Amazon Web Services has introduced runtime instances for Amazon Bedrock AgentCore, a new compute option designed to support persistent infrastructure for AI agents handling complex, long-running workflows and multi-agent coordination.
The update, announced by AWS, extends AgentCore’s capabilities beyond its previous stateless execution model. Runtime instances allow agents to maintain state across multiple steps, which is critical for tasks that require continuity and context over extended periods.
This development is part of AWS’s ongoing effort to enhance its generative AI services for enterprise use. AgentCore, a component of Amazon Bedrock, manages the lifecycle of AI agents, including their configuration, invocation, and monitoring. Previously, AgentCore operated in a serverless fashion, where each invocation ran in isolation, limiting its suitability for complex, stateful applications.
Key Features of Runtime Instances
Runtime instances provide dedicated compute resources that persist between invocations. This allows agents to retain memory, maintain conversational context, and coordinate with other agents in a multi-agent system. The persistent nature of these instances is intended to reduce latency and improve efficiency for workflows such as automated reasoning, multi-step planning, and tasks that require iterative decision-making.
According to AWS, the new compute option is purpose-built for production-scale deployments that demand reliable, long-running execution. It is also designed to simplify the development process by eliminating the need for developers to manage underlying infrastructure.
Implications for Enterprise AI
Analysts note that the introduction of persistent compute addresses a common limitation in agent-based AI systems: the inability to maintain state over time. By offering runtime instances, AWS aims to make AgentCore more competitive for business-critical applications, including customer service automation, supply chain optimization, and complex data analysis.
The service is expected to benefit organizations that require multi-agent collaboration, where multiple AI agents must share information and coordinate actions to achieve a common goal. This capability is increasingly relevant as enterprises adopt AI to automate more sophisticated workflows.
AWS has not disclosed specific pricing for runtime instances, but it is anticipated that they will follow a consumption-based model similar to other AWS compute resources. Users can expect the new option to become available in the coming months, with a gradual rollout across AWS regions.
Developers and organizations using Amazon Bedrock AgentCore can begin testing runtime instances in preview, with general availability expected later this year. AWS has indicated that it will continue to expand the service’s features based on customer feedback and evolving AI requirements.
