Deep Agents Architecture | Collier King | AIMUG September Showcase

Why “shallow agents” struggle with sustained context, planning, and complex workflows

How Deep Agents extend agent capabilities with:

✅ Planning tools for long-term task tracking ✅ Sub-agents to quarantine context and specialize behavior ✅ System prompts (reverse-engineered from Claude Code) ✅ A virtual file system for state management, parallelization, and delegation

Built-in tools for to-do lists, file creation/editing, and coordination between agents.

Real-world demo: a deep research agent analyzing marketing material vs. social media to uncover customer personas and gaps.

Growth trajectory of the Deep Agents package and how it leverages LangGraph under the hood.

✨ Why it matters: Deep Agents bring structure and scalability to agent design, making them better suited for research-heavy workflows, delegation, and complex pipelines than simple tool-loop agents.