@jakevanclief: I misspoke when I said that semantic Kernal was in C sharp originally it was just in C++ but now they have python as well. I didn't feel like rerecording.

The Real asdfasdfqwe :

yeah, that only works for an individual solo Dev. what you're talking about does not scale organizations of even small startups, two large corpse, but those frameworks you talked about scale very well for what Enterprise companies from the 10-20 engineer Dev team and up. yeah I can see how you have this opinion in your own isolated little. I don't have a job. I'm just a solo open claw user, but you know I think the real world leads itself to meaning a completely different type of semantic structure. but you're right for open claw or a single person's individual AI assistant file system is the perfect approach

calviivsqt0 :

Cat Black :

agents are not a waste of time, you don't have enough computational power to train and use generic type LLM you use that for orchestration. Agent's are your specialized problem solvers and tools are your per problem computational specialists. you can have multiple orchestrators on multiple levels, your tasks and subtasks are Agent's and tools, knowlage bases are connected. It all runs and communicates trough MCP which is a communications protocol (or you can use other communications protocols) Lastly it's not a tree and definitely not a simple tree as edges can go between tools, agent's and orchestrators in asynchronous way due to communications... a proper agentic system is an N dimensional Graph with multi node communication trough MCP... 🙄 If you don't get it you are doing it wrong... the biggest issue with this whole system and anything based on an LLM is that it still is PROBABILISTIC complex (NxM)^K dimensional function minimum finder... it's not smart, it's not reasoning, it's Random Drawing answers from training pool of information and than joining together a Markov Chain that is your answer, than you add more orchestarors and agent's which makes it even more complex but still a probabilistic problem solver. Humans can give repetitive deterministic answers to the same problem, AI it almost never gives the same answer 😏

nerd.chef :

Whiteboard did not help you.

Torrefaction :

terrible advice for operationalizing consistency for any agent. actor critic loops are one of the most powerful things about LLMs

JT Turner :

This is only one part the memory but I feel like you have different parts and determinate tasks should be code saving prompt by letting it write it's own tools. You also need agents to respond to tasks in near real time or on an action of some sort which would also need code. But I agree file trees are the best memory for Claude code and codex cli tools.

teromee :

I do this whole stack naturally because they are just subminds

kelargo :

RAG

Knowledge :

10/10 video I gave a talk on this a few weeks ago

JakeCondemn :

I'm not tech-savvy at all but 22 years in the Army taught me about hierarchy and "org charts" it's a prefect fit for this file system and I'm using something like this. One thing you should add, remove the 260 character limit, file paths, etc.

TBD0x00 :

You have that now with copilot studio

Joe | 🚬 🤖 :

Hell yeah let’s go

TUNG TUNG TUNG SAHUR 🪵🪵🪵 :

GENIUS! you're so FUUUUCKING right mate ♥️

Joseph A P Holsten :

I’m delighted that you’re just arguing for plan 9.

GregUlan :

omg ty for explaining open claw finally! I like this explanation. I could not understand why we need the same ai put on different hats of different experts when it's mostly the same ai.

IndependentFilm.com :

So Romeo is like the carburetor on my Harley. And Juliette is like the handlebars. Got it.

GALLUS :

F L O W I S E 🥰

wildgriffin :

Nice synthesis. Well done.

Paul Weitermachen :

nice 1

Cherry Gardens Cyprus :

God Mode

KalleGrad :

you got it!

track.jpalmer :

can you put this in context of skills and agents as marked down files?

Jay Black :

Agreed.

Markizano Draconus :

Wait "C+"? Are you referring to the the ABCL developed by Japan in the 1980s or C++?? I know you "misspoke"... But missing the correction too??

Ant :

bro talking for so long and I still don't understand what he is trying to say😳😳😳