On July 30, 2026, AWS started returning HTTP 403 errors to any account that tried to create a new Bedrock Agent without prior service usage. The message: "Bedrock Agents is in Maintenance Mode. New agent creation is not available for accounts without prior service usage." That's a hard gate, not a deprecation warning.
Amazon Bedrock Agents, launched in November 2023, is now Amazon Bedrock Agents Classic and will no longer be open to new customers starting July 30, 2026. If you already have agents running, they keep working - existing customers can continue to use the service as normal. But the model catalog available in Bedrock Agents Classic is frozen as of that date. No newly released foundation models will appear in the Classic orchestration layer. An agent that can't get a newer model is an agent with a ceiling.
The timing matters because it collided with another clock. The EU AI Act's most consequential tier of obligations - applicable to high-risk AI systems under Annex III - became enforceable on August 2, 2026. That's three days after the AWS gate went up. Teams rebuilding agent infrastructure this week are doing it inside a new regulatory environment, not before it.
What AgentCore actually is (and why it isn't a rename)
AgentCore is a framework-agnostic runtime for deploying agents - not a managed agent service. That distinction matters operationally.
Classic was a managed agent: you configured an agent in the console, AWS owned the orchestration loop, and your code only existed at the edges, inside action group Lambdas. AgentCore flips this. You own the loop - you write it using LangGraph, CrewAI, Strands, or your own code - and AgentCore provides the surrounding infrastructure: Runtime, Identity, Memory, Gateway, and Observability as modular services you assemble.
Amazon Bedrock AgentCore is framework-agnostic infrastructure for agents. Services include Runtime, Identity, Memory, Gateway, Browser, Code Interpreter, Observability, Policy, and Evaluations. You can bring Strands, LangGraph, CrewAI, another supported framework, or custom code.
AgentCore is not a renamed Bedrock Agents. It is a complete re-architecture. The old Bedrock Agents was a single-agent framework with basic tool calling; AgentCore lets you build systems of agents that share memory, delegate tasks, and route through a unified gateway.
The two migration paths AWS offers:
| Path | What you own | Best for |
|---|---|---|
| AgentCore Harness | System prompt, tools, model choice | Teams migrating from Classic; config-driven, similar feel |
| AgentCore Runtime | Full agent loop (Python / Node.js) | Teams with existing frameworks or custom orchestration |
AgentCore Harness is a config-based experience similar to Bedrock Agents Classic - declare your model, tools, and instructions, and AgentCore handles compute, memory, identity, and observability. This is the closest analog to the Bedrock Agents managed experience.
AgentCore Harness reached general availability on June 17, 2026. That's about six weeks before the Classic gate closed - not a lot of runway if you were watching the news casually.
AgentCore pricing is per-agent invocation plus Gateway request charges. The Gateway adds a translation layer that normalizes tool calls across agents, which means fewer custom integrations to maintain but more AWS service charges on your bill. Budget accordingly before moving volume workloads.
What the frozen model catalog means in practice
This is the quiet risk most teams are underweighting. The model catalog available in Bedrock Agents Classic is frozen as of July 30, 2026; new models released after that date will be available through AgentCore.
Amazon Bedrock itself - model inference, Knowledge Bases, Guardrails - continues to receive new models. This restriction applies only to the Bedrock Agents Classic orchestration layer.
Concretely: when Anthropic ships a Claude update or AWS adds a new Titan variant to Bedrock, you can call it via direct inference, but you cannot wire it into a Classic agent as the backbone model. Your Classic agent is frozen to whatever was in the catalog on July 29.
For most teams running stable production workloads, that's tolerable for now. For teams where model capability drives product quality - coding agents, document processing agents, anything that benefits from each new model generation - it's a meaningful constraint that compounds over 12-18 months.
The pragmatic path: keep Bedrock Agents Classic running in maintenance mode, build new features on AgentCore, and migrate incrementally as Classic reaches end of support. No hard deadline has been announced, but AWS typically gives 12 months of maintenance before full deprecation.
The EU enforcement collision that raises the stakes
The timing here is genuinely awkward. Most EU AI Act obligations for high-risk AI systems apply from August 2, 2026. For enterprises running AI agents and automated decision workflows, readiness means demonstrable runtime controls: risk management in operation, human oversight with intervention capability, automatic logging, transparency for deployers, and continuous policy enforcement - not pre-deployment documentation alone.
Key obligations taking effect in August 2026 include full requirements for high-risk AI systems - spanning risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy, robustness, and cybersecurity - plus deployer obligations, conformity assessment procedures, and post-market monitoring.
Worth being precise about what actually changed: the co-legislators agreed on a fixed timeline for the delayed application of high-risk rules - the new application dates are December 2, 2027 for stand-alone high-risk AI systems and August 2, 2028 for high-risk AI systems embedded in products. So if your agent operates in a domain covered by Annex III - employment decisions, credit scoring, critical infrastructure - the obligation clock moved out, not in.
What did not move out: GPAI obligations have technically applied since August 2, 2025, but without enforcement teeth. The Commission's power to investigate and fine arrives August 2, 2026. The rules already existed; the bite is new.
Supervisory review will expect attributable audit trails and evidence that agent and tool actions stay within approved risk bounds over the system lifetime. AgentCore's built-in Observability layer - traces through CloudWatch plus dedicated observability tooling - is structurally better positioned for this than Classic's console-based traces, which were described as "crude but one click away."
Beagle in action#platform-eng, post-migration planning docAmazon Bedrock AgentCore migration: common questions
What happens to my existing Bedrock Agents Classic agents?
Nothing, immediately. Only the Amazon Bedrock Agents Classic service is no longer open to new customers as of July 30, 2026. Your existing Amazon Bedrock models, Knowledge Bases, and Guardrails are not affected. Your agents keep running. The risk is longer-term: no new models, and an eventual end-of-support date AWS has not yet announced.
Is AgentCore harder to use than Bedrock Agents Classic?
For simple use cases, the Harness path is comparable in effort. Use the Harness unless you have a specific reason to own the loop yourself - for example, an existing agent codebase to migrate, or advanced orchestration the Harness does not yet express. The Runtime path requires more code but gives you full control over orchestration logic.
Does the EU AI Act apply to agents built on Bedrock?
It depends on what the agent does, not which infrastructure it runs on. The high-risk system classification includes AI systems used for biometric identification, critical infrastructure, education, employment, access to essential services including credit scoring and insurance, law enforcement, migration, and administration of justice. Internal IT automation and customer support agents in most B2B SaaS contexts fall outside that list.
Can I use AgentCore with models outside Amazon Bedrock?
Yes. AgentCore Runtime is a secure, serverless runtime purpose-built for deploying and scaling dynamic AI agents using any open-source framework including LangGraph, CrewAI, and Strands Agents, any protocol, and any model. You are not locked to Bedrock foundation models when you own the loop via Runtime.
Should I migrate now or wait?
Not immediately - your existing workloads are unaffected. However, AWS recommends evaluating AgentCore for new agent development and future migrations. The practical trigger to migrate sooner: if your Classic agent's quality depends on getting newer models, the frozen catalog is already costing you.