A common challenge in AI-powered research workflows is depth versus context. If your agent reads ten web pages, its context window (the amount of text a large language model (LLM) can process at once) gets filled with raw content. If it also runs data analysis code, chart-generation logic competes with strategic reasoning for limited space. Teams typically work around this with manual prompt-chaining or sequential processing.

A better approach is to delegate deep work to isolated subagents that return only concise results. LangChain Deep Agents handles this orchestration, spawning specialized ephemeral subagents and managing their lifecycle. Amazon Bedrock AgentCore provides the infrastructure each subagent needs. This includes a real browser in a MicroVM (a lightweight, single-purpose virtual machine) for web research and a full Python environment for data analysis. AgentCore is also available as a native sandbox provider in the Deep Agents CLI, so you can run deepagents --sandbox agentcore to try AgentCore CodeInterpreter without building a full agent.

In this post, you’ll build a competitive research agent that demonstrates this pattern end to end. This walkthrough targets developers building multi-step AI workflows who need isolated execution environments for their agents. In Part 2 of the notebook, you can deploy this same agent to Bedrock AgentCore Runtime using the AgentCore CLI, so it runs as a managed, session-isolated service.