What this is
Twenty-three years ago Rod Johnson wrote a book about J2EE, then open-sourced Spring and pulled enterprise Java out of the EJB quagmire. Now he is back with a new framework, Embabel, betting on something bolder: enterprise Agents should not let LLMs decide the next step.
Embabel is not a new model, and it does not replace Spring AI. It is an orchestration layer built on top of Spring AI—one handles wiring models and vector stores, the other handles action sequencing and workflow controllability.
The most counterintuitive design choice: do not let the LLM decide the action sequence. Embabel borrows GOAP (Goal-Oriented Action Planning) from game AI—the same algorithm F.E.A.R. used for NPC decision-making back in 2005. Developers declare goals and actions; a planner uses deterministic algorithms to compute the action sequence; the LLM only works inside each action. In Rod Johnson's own words: given the same starting state and goal, it generates the same plan every time—unlike LLMs, where you cannot explain a thing.
The framework is 95% Kotlin, but Java developers writing business logic never need to touch a .kt file.
Industry view
Supporters see the real pain points of enterprise Agents: mainstream frameworks (LangChain, AutoGen, and the rest) look slick in demos, then fall apart in production—actions orchestrated by LLM hallucination, workflows unexplainable, errors untraceable. Finance, telecom, and government cannot tolerate 10% randomness. Rod Johnson calls this "the most certain project since founding Spring"—no small statement from the man himself. And Spring's twenty-plus years of accumulated gravity in the enterprise Java ecosystem dramatically lowers migration cost for existing users.
Opposition is worth hearing too: the AI Agent main battlefield is Python, and Embabel trying to take share from Python is no easy feat; the framework just shipped, with zero production validation; "deterministic planning" carries a cost—it sacrifices the flexibility and autonomous exploration that LLM Agents offer; and Rod Johnson has been largely out of the frontline Java community since stepping back from Pivotal in 2017, so his current actual influence is an open question.
Impact on regular people
For enterprise IT: Traditional industries with deep Java stacks—banking, insurance, telecom—now have a relatively controllable path for landing Agents. But the framework just shipped; the journey from evaluation to production will take at least two to three years, and incumbent systems will not be swapped out overnight.
For individual careers: Java engineers have not been left behind by the AI wave—they now have one more direction worth tracking. But engineers who stick to pure Java and never touch Python will still need to catch up over the long term: most frontier innovation is still happening inside the Python ecosystem.
For the consumer market: No direct short-term impact. Things built on Embabel will reach us through enterprise services (bank call centers, insurance claims, government hotlines will become more stable), but the consumer AI products we use every day remain firmly in the Python camp's hands.