Re:Embabel Agent Framework for the JVM (by the founder of Sp...
by reactordev on Tuesday July 01, 2025 @11:44AM
(#44435069)
This is awesome work! It bridges the business gap between “We’re a Java shop” and “Here’s my Jupyter notebook of my agent”. Looking forward to seeing where this goes.
Re:Embabel Agent Framework for the JVM (by the founder of Sp...
by 0x457 on Tuesday July 01, 2025 @12:56PM
(#44435860)
IMO these agent frameworks do very little in terms of value add. Main thing they do is handling tool calling and some prompt wrapping and maybe naive chat history managment.
If a Java shop can figure how to do that in one day, then it's not a Java shop or they lack context around LLMs and framework won't bridge knowledge gap.
Re:Embabel Agent Framework for the JVM (by the founder of Sp...
by manishsharan on Tuesday July 01, 2025 @03:21PM
(#44437184)
This 100%.
Frameworks like these, and I include Microsoft Semantic kernel and Google' Agent2Agent, just get in the way. Fact is that Agent development is still in early stages. We daily learn more about how to get the most out of LLMs. It used to be "prompt engineering", this morning HN is talking about "Context engineering".
Abstractions and frameworks like these will obfuscate insights one would gain by using LLM +tools etc directly.
Re:Embabel Agent Framework for the JVM (by the founder of Sp...
by breun on Thursday July 03, 2025 @12:29PM
(#44456716)
What you're describing sounds more like what Spring AI provides for Spring applications. Embabel builds on top of Spring AI with higher-level abstractions that go beyond this. It's still early days for Embabel, but the approach seems interesting to me, so I'll be keeping an eye on it.
If a Java shop can figure how to do that in one day, then it's not a Java shop or they lack context around LLMs and framework won't bridge knowledge gap.
Frameworks like these, and I include Microsoft Semantic kernel and Google' Agent2Agent, just get in the way. Fact is that Agent development is still in early stages. We daily learn more about how to get the most out of LLMs. It used to be "prompt engineering", this morning HN is talking about "Context engineering".
Abstractions and frameworks like these will obfuscate insights one would gain by using LLM +tools etc directly.