I’ve just completed the AI Operating System training from Nate Herk’s AI Society community and decided to put it to work on a real business rather than a hypothetical project. My first test case is my wife’s dessert catering business, Treats2eat. The goal is to see how far an AI Operating System can go in helping run and grow an existing business across areas like: • Marketing• Meta ads• SEO• CRO• Customer research• Operations• Product development• Analytics• Strategy• Troubleshooting Rather than building a single AI assistant, I’m building a system of specialised agents that work together, challenge assumptions, review each other’s work, identify opportunities, diagnose problems, and recommend actions based on business impact. The experiment is simple: How much of the day-to-day thinking, planning, analysis, optimisation, and execution support can be handled by an AI Operating System while still maintaining quality and commercial judgement? I’ll be documenting what works, what fails, what needs human oversight, and where the biggest opportunities and limitations are. Happy to share learnings as I go and potentially parts of the operating system itself once I’ve cleaned out any business-specific or sensitive information. Would love feedback from anyone else who’s implemented Nate’s framework in a real operating business rather than a greenfield startup. What worked? What broke? What would you do differently?

Would love to get a gauge as to why people in the community have chosen either Claude code or codex and why? Or are you using both?

One of the most useful AI techniques I've come across recently is surprisingly simple. Most people write a prompt, hit enter and judge the result. Instead, I get the AI to: • Build the prompt • Critique the prompt • Score it against a rubric • Identify weaknesses and failure points • Improve it • Review it again • Improve it a second time • Then execute it For example, I’ll say something like: “Create this prompt first. Then review it with a strict rubric, identify what is weak, improve it, review it again, and only then produce the final output.” The difference in output quality is often significant. What you're really doing is separating creation from evaluation. Humans naturally do this when writing, designing or solving problems, but most AI users skip that step entirely. The process forces clearer thinking, better constraints, fewer assumptions and much stronger outputs. It's not just useful for AI prompts either. It works for strategy, marketing, business planning, creative work, proposals and decision making. The biggest lesson for me: the quality of the review process often matters more than the quality of the first draft.

This morning a Mythos identifier showed up on Anthropic's API, people screenshotted it, and the whole timeline decided a launch was days away. I break down what Mythos actually is, what really happened this morning, and why a leak plus widening access still isn't the same as a public launch. My honest bet is that the capability quietly folds into the next Opus before anyone ever logs into something called Mythos, and the thing you should really be watching is OpenAI's next move.

I've installed skills! noticed this doesn't automatically get added to your standard Claude browser? Anyway this deserves approximately 11 likes. 😁 Also, how do I know if I have given Claude too much access to my PC?