Amazon Web Services announced the new AgentCore runtime, a capability of Amazon Bedrock AgentCore that serves as a managed compute layer for deploying and running agents without building or maintaining infrastructure. Since launch, thousands of teams have used it to run production agents. The new runtime is purpose-built for speed, flexibility, and cost efficiency, offering better memory management that reclaims memory as a session releases it instead of holding it at the peak, and consistent cold start times regardless of container size or agent concurrency. It retains the serverless model, with memory released back the instant a session ends, startup times staying consistent regardless of size or concurrency, and billing that tracks the work an agent does. The first version provided serverless operation, session isolation, scale to zero, and pay-only-for-what-you-use. The new runtime extends that foundation across the full spectrum of agents, staying fast and consistent for interactive agents and durable and affordable for long-running, more autonomous agents. With AgentCore runtime, customers pay only for what they consume, not for idle CPU waiting for I/O, with billing following resource usage and no standing charge for provisioned capacity; the platform scales all the way down to zero, so when an agent isn't handling work, nothing runs and nothing is paid for.