Jonny9811

[Jonny9811 @Jonny9811]

@mehdidjabri Mines better and simpler https://t.co/5efnvO9M9u

[Jonny9811 @Jonny9811]

@SamVale29 https://t.co/UjMIMPKk5H

[Jonny9811 @Jonny9811]

@ProductHunt Harness the harnesses https://t.co/KBUgehk2HV

[Jonny9811 @Jonny9811]

Ranking narrows candidates from git history, knowledge base and rules. A decision model judges each one’s relevance, then an LLM curates memory for context.

So it covers all kinds of memory: the decision history in git (episodic), the knowledge base (semantic) and rules (procedural).

[Jonny9811 @Jonny9811]

@skelz0r @marmotize_ @JoshARosen Git commit messages are prefect for this, I’ve been doing this for a while, but now after adding basic ranking and a decision model the results are beyond what I expected

[Jonny9811 @Jonny9811]

Yes they’ll experience state of the art inference, but what if you want them to follow clear process every time without forgetting something? That’s just not possible right now, and skills don’t enforce it either. Workflows that drive existing harnesses/models in the answer, they progress with the frontier harnesses not instead of them.

[Jonny9811 @Jonny9811]

@lightsilver323 https://t.co/KBUgehk2HV

[Jonny9811 @Jonny9811]

@spacedevin https://t.co/KBUgehkAxt

[Jonny9811 @Jonny9811]

@Joi2James https://t.co/KBUgehk2HV and my marketing site is currently really shit

[Jonny9811 @Jonny9811]

@hanakoxbt This is great! I gave this to my agent and had it built the workflow with https://t.co/KBUgehk2HV, this aligns very closely with my thinking when I built it, although in my opinion the loop lives outside the node.

[Jonny9811 @Jonny9811]

@Lummox_eth Build process around inference, keep agents doing small focused tasks within a set workflow. The results are much better.

[Jonny9811 @Jonny9811]

@zachlloydtweets I’m building this. https://t.co/cznVzaeIEh

[Jonny9811 @Jonny9811]

@housecor Better yet, it should be a skill that launches a coded workflow

[Jonny9811 @Jonny9811]

@KingBootoshi 1000% agree

[Jonny9811 @Jonny9811]

@dillon_mulroy Replying off the cuff feels quite freeing in a world where people are getting ai to write or review even the most trivial content

[Jonny9811 @Jonny9811]

@dillon_mulroy The only model that actual sticks to doing what you ask it, both in terms of chat output/communication style and what it produces

[Jonny9811 @Jonny9811]

@inverily Creativity

[Jonny9811 @Jonny9811]

@mikehostetler Surely even better now that we can bring ideas to life so much faster?

[Jonny9811 @Jonny9811]

@unclebobmartin And quality hasn’t dropped?

[Jonny9811 @Jonny9811]

@arscontexta Yeah, the harness launches a lightweight micro ui automatically, I do what I need to do, then close it and it feeds back to the agent and the workflow continues.

Can launch manually too.

Some basic docs for it here - https://t.co/bzdU04W8ZE