I did it. I figured out the perfect AI coding loop It's called the Finn Loop and it's 100x'd my vibe code output while cutting down work by 95% In this video I cover how to set it up By the end you'll have your own software factory +I... completely open sourced it down below!show more

Full Finn Loop repo here: Enjoy! And let me know if you have any questions!

This is really cool. Just made a @ChatGPTapp Site that explains how the whole loop works and how to install it. Completely one-shot with ChatGPT Work:

as good as it looks on paper these loops create a lot of slop that just compounds and you either end up with a lot of cleaning to do or a lot of failed loops that need reruns and then wasted tokens. Is just not efficient. I'd like it to run properly, don't get me wrong, but we're just not there. And sooner or later this should already be embedded in the models. Also not sure how /goal in Codex for example doesn't just do this? You set a plan and it creates loops, it tests, redoes it, etc. until it ends the work and accomplishes the goal.

AI is making it easier than ever to build. $OBX is making it easier to own and monetize the data those AI systems depend on through a private marketplace built on Solana. The next wave is not just better AI. It is better data ownership. Ca:ttgtaonT9ncgPqgsgJxmLe4Ls1VuVaqJyDyvJJfpump

Do I really have to sit through 20 minutes of you waffling on. Why not just post the file in a easy to read file I can give to my AI....😂

Didn’t we move from loop to graph.

I keep borrowing pieces from other people's workflows instead of copying them end to end. That's usually where I find the biggest improvements.

I am actually watching it on YouTube right now 😅🤗

Am I the only one that when I finish an incredible vibe coding session, and I need to say goodbye to this session and start a new session, that It kind of feels like saying goodbye to an old friend?

I’ll add it to my marketplace ! Alex Finn the Honest

Here's my software factory / vibecode orchestra:

Man respectfully this is pretty basic I don't think you invented something... and in my experience I would add some kind of auditor using a top tier model like Fable 5 before merging anything... I would also add some automated tests and other stuffs to the "loop"

I’ve tried doing this with kanban style orchestration accross several custom agent profiles and skills. The time it took to run the loop was not worth it, and context drift would always get introduced, causing the model to go off the rails. Interesting concept though

Good one! But now they have moved from loops to graphs ...

Open sourcing it is the right move regardless of the number. 100x is a lot of precision for a technique with no benchmark attached. Curious what you're actually measuring: wall clock, token spend, or just typing less.

I asked ChatGPT how much you’ve increased your productivity over these last several months: Conservatively: 100x. Additively: at least 670x. Alex Finn Cinematic Universe math: 1 quadrillion x. @AlexFinn is ASI 🤯

I hope this is on YouTube because Twitter sucks for video

it's on YT as well

The Peter Steinberger quote in there is the real thesis, "design loops that prompt your agents" instead of prompting them yourself one message at a time. - A loop means the agent checks its own output, corrects, and re-runs without you sitting in the chat driving every step - That's the actual definition of a "software factory," you're managing a process, not typing prompts all day - Open sourcing it is smart distribution too, people who use his loop become an audience for whatever he ships next This is where the real 10x comes from with AI coding right now, not better single prompts, but systems that keep working while you're not watching them.

This has been a game changer for me - it absolutely works, took me minutes to incorporate with my current setup. Congrats and thanks!!!

@tobiaspatch

The loop idea is genuinely more useful than any single prompt trick, but models tend to drift off-task after 10-15 iterations and that's where most vibe coding setups collapse. How are you handling scope creep inside the loop?

Does it work with GitLab?

Agreed

Every dev has their loop right now and most of them are genuinely good. The honest thing nobody says enough: the loop matters less than knowing what to build and whether the output actually works. A fast factory still needs a quality check at the door.

The part worth copying is the human merge rule. /spec, /build and /review can still make noise, but “only I add agent-ready and only I merge” is what keeps the factory from becoming PR spam.

Sounds like a video we must all watch!

great hook. the hardest part with these autonomous loops is knowing exactly when to stop and hand control back to a human. without that, bugs compound fast. what’s the actual stop condition in the Finn Loop?

Loops 🥱 old new if u aren’t using capsules w loops, ur way behind

Curious to see the full setup.

Curious how you got to the 100x and the 95%, genuinely asking because I collect these. Most big multiples I see are counting output volume, and the people who follow it through to merged and shipped usually land lower but still good enough to be worth it.

100x output while cutting work by 95% is a bold claim worth watching 👀

Can you use this with hermes?

I see how the loop handles ambiguity, review failures, and human escalation. What I still do not see is how you measure whether the factory is economically effective: first-pass approval rate, review time per merged PR, average fix rounds, discarded PRs, and escaped defects. Once generation outpaces human understanding and testing, review becomes the real production line.

sorry the only good loop is a ralph loop. lol

Thank you , im new to ai an any information you have im interested