Amazon Bedrock AgentCore review: AWS's enterprise agent platform for any framework and any model

Answer in brief: Amazon Bedrock AgentCore is AWS's platform for running production AI agents, summed up by AWS as "Any framework. Any model. Secure at scale." On our market map it is a hyperscaler platform. It is a set of modular, separately billed services (Runtime, Gateway, Identity, Policy, Memory, Browser, Code Interpreter, Web Search, Web Answer, evaluations) that you can use alone or together. It is the strongest fit for AWS developer teams that already have an agent framework. Pricing is pay-as-you-go with no minimum fees.

Top pick for AWS engineering teams. Bring LangChain, the OpenAI Agents SDK, the Claude Agent SDK, Strands or your own framework. Per-second microVM billing that stops CPU charges during I/O wait. Visit Amazon Bedrock AgentCore →

Independent research, last checked November 9, 2026. We have not tested this product hands-on; facts come from AWS's own pages.

Where it sits on the market map

AgentCore is pure infrastructure: the runtime and control layer underneath agents you write. It is the natural home for teams using open-source frameworks, which is the fourth type on our market map, without running their own servers.

What it does for enterprise agents

Runtime: a secure, serverless runtime with two compute types. MicroVMs are serverless and billed by consumption with hardware-enforced isolation. Instances run on AWS-managed EC2 in your own account, with sessions lasting up to 14 days.
Gateway: secure access to tools, models and other agents (MCP servers, APIs, Lambda functions), with authentication and access control handled for you.
Identity: agent identity and access to AWS and third-party resources on behalf of users. There is no extra charge when you use it through Runtime or Gateway.
Policy: checks every tool call through Gateway against your rules, with natural-language authoring that converts to Cedar policies.
Memory, Browser, Code Interpreter, Web Search and Web Answer: context-aware memory, cloud browser sessions, sandboxed code execution and grounded web answers with citations.
Debug and optimize: see the steps an agent took and run evaluations against real traffic.

Who it fits

Developer teams on AWS that want managed, secure agent infrastructure and the freedom to change frameworks or models without rewriting it.

Strengths that win deals

Framework- and model-agnostic by design.
CPU charges stop during I/O wait (waiting on LLMs, tools or databases), and Runtime v2 reclaims idle memory after 120 seconds.
EC2 Savings Plans and Reserved Instances apply to Instances compute.
Security policies are enforced at the platform layer and verified by automated reasoning, which AWS says is the technology behind IAM and S3.

Honest drawbacks

Built for developers, not a no-code builder for business users.
Many separately metered parts make cost modeling an exercise.
Models, storage (ECR, S3, EBS) and data transfer are billed separately.

Pricing model

No upfront commitment or minimum fees. New AWS customers get up to $200 in Free Tier credits.
Runtime microVMs, Browser and Code Interpreter: per-second CPU and memory consumption, 1-second minimum, 128MB minimum memory billing.
Runtime Instances: EC2 cost plus a management fee calculated as a percentage of the EC2 On-Demand price.
Web Search: $7 per 1,000 queries. Web Answer: $10 per 1,000 tool calls.
Gateway: per operation, plus $0.006 per GB for data egress to customer VPCs.

Customers AWS names

AWS features Cox Automotive (scaled from zero to 17 production agents in under a year), Druva (solves 68% of support issues without human intervention) and Thomson Reuters (70% automation in platform engineering). These figures are as reported by AWS.

How to evaluate AgentCore

Run your existing agent code on Runtime first, then add Gateway and Policy.
Measure I/O wait in your agent loop, because that is where consumption billing saves money.
Budget model tokens separately; see bestinferenceproviders.com for model-hosting comparisons.

Best alternatives

Google Gemini Enterprise Agent Platform: the Google Cloud equivalent, with a low-code studio.
OpenAI Agents API: a managed harness if you are standardizing on OpenAI models.
IBM watsonx Orchestrate: a governed option that also runs on AWS or on-premises.

FAQ

Which frameworks does AgentCore support?

AWS lists LangChain, the OpenAI Agents SDK, the Claude Agent SDK, Strands and your own framework, with any model.

Is there a minimum fee?

No. AgentCore is consumption-based with no upfront commitments or minimum fees.

Can agents run for long periods?

Yes. With the Instances compute type, sessions persist up to 14 days on EC2 in your own account.

Is AgentCore a no-code builder?

No. It is developer infrastructure; business-user builders are better served by suite platforms such as Copilot Studio.

Sources

Visit Amazon Bedrock AgentCore →

← Back to the ranked list of enterprise agent platforms · Pricing comparison