Hatsu | Ai
Region: US
Wednesday 12 August 2026 18:53:34 GMT
10965
538
147
174
Music
Download
Comments
Echoflux :
I'm building a MAS utilizing a combination of the insights of Stafford Beer's VSM and of Holocracy.
2026-08-16 20:29:54
2
a007frankwhite :
i am glad i found you but found you last night after changing from hermes to deepseek had just linked my ollama then learn about FREE TOKEN which would allow me to actually run the new HUGE GLM MODEL AND NOW THIS ITS LIKE ILL NEVER FINISH BUILDING MY SETUP
2026-09-07 18:38:23
1
michaelgannon :
Use a file system based communication method between agents. Treat them like engineers with tasks, handoffs and run as many as you like
2026-09-04 02:01:32
1
Charles Streicker :
I like this but I need to figure out how to do this
2026-08-26 20:27:17
1
replicant432 :
still sub optimal. there is a better way
2026-09-05 16:02:00
1
thechrislindsayclegg :
I run 40 sessions through 1 session and barely hit max limits 😂😂😂 wym
2026-08-20 07:49:46
1
AlamIndah :
Though this might work, it still burn your tokens because it works against prompt caching
2026-08-13 03:43:32
1
Arsenicx2 :
What are you talking about? I can have 1 agent run asynchronous and use multiple LLMs on different servers with no bottle neck. While spawning sub agents. Sounds like the Stanford researchers forgot what running in parrell is.
2026-08-13 09:59:27
3
spine influencer :
Langraph?
2026-08-13 12:34:17
1
Sweaty Pants :
Lol This is exactly what I built . Structure instructs
2026-08-13 08:07:15
1
Vchip :
I need implementation details
2026-08-12 20:08:27
1
Ben the Magnificent :
OK, but we need to get away from these videos and talk about how to actually set that up. I’ve been trying to set something up like this now for several weeks and still can’t get to run smoothly.
2026-08-12 23:29:56
2
The AI Syndicate :
I hate posts like this, "You're doing it wrong!"... Because a paper came out yesterday. Do you know if people have a data pruning tool in their system? Do you know if they are using deferred tool laoading? You make it sound like one paper comes out and everyone is invalidated. AI systems are like fingerprints right now, but just because this paper comes out, we're all doing it wrong. give me a break. And hey, maybe my system IS wrong, but I've already thought about ever growing files, context, and memory. So maybe it ain't.... oh wait... that paper said I am...
2026-08-12 22:34:19
3
Cory Bohn :
I don’t know why you just don’t give the agents separate obsidian vault to work with and communicate with each other
2026-08-21 07:20:04
1
Primetalk GottePåsen 🇸🇪 :
Agentic mesh-organism
2026-08-12 19:14:55
1
𝘉𝘉𝘓 𝘋𝘙𝘐𝘡𝘡𝘠 𓅓☔️ :
im currently building one for my platform and I have 0 clue what im doing... im currently looking for a cto
2026-08-14 08:13:37
1
Sally Sunshine :
So do you put each agent in its own project folder?
2026-08-13 02:15:16
1
W3zton99 :
I don't think we need Stanford for this
2026-08-19 09:23:18
1
benjamin :
well done
2026-08-12 21:43:10
1
. :
can you provide the link to the study please?🙏
2026-08-14 05:45:29
2
Jeff.w :
soo a git repo? the same way real devs work?
2026-08-24 14:08:24
1
devos0820 :
If your orchestrator is running as a full context central manager reading every single raw output, thought trace, and payload from 8+ agents in real time, the flaw isn't the architecture pattern it's how you built it but who builds like that 🤦🏽♂️a true multi agent dashboard needs a orchestrator thats the creators jarvis
2026-08-14 07:55:43
1
Mark Riechers :
I love that this also explains why top-down management doesn’t work
2026-08-14 05:14:14
3
3x+1 :
You should never depend on a single agent.It should always be hot swappable for any local agent or a p.I call agent.🤔 that's how I do it at least, agents are scoped with boundaries and approved tasks. intent is the only sys call. everything tracked.Every agent has to be approved for every response based off of certain criteria.. approved blocked denied or quarantined. no real world action is implemented unless authentication and approval are met.
2026-08-12 19:25:10
1
JP4NY :
Just need to be more intentional about the context of your top level agent right?? Top dude runs show, only job is to delegate out work to series of orchestrators running concurrently all on the same rules… eliminates worktree headaches and no context bloat as long as roles are clearly defined… kinda like you’d want in a legit organization… I got my C-suite, you guys seem like you are trying to optimize a LAN party
2026-08-14 04:35:09
1
To see more videos from user @hatsunama, please go to the Tikwm
homepage.
