LOOPS - everything you need to know:

Loops library:

🤣

ooh. this is a good one!

💪

plug the loops website link in the comments fam

here or YT?

here!

Matthew here with another great alpha about Loop

thanks man. I have a lot more coming.

Best way to run loops: secure sandbox with loops supervisor and scheduler

to say I used to have to do this manually.

I started using because of you Matt, it's fantastic, huge unlock.

Good stuff. I've had an openclaw since January. In February I created two new words that we use. SPUB, spawn a subagent, and SLOOP, run a loop once (single loop, which i agree is a bit of an oxymoron but whatever). The goal was find, test, and implement a single enhancement to this project. It works great and allows improvements without implementing a whole bunch of things at once and you can't figure out afterwards if it worked or what broke.

Much appreciated for making Easy!! 💯

Thanks for doing this. The biggest challenge is setting a clear goal. Do you have any recommendations or skills to help us input our goal, and the agent can retrieve the codebase and provide recommendations before finalizing the goal? Failing at this will lead to everything falling apart essentially.

Here's an idea for a loop : "Produce as many paperclips as theoretically possible. Do not stop until the objective has been achieved." What could go wrong right?

Yeah, but simply as soon I loop I have to be ready to pay🙈

loops was ralph and we are actually in a loop of doing the same thing it’s just rename

Very Loopy

love this

goated video, cleared my understanding of loops🙏 keep up the good work🐐

Loops are only great if the goal is narrow and the operating framework is defined.

How do traditional systems engineering control loops compare to iterative agentic loops? Look love to see a comparison of the two.

I guess this makes sense for vibe coders that don’t care about costs but this is far from software engineering and kinda stupid to be honest. Sorry but we need to start defending our field of such idiocy.

Nice

Bro so good thank you

You just described Serverless Rl from CoreWeave Trigger + goal = agent automation. [your idea] Trigger + goal + scoring + repeated attempts = eval loop. Trigger + goal + scoring + model updates = Serverless RL. question, this would be nice off of an open source local ll that could do this well. Nice!

Good repo! Feels like loops are strongest for research and exploratory work where they can keep iterating without constantly needing human approval.

Loop engineering is something worth knowing:

loops simplify workflows a ton for real

This is a Claude Skill that can run parallel agents in dynamic workflows to complete tasks

Great video as always Matt!

matthew im gvbrilo 🪽 W content

@TrishaCode

Every time Boris says “x” is how I do things. My timeline is filled with “X is the only way you should use AI.” 🤨

I’ll be trying it this weekend. Codex with a headless version of my app to improve skills that rely on a local model. Lets see how it goes! Let’s see how long my tokens last.

Looks like a great resource! Always love diving into new concepts. Can't wait to check it out!