and guardrails.

How is ACXD different from Amazon Lex?

Amazon Lex is an intent-and-slot conversational bot service. ACXD is a higher-level design environment for agentic conversational applications that combine deterministic workflow logic with generative AI, knowledge bases, integrations, and guardrails, built visually on a canvas. For many new builds ACXD is the more modern starting point, though the two can coexist.

Can ACXD be managed as infrastructure as code?

Yes, via the ACXD SDK, which is the only infrastructure-as-code path for ACXD itself today. There is no CloudFormation support, and because the AWS CDK synthesizes to CloudFormation, CDK (like Terraform) can only manage the surrounding AWS resources, not the ACXD application, flows, nodes, or guardrails.

Conclusion

ACXD — Amazon Connect Agentic CX Designer — is AWS's no-code environment for agentic self-service: one workspace to build conversational applications that understand natural language, follow structured logic, tap knowledge and APIs, stay within guardrails, and hand off to humans. Start with the beginner's guide, then use the hub above to go deep on guardrails, CI/CD, infrastructure as code, runbooks, and evals as you move from first build to production operation.