i've been using loads of ai agent frameworks (crewai, langgraph, autogen) and i just don't get why they are so popular. i can literally do everything what they offer spending a few hours with cursor, and better. if i want easy agents i want a platform at a higher abstraction layer where agents are literally ai generated for me and offer robust scalable methods around it like agent workflows, not frameworks that i can generate the code for with my ai. i've been looking at two agent platforms phidata (although still requires quite some coding and workarounds) and omega . ai (more do it for you kind of platform). both pretty impressive and getting some serious agents up, now have a marketing agent doing my daily blogs, social media posts and twitter replies. what are some other non-framework ai agent platforms you use? looking to try out more
bsenftner|1 year ago
Case in point: I don't use any frameworks whatsoever. I wrote a conversational AI Agent that helps me write AI Agents, integrated that into an office suite, unleashed that at the law office where I'm CTO, and we currently have just shy of 900 agents created by me, the attorneys and their staff. They support new client interviews, legal research, document authoring, case financial modeling, and pretty much whatever the staff needs to help them do what they already did before AI, just now they have idiot savant help.
Everything is based upon chat completion with most using structured output that is I/O with the office suite's internal data structures. The AI Agents act as virtual co-workers inside the office software used by staff, and they each personalize their agents to their needs.
If I tried to do this with these frameworks, I'd still be dinking around with their abstractions and lack of documentation.
2024user|1 year ago
lunarcave|1 year ago
Most "agent frameworks" are just workflow builders with LLMs as a node in the workflow. Zapier can do most of what "agent frameworks" do with having OpenAI / Anthropic as a node. IMHO, to be an "agent", you need agency. And a lot of that agency has to do with having control over the control flow. (Design its own graph at runtime)
Agents with true agency - that is, generating the control flow at runtime is a hard problem (we've spent months into it - and it's still not generalizable). We've got a long way there with good old engineering (deterministic guardrails, progressively enriching context etc). (This is on top of the usual distributed systems problems - for example - solving for invoking tools idempotently / at least once delivery etc)
But the path of least resistance here is claim "How an AI agent 3x'd my inbound and mowed my lawn" and make a youtube video about how agents are going to take over the world.
thiago_fm|1 year ago
Maybe they eventually find a way, but right now it's worthless imho.
The updates to the foundational models is what makes a big difference.
xzzzzzz|1 year ago