Deep Agents with LangGraph: From Planning to Persistent Reasoning | Community Webinar
•December 11, 2025
0
Why It Matters
Deep Agents enable enterprises to deploy AI that can plan, track, and iteratively refine complex tasks, reducing hallucinations and scaling beyond the limits of traditional single‑shot agents.
Summary
The webinar introduced Deep Agents built on LangGraph, positioning them as the next evolution in multi‑agent AI systems. Presenter Sajir Heather Zaddi, a senior software engineer specializing in LLM fine‑tuning and agentic workflows, framed the discussion around a recent tweet by Andrew Ng that AI agent workflows will outpace foundational model advances this year. The session promised a blend of theory and a live demo, targeting developers with basic Python and LangChain knowledge.
Zaddi outlined the shortcomings of traditional agents—single‑shot prompting, limited context windows, and an inability to decompose complex tasks. He argued that task complexity is doubling roughly every seven months, creating a bottleneck for shallow agents that cannot plan, track progress, or manage large contexts. Deep Agents address these gaps through four pillars: a detailed system prompt, an integrated planning tool that breaks goals into discrete steps and monitors status, specialized sub‑agents that isolate context for distinct functions, and a file‑system interface that offloads data to persistent storage, mitigating context overflow.
Illustrative examples included a comparison to OpenAI’s Deep Research workflow, where a query triggers a multi‑step plan, iterative tool calls, and citation‑rich output. Zaddi demonstrated how sub‑agents can separately handle research, literature review, and markdown conversion, each operating on isolated file contexts to prevent cross‑contamination. The planning tool’s progress tracker was highlighted as a safeguard against hallucinations, ensuring each sub‑task is completed before moving on.
The implications are clear: Deep Agents promise more reliable, human‑like problem solving for enterprise applications such as insurance claim analysis or educational content generation. By embedding planning, context isolation, and persistent storage, they aim to reduce hallucinations, improve scalability, and enable longer‑running, multi‑step workflows that were previously untenable with shallow agents.
Original Description
Understanding Deep Agents: The Future of AI Autonomy
Explore how Deep Agents take AI beyond simple tool-calling into persistent, stateful intelligence. This session breaks down the core pillars of Deep Agents and how LangGraph enables scalable, production-ready autonomy.
Key Takeaways:
• Why traditional agents fail on long, complex workflows
• Core pillars: planning, sub-agents, memory/state, and virtual file systems
• How LangGraph delivers human oversight, scaling, and error recovery
• Practical patterns for building reliable, multi-step AI systems
• Real-world use cases from research to decision support
#DeepAgents #AgenticAI #LangGraph #AIAutonomy #MultiAgentSystems #AI
-------
Table of Contents:
0:38 – Welcome & Opening Remarks
2:38 – Speaker Introduction
4:22 – Introduction to Deep Agents
6:50 – Limitations of Traditional Agents
8:18 – What Are Deep Agents?
10:36 – The Four Pillars of Deep Agents
31:16 – Deep Agents Live Demo
50:45 – Audience Q&A
1:01:35 – Agentic AI Bootcamp Overview
1:03:35 – Closing & Final Notes
---------
👉 Learn more about Data Science Dojo:
👉 Explore video tutorials:
👉 See community feedback and success stories:
At Data Science Dojo, we believe data science is for everyone. Our in-person and virtual training programs have helped 10,000+ professionals from 2,500+ companies — including Microsoft, Apple, and Meta — apply AI responsibly and effectively.
🔗 Subscribe to our newsletter for more AI tutorials and events:
0
Comments
Want to join the conversation?
Loading comments...