Spring Creator Rod Johnson Launches Embabel Java AI Agent Framework

The creator of the Spring Framework has entered the AI agent arena with a new open-source project designed to bring enterprise-grade predictability to generative AI workflows. Rod Johnson unveiled Embabel at Microsoft's JDConf developer conference, positioning the framework as a Java-native alternative to Python-dominated tools like LangChain and Crew.ai.

Embabel is Apache-licensed and hosted on GitHub, built directly on top of Spring Boot. The framework is written in Kotlin but maintains full Java interoperability, meaning developers already embedded in the Spring ecosystem should find the transition frictionless. Johnson's stated goal is straightforward: demonstrate that Java isn't inferior to Python-based solutions for agent systems, particularly in enterprise environments where reliability matters more than novelty.

According to The New Stack, Johnson launched Embabel in May 2025. The project currently employs over five full-time engineers and has accumulated more than 3,000 GitHub stars. That's respectable traction for a framework still in its early commercialization phase, though it remains to be seen whether enterprise adoption will match the GitHub enthusiasm.

The core technical differentiator is Embabel's use of GOAP—Goal-Oriented Action Planning—a non-LLM pathfinding algorithm. Here's the problem Johnson is trying to solve: pure LLM-driven agents produce unpredictable outputs (fine for chat, terrible for invoicing), while fully prewritten code lacks flexibility. GOAP sits in the middle. If you define possible action units, the system finds the optimal path to meet a goal and can explain why it chose that path. This enables verification and auditing, which is non-negotiable for most corporate IT departments.

Embabel supports multiple LLMs by default, including OpenAI, Anthropic, and Meta Llama. Different models can be assigned at each workflow step, and the framework integrates with coding agents like Claude Code and GitHub Copilot. The physical reality of this means developers aren't staring at a black box—they're configuring typed interfaces, watching build logs, and debugging actual stack traces instead of hoping a prompt works.

James Governor, co-founder of industry analyst firm RedMonk, noted in a blog post that Embabel brings determinism to project planning by using a non-LLM model for the plan itself, then using autonomous agents to generate code that maps to that plan. Not everything is decided by LLM. That distinction matters when you're building systems that process financial transactions or manage customer data.

Simon Ritter, deputy CTO at Java runtime provider Azul Systems, told The New Stack that increasing availability of Java frameworks could help close the widening gap between Python and Java usage in the AI agent space. The JVM has decades of production-hardened infrastructure behind it. Python's AI ecosystem is vibrant, but enterprise Java developers have spent years building resilient, distributed systems. Embabel attempts to bridge those worlds without forcing developers to abandon their existing toolchains.

Chad Arimura, formerly vice president of developer relations at Oracle, mentioned in a briefing about Java 26 that the latest version supports the "natively agentic" nature of Embabel. Java 26 was released last month, which suggests the language itself is evolving to accommodate these new workload patterns. Whether that translates to actual developer adoption is another question entirely.

Johnson's keynote at JDConf 2026, alongside Bruno Borges and Ayan Gupta from Microsoft, explored where the Java ecosystem is headed in the agentic era. The conference documentation emphasizes practical themes: connecting agents to enterprise systems with clear governance, frameworks adapting to AI-native workloads, and how testing pipelines evolve as automation gets more capable. This isn't theoretical speculation—it's about shipping code that won't break production.

One workshop at the event covered building durable, production-ready agents with Spring AI and Temporal. The session addressed real distributed systems problems: How do agents persevere through flaky networks? How do they function when LLMs are rate limited? How do they run for hours, days, or weeks when infrastructure is rarely stable that long? Embabel doesn't solve all of these alone, but it provides the foundation for addressing them systematically.

Johnson has been clear about his positioning. He doesn't regard himself primarily as a Java person, but rather as someone who wants to solve problems in enterprise software. In the early 2000s, the problem was productivity with Java. In the mid-2020s, the problem is making GenAI relevant to business applications. The framing is pragmatic, which is refreshing given how much AI coverage focuses on hype cycles rather than actual deployment challenges.

The commercial model is planned to be similar to the SpringSource approach—open source core with commercial support and enterprise features. That's a proven model in the Java ecosystem, though it requires building a sustainable business around community adoption. Five engineers is a lean team for what's essentially competing against well-funded Python frameworks with larger developer communities.

Whether Embabel gains significant traction depends on whether enterprise Java developers actually need a dedicated agent framework or if existing Spring AI capabilities suffice. The framework solves a real problem—predictability in AI workflows—but it also adds complexity to an already crowded tooling landscape. Developers will have to weigh the benefits of GOAP's structured planning against the overhead of learning and maintaining another abstraction layer.

For now, the code is available on GitHub and the framework is Apache-licensed. That means anyone can inspect, fork, or build on top of it. Whether organizations actually deploy it in production, or whether it becomes another well-intentioned project that gathers dust in a developer's local repository, remains to be seen. The technology is sound, but adoption is always the harder problem to solve.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn