Bedrock Agents Classic Is Being Retired

Part of a series on running AI workflows across model providers — this one is a heads-up for anyone still building on the original Bedrock Agents feature, with a real date attached.

If you built an agent on Amazon Bedrock before AgentCore existed, you were using a feature that launched back in November 2023, simply called "Bedrock Agents." As of a June 30, 2026 AWS announcement, that original feature has been renamed "Amazon Bedrock Agents Classic" and is moving to maintenance mode on July 30, 2026 — after which AWS will stop accepting new customers onto it, though existing users can keep running it with continued support. If your team has an agent built on the original Bedrock Agents and hasn't looked at AgentCore yet, this is worth reading before that date passes.

What "maintenance mode" actually means here

AWS grouped this transition with two other first-generation AI services moved to the same status on the same date — Amazon Kendra and Amazon Q Business — which is a useful signal about what's happening industry-wide: several of AWS's earlier-generation, more constrained AI products are being succeeded by newer, more flexible ones, while the older versions stay operational for existing customers rather than being shut off abruptly. Concretely, that means: if you already have an agent running on Bedrock Agents Classic, it keeps working, and AWS continues supporting it. What changes is that new customers can no longer start there — AWS is steering all new agent-building work toward AgentCore instead.

Roughly two and a half years between the original feature's launch and its formal succession by AgentCore.

Why AWS built a replacement instead of just improving the original

Bedrock Agents (the original) was a single, fairly opinionated feature: you defined an agent, gave it access to specific AWS-native tools, and Bedrock ran it. AgentCore, by contrast, is a set of separately composable services — Runtime, Memory, Gateway, Identity, Observability, Browser, and Code Interpreter — that work with any agent framework and any foundation model, not just ones native to Bedrock. Rather than extend the original feature's more constrained model, AWS built a broader, more flexible platform alongside it and is now formally directing new work there.

What this means if you're currently on Bedrock Agents Classic

Nothing breaks on July 30, 2026 itself — existing agents keep running, and AWS support continues. The practical planning question is whether to migrate proactively or wait. Reasons to move sooner rather than later: AgentCore's broader framework support (Strands Agents, LangChain, LangGraph, LlamaIndex, CrewAI) may already fit your stack better than the original feature's more constrained tool-integration model, and getting ahead of a maintenance-mode feature avoids being caught flat-footed if AWS eventually sets a harder end-of-life date down the line (none has been announced as of this writing, but "maintenance mode" is often a step on the way to eventual retirement for AWS services, not a permanent state).

The model doesn't have to change during a migration — it's the surrounding agent infrastructure that moves.

A real use case: a team that migrated before being forced to

An internal-tools team at a mid-size company had built a document-classification agent on the original Bedrock Agents back in 2024. When the June 2026 maintenance-mode announcement came out, they didn't wait for a forcing function — they used the migration as an opportunity to also solve a problem they'd been living with: their original agent couldn't easily call an internal API without custom Lambda glue. Moving to AgentCore Gateway let them expose that internal API as an agent-callable tool directly, removing a piece of custom code they'd been maintaining since the original build. The migration cost a sprint; the tool-access improvement was a genuine upgrade they got for free while they were in there anyway.

How this connects to model flexibility more broadly

Whether a team builds on Bedrock Agents Classic, AgentCore, or an open-source framework like LangGraph or CrewAI running on AgentCore Runtime, the underlying lesson is the same one that shows up across every provider in this series: agent infrastructure and the model underneath it are separate decisions, and coupling them too tightly is what makes a migration like this one painful. A team whose agent logic lives in DNotifier's defineAgent/Workflow primitives, calling whichever model provider — Bedrock or otherwise — fits a given step, doesn't face this kind of forced migration when an underlying platform's roadmap shifts, because the orchestration layer and the model layer were never coupled together in the first place.

Frequently asked questions

Will my existing Bedrock Agents Classic agent stop working on July 30, 2026?

No — existing customers can keep running it with continued AWS support. The maintenance-mode date only stops new customers from starting on it.

Has AWS announced a hard end-of-life date for Bedrock Agents Classic?

Not as of this writing. Maintenance mode is a distinct status from end-of-life, though it's often a step toward eventual retirement for AWS services.

Do I need to rewrite my agent's logic to migrate to AgentCore?

Not necessarily — AgentCore supports popular open-source frameworks directly, so depending on how your original agent was built, migration may be more about infrastructure than a full logic rewrite.

Were Amazon Kendra and Amazon Q Business retired for the same reason?

They moved to maintenance mode on the same date as part of the same broader announcement, but each is a separate product with its own migration path — this article covers Bedrock Agents specifically.

The Bottom Line

Next in this series: how to actually choose between the model families available on Bedrock — Nova, Claude, Llama, and Mistral — instead of defaulting to whichever one you tried first.

Read part five: Choosing a Model on Amazon Bedrock. Explore dnotifier.com.