AWS: Amazon Bedrock AgentCore Harness Reaches General Availability — Deploy in 2 API Calls
Amazon Bedrock AgentCore Harness has exited preview and is now available to everyone in production. Deploying a production AI agent now requires only 2 API calls, and the platform supports Claude, Nova, Llama, DeepSeek, GPT-5.5, and GPT-5.4 with the ability to switch models mid-session without losing context.
This article was generated using artificial intelligence from primary sources.
Amazon Web Services has announced general availability (GA — meaning the product has exited preview and is now available to everyone in production) of the Amazon Bedrock AgentCore Harness platform. The service enables creating and running a production AI agent with just 2 API calls: CreateHarness for initialization and InvokeHarness for execution. Compared to earlier, more complex orchestration approaches that teams had to build themselves, this represents a significant reduction in setup complexity.
Supported Models and Mid-Session Model Switching
AgentCore Harness from its GA version supports a wide range of models: Anthropic’s Claude (including Opus and Sonnet lines), Amazon Nova, Meta Llama, DeepSeek, and the newly added OpenAI GPT-5.5 and GPT-5.4. A key technical feature is the ability to change model providers within the same session without losing context — an agent can, for example, start with Claude Opus, switch to GPT-5.5 for a specific subtask, then continue with Gemini, while the platform transparently transfers the full conversation context. This capability was previously only available with custom orchestration logic.
Pricing and Billing
Billing is based on actual resource consumption (active-consumption model): $0.0895 per vCPU-hour and $0.00945 per GB-hour of memory. AWS emphasizes that charges are incurred only when the agent is actively executing tasks, not during waiting or idle periods.
What This Means for Development Teams
AgentCore Harness aims to remove the infrastructure complexity that previously slowed down production deployment of AI agents. Teams that previously had to build their own orchestration logic, context management, and load balancing between models now get this as a managed service. The two API operations — CreateHarness and InvokeHarness — abstract details that previously required significant engineering investment. The combination of a model-agnostic approach and mid-session switching capability is particularly relevant for complex multi-step agents that use different models for different task phases.
Frequently Asked Questions
- What does GA (general availability) mean for AgentCore Harness?
- GA means the product has exited preview or beta and is now available to all AWS users in production environments with full support, SLA guarantees, and standard billing.
- Can an agent switch model providers without losing context?
- Yes — AgentCore Harness enables mid-session model switching, for example from Claude Opus to GPT-5.5 or Gemini, while the platform transparently maintains the full conversation context without requiring session reinitialization.