Skip to main content

November 2025: LangGraph, RAG Failures, and Deep Agents

Welcome to our November 2025 Mixer & Showcase! This month featured practical deep dives on LangGraph fundamentals, enterprise RAG implementations, production deep agents, and AI project management anti-patterns.

Video Recording​

Featured Content​

Thunderstorm Talks​

  1. Introduction to LangGraph

    • Controllability: balancing agent autonomy with developer control
    • Bulk synchronous parallel architecture
    • State persistence and fault tolerance
    • Human-in-the-loop workflows
    • Production use cases in FDA compliance and ERP systems

  2. Why RAG Use Cases Crash and Burn in Enterprises

    • 42% of enterprise AI use cases failed in 2025
    • Five pillars of RAG failure
    • The "goldfish vs elephant" memory problem
    • State management architecture
    • PIMCO's Microsoft-featured success story

  3. Deep Agents in Production - Real World Experience

    • $80, 3.5-hour pipeline analyzing 400 companies
    • Four specialized subagents architecture
    • Middleware as control mechanism
    • The "toddler" approach to agent correction
    • When to use Deep Agents vs LangGraph

  4. AI Project Management Anti-Patterns

    • 95% of AI pilots fail due to organizational approach
    • Legacy SaaS/ERP mental models don't work for AI
    • Framework comparison: Accenture, Google, AWS, IBM
    • Red flags to watch for in AI projects
    • The 6-month notebook problem

Event Details​

November Monthly Mixer & Showcase (November 5, 2025)​

Location: Austin Community College RGC 3000

Attendance: 50+ in-person attendees, remote participants from Panama chapter

Schedule:

  • 6:00 PM - Welcome & Networking
  • 6:20 PM - Introductions & Kickoff
  • 6:30 PM - LangGraph Introduction (Colin McNamara)
  • 6:50 PM - RAG on LangGraph (Anupama Garani)
  • 7:05 PM - Deep Agents: Building on LangGraph (Collier King)
  • 7:25 PM - Project Management for AI (Paul Phelps - Remote from Panama)
  • 7:40 PM - Buffer & Open Q&A
  • 8:00 PM - Walk to The Tavern / After-Party

Key Topics​

LangGraph Fundamentals​

  • Graphs vs chains vs agents
  • Controllability through node-edge structures
  • State persistence in Postgres, Redis, file systems
  • Human-in-the-loop native support
  • Time travel debugging capabilities
  • When to use LangGraph vs React agents

Enterprise RAG Reality​

  • Industry failure statistics (S&P Global survey)
  • Data quality and metadata challenges
  • Prompt engineering evolution
  • Evaluation and feedback loops
  • Governance and compliance requirements
  • State management for multi-turn conversations

Production Deep Agents​

  • Subagent specialization patterns
  • Middleware for agent control
  • Validation tracking and logging
  • Cost analysis: $80 for 11M tokens
  • Performance metrics and success rates
  • LangGraph vs Deep Agents decision matrix

AI Project Management​

  • Non-deterministic vs deterministic software
  • Evolving requirements as learning signals
  • Developer involvement from day one
  • Framework evaluation criteria
  • Organizational barriers to AI deployment
  • Building developer-focused AI frameworks

Community Highlights​

  • International Presence: Remote participation from AIMUG Panama (100 members)
  • Production Focus: All talks featured real-world production experiences
  • Honest Insights: Speakers shared costs, failures, and lessons learned
  • Technical Depth: Advanced topics on architecture, middleware, and state management
  • Community Training: LangChain team provided training materials

Speaker Profiles​

Labs & Resources​

Code Repositories​

Learning Resources​

Prerequisites​

  • Understanding of:
    • Basic LangChain/LangGraph concepts
    • Agent architectures
    • Python development
    • RAG fundamentals
  • Development environment with:
    • Python 3.9+
    • LangChain/LangGraph packages
    • Access to LLM APIs

Key Takeaways​

Technical Decision Matrix​

Use LangGraph when:

  • Deterministic workflows required
  • Compliance/safety requirements
  • Cost efficiency critical
  • Repeatability needed

Use Deep Agents when:

  • Open-ended research tasks
  • Complex multi-step planning
  • Managing large context
  • Long-term memory required

State Management Principles​

  • Context = current conversation
  • Memory = conversation history
  • State = orchestration layer persisting both

Enterprise AI Success Factors​

  1. Developer involvement from day one
  2. Acceptance of non-deterministic behavior
  3. Iterative discovery vs complete upfront requirements
  4. Cross-functional collaboration
  5. New PM frameworks (not legacy SaaS playbooks)
  6. Golden datasets for evaluation
  7. User feedback loops
  8. Audit trails for compliance

Next Events​

  • Tuesday Office Hours: Every Tuesday at 5:00 PM Central on Discord
  • Hacky Hour: Mid-month (check Discord/Meetup for updates)
  • December Showcase: First Wednesday of December

Join us on Discord or visit aimug.org for the latest updates!