Most AI setups fail because the same agent writes the work and grades the work. Quick-Connect starts you at Chapter 2 with a stack that already separates makers from judges.
These ten rules are the doctrine that survived ten independent videos. Share them. Implementation — NODE, pair, Marco on your channels — is the paid product.
The thing that does the work never judges the work. A second checker, with different context, signs off.
The human gate is sized to the cost of the mistake. A private note is light. Customer mail and money are heavy.
Find the constraint before you build the machine. Do not automate the wrong bottleneck.
Run it by hand before you automate it. Manual first. Then scripts. Then agents.
The smallest version that improves quality wins. Fancy harnesses lose to a tiny check that actually fails.
Every run must leave something behind. If there is no text file, the work is not done.
Couple to the interface, not the vendor. Skills and standard connectors travel; one-brand lock-in does not.
What looks like a boundary usually isn’t. Context, tools, and memory leak. Design as if they will.
Quota and cost are operational, not footnotes. Caps, free lanes, and paid lanes are part of the SOP.
Running processes are not durable. If it is not scheduled, versioned, and restartable, it will vanish.
Free to share · paid to implement
Print these rules. Teach your team. When you want the starting line done for you — one card, one NODE, Chapter 1 handled — that is Quick-Connect.