Luigi Marcon

Exploring the weird edges of AI.
Hidden projects, strange tools & experiments.
Human × Agent.

[Luigi Marcon @LuigiMarcon28]

I’m a construction worker from the Italian Dolomites.

At night, I’m running a public experiment.

I’m building businesses, AI systems, and searching for better ways to create freedom and leverage.

Some experiments will work.
Some will fail.

I’ll share both. 🚀
#Entrepreneur

[Luigi Marcon @LuigiMarcon28]

@MarkusEicher70 @witcheer Make sense so you can, in some way, personally optimise the skill. But how do you make sure the skill is good and not “waste tokens”.

Do you think the prompt already does it efficiently?

[Luigi Marcon @LuigiMarcon28]

@Ionkosm So you are an early person! That’s good time then! 💪 everyone are different!

[Luigi Marcon @LuigiMarcon28]

Hey thanks so much! You did help a lot! I’m saving money to upgrade to a good hardware.. unfortunately now I only have a Mac mini 8gb from 2 years ago. So I’m “running” 😂 a g9v3-3b.. barely good for conversation..

But your explanation help me understand what kind of hardware I can aim to buy..

16gb is the bare minimum as I see.. 34Vram would be a good starting point.

Hope while I save some money people will optimise local models to run good on less powerful pc so I can optimise my buy! 🙂

Are you running some local? Or not..

[Luigi Marcon @LuigiMarcon28]

@Ionkosm 6am is personally to early for me.. and midnight too late 😂I’m getting old.. I have a peak of concentration around 10.

What about you

[Luigi Marcon @LuigiMarcon28]

@MarkusEicher70 @witcheer Hermes should create the skill automatically right?

[Luigi Marcon @LuigiMarcon28]

@HermesAgentTips MAC OS vs Omarchy?

[Luigi Marcon @LuigiMarcon28]

@Jonnykov @AlexFinn I totally agree with you! 🙂 the way you can create with a local model is impressive and also the “freedom of speech” 😂if we want to call it that way..

Which that said.. I don’t know.. for a Qwen 3.8 27b what hardware would you suggest? Its the bare minimum a 27B in my opinion

[Luigi Marcon @LuigiMarcon28]

Would you test your agent against another agent? To see who built the best one? #LLM #AI

[Luigi Marcon @LuigiMarcon28]

@DrewWakefield5 4 for me! 🤪 but one is finished and I’m using it.. just a webapp for myself. I think with ai agent one of the best uses you can realistically have is building things you want or need without going to pay someone else..

[Luigi Marcon @LuigiMarcon28]

During the next days and weeks I’ll continue build with MiMo flash! Giving it tasks that become more difficult every time!

I’ll keep you updated with the future experiments! 😃hope you enjoy reading and testing them! Thanks in advance

#LLM #AI

[Luigi Marcon @LuigiMarcon28]

@AbhiCodes15 Creativity in my opinion!

[Luigi Marcon @LuigiMarcon28]

@niqudrift EU here! 👋

[Luigi Marcon @LuigiMarcon28]

@udiWertheimer What about obsidian+cronjob? To keep all you do in check and the agent have everything there

[Luigi Marcon @LuigiMarcon28]

CAN YOU BEAT MY SCORE? 🎮👀

I gave Hermes Agent + MiMo a challenge: build a complete 3D game autonomously.

It built, tested, debugged and deployed this game. No manual coding.

My record: 18 seconds, 10/10 coins, ZERO falls. 🏆

Think you can beat me? #LLM #AI

Play: https://t.co/uBVWZzVoPe

Post a screenshot of your best run below! 👇

[Luigi Marcon @LuigiMarcon28]

@Jonnykov @AlexFinn At the same time.. you have freedom with a local model.. but again.. money is king

[Luigi Marcon @LuigiMarcon28]

@AlexFinn I agree but it is too expansive if you want any a good model.. we need a way to make a good model function local without spend 4000$ in my opinion

[Luigi Marcon @LuigiMarcon28]

I used to use deepseek but guys..

MiMo -V.2.6 flash has been a game changer for me..

Honest feedback: if you give him something to do and tell him to not ask question. He will do an amazing job autonomously (also multiple test to give you back the best product he can) BUT it is much slower then deepseek..

So In my opinion you have a choice to make.. speed vs autonomously