@Bryan Alva Anything I need, I build. So far I finished: - Shopify App for local Jewelry Store for easy content generation and anything they need, I build on top of it - eCommerce pure profit after Ad spend tracker - custom app for iPhone that talks directly to my terminals on iMac. It's only the beginning. Always be shipping 🚀

this is a trading book, but it's really good way of thinking about risk for life and business

@Diana Sanger Do not judge a decision only by its outcome. Judge the process, the risk taken, and what could have happened but didn’t.

The folder hierarchy can absolutely become both: - the human control surface, where you encode intent, rules, priorities, templates, routing, and “definition of done” - the agent’s operating map, where Claude decides what to read, what context matters, and which sub-agents or workflows to run You can drill very deep: folder-level instructions, examples, checklists, guardrails, output formats, handoff rules, review steps. That will make Claude much more likely to execute your setup faithfully. But it still won’t be perfectly deterministic. Claude can misread, skip context, over-apply a rule, or make a probabilistic judgment you didn’t intend. So your responsibility is to design the system, constrain the agent, and verify the output. Yes, you are the orchestrator of the operating environment. Claude is the execution layer. But the execution layer still needs guardrails, tests, review, and correction because it is not deterministic software.

Hi Bagu, you need JSON. - find example of design, picture or whatever and ask your agent to create a JSON template of the image. - now ask your agent to create a picture with that JSON. - a lot of sloppy text comes from an agent trying to add text on top of the image via code, if this keeps happening, just ask your agent to only use Image LLM to generate full picture.