This is great! Better agent frameworks are sorely needed in a more diverse set up languages beyond Python and TS. Thank you for the work.
reactordev 221 days ago [-]
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.
0x457 221 days ago [-]
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.
justincormack 220 days ago [-]
This framework uses Goal Oriented Action Planning to plan based on typed outcomes, which is interestingly different from other frameworks.
manishsharan 220 days ago [-]
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.
breun 219 days ago [-]
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.
0x457 215 days ago [-]
Sure, but what it provides can be built quickly if purpose built on top of Spring AI (i.e. your use case and not a framework)
Rendered at 06:05:07 GMT+0000 (Coordinated Universal Time) with Vercel.
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.