Why Do We Need an Agent Framework? • Rod Johnson • YOW! 2025

2026-06-05 · Source: GOTO Conferences · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Advanced, extended

Summary

Rod Johnson's YOW! 2025 talk highlights the disparity between GenAI's success in personal productivity and its struggles in enterprise business processes. While coding agents benefit from user intervention and "git revert" capabilities, enterprise applications lack this fault tolerance for customer interactions, as seen in the Air Canada chatbot case. Enterprise GenAI projects frequently fail due to inherent non-determinism, "prompt engineering as alchemy," and performance costs, alongside organizational issues like top-down mandates and siloed teams. Johnson advocates for breaking down large tasks into deterministic, structured steps, integrating with existing systems, and applying "domain integrated context engineering" that leverages established domain models. He introduces Embabel, a new JVM framework built on Spring, designed to maximize determinism, facilitate integration, emphasize domain modeling, and enable type-safe mixing of code with LLM operations, supporting various models and using GOAP for predictable planning.

Key takeaway

For AI Engineers or Architects building enterprise GenAI applications, you must prioritize deterministic orchestration and structured data exchanges over "god model" approaches. Focus on integrating GenAI with existing JVM systems and domain models to achieve predictability and testability. This strategy, exemplified by frameworks like Embabel, mitigates risks of non-deterministic, text-only outputs and ensures robust, maintainable solutions, avoiding common enterprise failures.

Key insights

GenAI's enterprise adoption requires deterministic orchestration, structured interactions, and integration with existing systems, unlike personal productivity tools.

Principles

• GenAI for business needs predictability; personal use tolerates non-determinism.
• Organizational best practices remain crucial for new technologies.
• Deterministic step planning is vital for reliable GenAI applications.

Method

Break large GenAI tasks into smaller, structured steps returning data classes (e.g., Pydantic models), not just text. Use deterministic orchestration (e.g., GOAP) and integrate with existing domain models and code.

In practice

• Use type-safe frameworks like Embabel for JVM GenAI.
• Implement JSR 380 annotations for LLM input validation.
• Track LLM prompt actions and costs for process visibility.

Topics

• Agent Frameworks
• Enterprise GenAI
• JVM Development
• AI Orchestration
• Deterministic AI
• Embabel Framework

Best for: Machine Learning Engineer, CTO, VP of Engineering/Data, Software Engineer, AI Engineer, AI Architect

Related on AIssential

• Presentation: Powering the Future: Building Your GenAI Infrastructure Stack — Intuit's GenOS platform and "fixed, flexible, free" framework accelerate enterprise-scale AI agent development through structured governance and continuous evaluation.
• Architecting Production-Grade Agents through LLM Orchestration and Agentic Loops — Reliable AI systems require deterministic agentic architecture over prompt-driven LLM experimentation.
• SAP recasts Joule as the front door to autonomous enterprise AI — SAP is transforming its Joule AI assistant into the core engagement layer for an "autonomous enterprise" vision.
• LangGraph vs CrewAI vs AutoGen: Which AI Agent Framework Should Your Enterprise Use in 2026? — Choosing an enterprise AI agent framework depends on production needs, development speed, and ecosystem integration.
• The Missing Half of the Enterprise Context Layer — AI agents require operational context from orchestrators to trust data, beyond just semantic definitions.
• Why GenAI Isn’t Ready for Prime Time — The author contends that Generative AI (GenAI) is not yet ready for "prime time" and will not replace human jobs soon, citing its immaturity and inherent limitations despite significant hype.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by GOTO Conferences.