parthknowsai
Region: US
Thursday 09 April 2026 12:16:43 GMT
38932
2848
77
476
Music
Download
Comments
david :
Can i get a meta harness for my life
2026-05-03 15:33:49
1
reed :
Great points you made there. For me, the struggle was always the execution part. Workbeaver has been a huge help in getting my projects across the finish line. I’m seeing way better results now that I’m not spread so thin. It’s basically the only way I’ve stayed sane while trying to scale up.
2026-05-11 17:21:59
60
Vicky :
I built my own harness. Did it AI native and AI first or whatever the term is. I only hand coded a little bit. The agents themselves monitor and fix/build the rest. Anyway, here’s the best advice someone who has been there can give: build for a human. If a human can use it, an LLM can definitely use it, even a tiny model. Also whenever you get a hallucination, that’s what I call a model desire line. It’s evidence of a model reaching for something, failing to find it, and making crap up as a result. To fix it, build the missing tool.
2026-04-09 16:17:41
7
airforce guy :
you need to train your agents brain(i.e rag memory on good code examples and then the ai can generate up to 95% good code based on the training), i did that on my own agent builder software and it worked on some complex workflows(i.e pipelines that have agents and they can wait for each others), and if you have your own software of ai agents, you can literary ask ai to write the code instead of everytime waiting the result of your prompt.
2026-04-11 19:36:50
1
Darius H :
this is becoming realy boring
2026-04-10 00:46:06
6
iberianpoet (Kiko) :
even optimizing the harness is just one layer that can be optimized/ one of the many failure points. more info in this public repo: https://github.com/hjasanchez/agentic-engineering
2026-04-09 13:11:58
48
Natron308 :
I’ve spent the last 2 months solving all of these issues. 19 projects 330k new lines of code and 8800 automated tests. Red team, governor, auto start autonomous sessions, work decomposition and assignment, 3 tier memory. Typed Message and work assignment bus.
2026-04-09 15:00:00
7
edward.lcl :
ontolgy is key
2026-04-09 20:44:48
8
The AI Dividend :
/meta-bench, create a skill to assess the performance of your agentic dev workflows, /loop till u hit your #’s, extract rules to memory
2026-04-13 13:18:57
3
zaldabus :
Using the Haiku model is setting the bar pretty low
2026-04-09 12:25:05
7
WaterDog :
why not make the "harness" be a vector database similar to the system used in RAG, train the ai to output a vector that will be used to query the toolsbase (NNS style) and it will use that tool, appending its output to its context. this will mean that adding more and more tools no longer will require taking any context space, and since it is all done in the same embedding, results will improve as more tools are added since the NNS will always return the closest one to the "ideal" tool
2026-04-09 13:38:17
0
Mishmish :
Gee, what could go wrong?
2026-04-09 13:52:43
3
George Goosen :
here comes 'Harness engineering'
2026-04-10 03:10:02
3
helgenuuk :
This is literally what I’m doing right now 😅
2026-04-09 17:28:26
1
Raramente Real :
Basically AI is way better than humans using AI 😂
2026-04-09 14:47:47
2
Kevin :
Stanford is junk !
2026-04-09 12:20:51
0
poki6041 :
i editee your harness -500k lines + 500k lines
2026-04-09 16:39:28
1
dakon :
is that the same approach used in minimax 2.7 LLM?
2026-04-09 14:45:25
0
CRYINGAI :
The key principle is your system design
2026-05-03 19:19:47
0
KiloEthereal🇨🇦 :
goose
2026-05-20 05:23:36
0
Loris :
I know how to book a flight. Isn't this a really marginal gain,?
2026-05-22 07:05:35
0
warpcode :
I'll save the $100 dollars a month and months of constant tweaking and just book the damn flight myself
2026-05-24 11:58:12
0
e9n.dev :
Kinda, the harness decides which tool to tell the LLM about. The LLM then decides which tool to use
2026-05-06 15:59:55
0
CRYINGAI :
Stop trying to sidestep the crucial design principles with frameworks
2026-05-03 19:20:04
0
X :
Meta harness was built by Meta super intelligence labs ?
2026-04-09 14:27:17
0
To see more videos from user @parthknowsai, please go to the Tikwm
homepage.
