So I started using the Folder structure for my agency and I've got a few clarifying questions with growing folders. I work mainly with Antigravity in the terminal My folder structure right now is build like this: Main Folder/ ├── CLAUDE.md ├── STARTUP.md │ ├── core/ │ ├── platform/ │ │ ├── REFERENCE.md │ │ ├── apps/ │ │ │ └── dashboard/ │ │ │ ├── CLAUDE.md │ │ │ ├── AGENTS.md │ │ │ ├── public/ │ │ │ └── src/ │ │ │ ├── app/ │ │ │ ├── components/ │ │ │ └── lib/ │ │ ├── docs/ │ │ │ └── DEPLOY.md │ │ ├── infra/ │ │ │ ├── docker/ │ │ │ ├── nginx/ │ │ │ └── pocketbase/ │ │ └── pb_data/ │ │ │ └── website/ │ └── REFERENCE.md │ ├── delivery/ ├── blueprints/ │ ├── README.md └── clients/ ├── CLAUDE.md ├── client1/ │ └──PROJECT.md └── client2/ └── PROJECT.md My questions: When working on something specific for example client1 do i go into the client1 folder and then open claude or from the main folder since the routing is in the CLAUDE.md? Are there enough md files or am I missing crucial md files? Is the layout too big or is it fine? When in the example above do I still give the model a role when prompting and what to do and not do even if most of that is in the md file already? I guess thats about it, thank you in advance for any help!

After reading the full history of thinking machines: Vote below, then tell us in the comments: which part of the timeline surprised you the most?

Even with AI, Cursor, vibe coding, and modern tooling… building a real product is still hard. People online sometimes make it sound like you can casually prompt your way into production-grade systems overnight, but once you start dealing with: - architecture, - validation, - edge cases, - workflow coordination, - reliability, - testing, - approvals, - and keeping everything maintainable… you realise very quickly why experienced developers are valuable. BuilderCore V1 took weeks of intense work just to reach a controlled foundation stage — and that’s with AI assistance accelerating parts of the process. Honestly, massive respect to developers who were building complex systems before modern AI tooling even existed. AI can massively increase leverage, but it doesn’t magically eliminate engineering difficulty or system complexity.

You just used Claude Desktop as a thinking partner instead of a vending machine. Vote below, then tell us in the comments: what was the problem, and what was the insight you walked away with?

You just watched the full 23-minute folder architecture walkthrough. Vote below, then drop your use case in the comments. What are you building this system around?