There’s a mysterious new model called Ox Alpha causing a storm right now, and nobody even seems to agree on who actually built it. People are testing it for coding, design, agents, and full software projects, and the crazy part is how accessible these models are becoming. This is one of the biggest reasons I wanted to build this community in the first place. If AI gets to the point where almost anyone can build software for free—or close to free—then simply being able to build software isn’t the moat anymore. The valuable part becomes knowing what to build, how to use these tools, how to distribute what you make, and having a community of people learning and building alongside you. A year ago, access to a really powerful coding model felt like an advantage. Now a random mystery model can appear overnight and suddenly thousands of people have access to another insanely capable developer. That makes me even more bullish on community > individual software products. Software is getting cheaper to create. Knowledge is getting cheaper. Models are getting better every month. But trust, relationships, distribution, and a group of people actually implementing this stuff together are much harder to commoditize. That’s what I want this Skool to become: not just a place where I post AI news, but a group where we find things like Ox Alpha early, test them, share what actually works, and use them to build real stuff before everyone else catches up. Resources OpenRouter — explore and compare AI models: OpenRouter Models OpenCode — open-source AI coding agent that can connect to models from many different providers: OpenCode Curious what everyone thinks: if building software keeps getting cheaper, what becomes the actual moat?

This went live this morning, so if you're reading this today you're genuinely early. What actually happened Replit shipped Free Mode today, announced jointly with OpenAI. It runs on GPT-5.6 Luna — the cheapest model in OpenAI's 5.6 family — and it's now the default mode when you open Agent. The reason it exists: OpenAI cut Luna's price by around 80% on July 30. That's the whole story. Replit didn't get generous, the floor dropped. Reality check on the name, because people are already confused: - It is not a free plan. Starter gets Free Mode inside the limits Starter already had. - Core ($20/mo) and Pro ($100/mo) get a Free Mode allowance that refreshes every ~5 hours, with a weekly cap on top. Pro's weekly cap is bigger. - "Free" means doesn't burn your credits. Not costs nothing. - Also, the modes got renamed and this trips people up: Old name New name What it is Lite Free No-cost everyday agent work, always on Auto Economy Power Paid mid-tier, same price as Economy was Power Max Paid top-tier, frontier models Your old settings carried over under the new names. Why this one matters more than a normal pricing update I've been saying the same thing for months: you don't need a frontier model to rename a button. Cheap model for cheap tasks, expensive model for the hard 10%. Replit just took that idea and made it the default setting for everyone. The frontier model is now the exception you opt into, not the thing running by accident while you ask it what a useState hook is. That's the shift. Not "free." The default flipped. The actual tips 1. Set the mode before you type, not after. Mode selector sits under the prompt box. ⌘+Shift+I (Ctrl+Shift+I on Windows) cycles Free → Power → Max without leaving the input. Most wasted credits happen because someone starts typing in whatever mode was left open. 2. One step per prompt. Always. Free Mode falls over when you dump "build me a full CRM with auth and Stripe" into it. It's strong when you give it one screen, one feature, one interaction at a time. Build the thing that proves the idea works, then add to it.

Been messing with an uncensored version of Qwen3.8 (27B) running locally in LM Studio, and it's genuinely wild how unfiltered it is. No refusals, no "I can't help with that," no lectures you can ask it literally anything and it just answers straight. Made a video where I used it to break down my own business what I actually need to improve, blind spots I've been avoiding, and where I should be focusing next. It gave way more direct/honest feedback than ChatGPT or Claude typically will, since it's not softening anything. If you're using AI for brainstorming or business strategy, an uncensored local model like this is worth trying curious if anyone else here has messed with it. Dropped the full breakdown here if you want to see it in action: