Is Dialogflow CX, Azure AI, or Bedrock Right for You?
In this article
In September 2026, CX leaders compared agent platforms. They usually got to the same three finalists. The first is Google’s Dialogflow CX. The second is Microsoft’s Azure AI Foundry Agent Service. The third is Amazon’s Bedrock AgentCore. Vendor pages compare feature counts. But a CX team must answer two other questions. Who on our team builds and runs this agent? And what happens at a handoff to a person?
ICX builds on all three platforms. This post gives a decision table from that build work. It is not a feature review. First, it explains the name changes. A buyer must see through them.
Why does the naming keep changing?
This post’s title has three names. Two are already out of date. That fact is itself a signal about platform risk.
Google. Google renamed Dialogflow CX to Conversational Agents. Google also merged its console with another one. That was the older Vertex AI Agent Builder console. Existing agents moved over on their own. Google’s Dialogflow change history records this. Then came April 2026. Vertex AI Agent Builder became the Gemini Enterprise Agent Platform. So search for “Dialogflow CX” today. You land on pages with both newer names. A job post from a year ago can use an old name. The product no longer uses it.
Amazon. Amazon did more than rename. Amazon launched Bedrock Agents in November 2023. That feature is now Bedrock Agents Classic. It stopped taking new customers on July 30, 2026. The replacement is Amazon Bedrock AgentCore. It is a different kind of product. AgentCore gives you a managed runtime. It adds an identity layer and an observability layer. It is not a single agent-building feature. In September 2026, AgentCore added a managed consent portal. It is for OAuth sign-in flows. Teams used to build that part by hand.
Microsoft. Foundry Agent Service is the newest of the three names. It came from a rebrand in 2025 and 2026. Azure AI Studio became Microsoft Foundry. The Build 2026 release added a stable orchestration framework. It also made the VS Code toolkit generally available. Other features are still in public preview. Examples are Toolboxes and multi-type memory.
These changes do not make any platform unstable. They do mean old comparisons are partly wrong. Confirm the current names before a procurement talk. Do not wait until after.
The five-question platform framework
ICX compares agent platforms for CX work with five questions. Each question is a decision your team must make. Vendors tend to highlight other things.
| Question | Dialogflow CX (Conversational Agents) | Azure AI Foundry Agent Service | Bedrock AgentCore |
|---|---|---|---|
| Where does the conversation logic live? | In a visual flow builder of pages, intents, and routes. A flow can use generative fallback. | In code, with the Microsoft Agent Framework. One Responses API calls it. | In code, in your own agent framework. AgentCore runs the logic. It does not define it. |
| How do you build escalation to a human? | With a native transfer-to-human step in the flow. Google built it for contact center handoff. | With a custom tool call. Teams often connect it to Teams, Dynamics, or Omnichannel. | With a custom tool call that the team writes. There is no built-in contact center handoff. |
| How do guardrails and evals attach? | Through limits in the flow design, plus Vertex safety filters on generative fallback. | Through Foundry’s own evaluation and tracing, plus Azure AI Content Safety. | Through Bedrock Guardrails, plus AgentCore’s identity and observability parts. |
| What must the team already run? | A Google Cloud project. Often a contact center integration too. | Microsoft 365 and Teams, so the agent can publish where people work. | AWS, and a team that can build the harness itself. |
| How often has the product renamed? | Twice in under two years: first to Conversational Agents, then into the Gemini Enterprise Agent Platform. | Once, from Azure AI Studio. Several features are still in public preview. | Once. The original feature became “Classic” and is closed to new customers. |
Read the table one row at a time. A platform can win one question. It can lose the deal on another. Which question matters most? It depends on what your team lacks most. It can be time for conversation design or engineering time. It can be an existing platform investment.
What ICX has observed building on all three
The real difference is the defaults. The other part is what it expects you to build.
Dialogflow CX. The flow builder gives a stakeholder a working demo fastest. But the flow design is the guardrail. A flow node can enforce a rule. But you cannot test that rule alone. Change one page, and the agent can say new things later. The effect can show up two steps later. Some teams add generative fallback to an existing flow. They must test again every node the fallback can reach.
Bedrock AgentCore. The runtime has no opinion about contact centers. So you write the escalation path yourself from day one. ICX sees this gap as useful. Some teams write their own handoff logic on day one. They tend to design a better handoff. Some teams get a default handoff. They often touch it only when something breaks. This matches ICX’s read on harness engineering. Some platforms have the least built-in structure. They often lead to the most careful structure. No step gets skipped by default.
Azure AI Foundry Agent Service. A finished agent fits well into Teams and Microsoft 365. That matters more for internal service desks. It matters less for customer support. We treat the newer parts as preview features. These are the extra memory types and Toolboxes. We do not use them in production yet. We wait until Microsoft makes them generally available. A preview feature can still run a live agent. It must not be the only path to escalation.
These observations do not replace a pilot. We ask these questions before we recommend a pilot. They come from work on all three platforms. They do not come from one vendor’s documentation.
Which platform fits which team?
Start from what your team already runs. Do not pick the newest-looking platform in a demo.
- On Google Cloud, with a contact center: choose Conversational Agents. The native handoff and the flow builder do real work. The other two platforms leave that work to your team.
- On Microsoft 365 and Teams: choose Foundry Agent Service. Your agent can already reach the people who use it.
- On AWS, building custom agent logic: choose AgentCore. It also fits a team that wants full control of the harness. A default harness is not enough for them.
In every case, the platform is not the hard part. Two of our posts point to the same result. One is the measurement framework for agentic AI. The other is the production gap in agent pilots. Pilots rarely fail because of the platform. They fail on conversation design, escalation rules, and evaluation. Without those, all three platforms show the same problems.
Buyers also ask AI search engines who to trust here. Our AI Citation Index found a pattern. Comparison and best-practice pages earn far more of those citations. Short opinion posts earn fewer. So this post is a table and a framework. It is not a verdict.
Where to start
Pick the platform that matches what your team already runs. Spend the saved time on conversation design and escalation rules. Those decide whether customers trust the result. ICX’s guide to enterprise chatbot design covers that work. Does a pilot stall on the platform question? Then the team usually asks the wrong question first. The design question comes first.
Want a second opinion on your platform? See ICX’s services or get in touch. ICX would love to hear what you’re weighing.
We are building a newsletter. It will track these comparisons as they change. It is coming soon. Bookmark the blog and check back.
How ICX verified this post
We checked every product name, rename date, and feature status. We used vendor documentation on September 27, 2026. The links are inline. Google Cloud’s change history records the Dialogflow rename. It also records the console merge. AWS’s developer guide and blog record two facts. One is the Bedrock Agents Classic retirement date. The other is the AgentCore identity feature. The Foundry feature list follows Microsoft’s Build 2026 announcement. Several named features are still in public preview. The post says this. We did not check these facts again for the November 4, 2026 update. That update changed only the wording and added the diagram.
We draft with AI help. We review every post before it goes live. Read the full policy on the AI disclaimer page.
Sources
Frequently asked questions
What's the actual difference between Dialogflow CX, Azure AI Foundry, and Bedrock AgentCore?
Dialogflow CX gives you a visual flow builder. The conversation logic lives in pages and intents. Azure AI Foundry Agent Service and Bedrock AgentCore are code-first. You write the agent logic in a framework. The platform runs it with managed memory and identity. Only Dialogflow was built mainly for conversation design.
Does Dialogflow CX still exist, or was it renamed?
The product still exists. But Google is phasing out the name. Google renamed Dialogflow CX to Conversational Agents. It merged the console with the Vertex AI Agent Builder console. Then Vertex AI Agent Builder got a new name. In April 2026, it became the Gemini Enterprise Agent Platform. Existing agents moved over on their own. But the documentation now uses the newer names.
What happened to Amazon Bedrock Agents?
Amazon launched Bedrock Agents in November 2023. It is now called Bedrock Agents Classic. It stopped taking new customers on July 30, 2026. Amazon Bedrock AgentCore is now the main managed runtime on AWS. New work must start on AgentCore.
Which platform has built-in escalation to a human?
Only Dialogflow CX has a native transfer-to-human step. It is in the flow builder. Google built it as a contact center product first. The other two platforms are different. There, the team builds escalation as a tool call or a custom integration.
Which platform is easiest for a non-technical team to start with?
Dialogflow CX. Its flow builder does not need orchestration code. Foundry Agent Service and AgentCore both expect a developer to build the agent. Each one needs less custom code in 2026 than before.
How should a CX team actually choose between these three?
Start with what the team already runs. Some teams use Google Cloud with a contact center integration. They get the most from Conversational Agents. Teams on Microsoft 365 get the most from Foundry Agent Service. Some AWS teams build their own agent logic. They get the most from AgentCore. In every case, the platform matters less. Conversation design and escalation rules matter more.
Get an email when we publish a new post.