Six rules refined across 40+ live tools built with AI coders (Claude Code, Cursor, Gemini CLI):
- 1. Share - context is half the code: give background, logic, schema, goal, constraints, sample rows, even failed attempts. Garbage intent in, garbage code out.
- 2. Tell - state clearly what you want, the scope, format and integrations upfront. It is a coder, not a mind reader.
- 3. Ask - interrogate first: never execute blind, demand a plan and 2-3 alternatives with trade-offs before running.
- 4. Iterate - there is no magic prompt; you still build in loops, just faster, and ask it to do security and performance reviews.
- 5. Validate - inspect what you expect, stress-test it, try to break it. AI or not, the buck stops with you.
- 6. The grind doesn't go away - debugging, edge cases, logs and data errors still take real hours. Faster and scalable, but still a grind.
The point: the skill now is turning domain knowledge into working systems, not syntax. Full post: https://www.tigzig.com/post/ai-coders-give-you-the-edge-the-6-rules-i-follow-when-working-with-ai-coders. How Amar works: https://www.tigzig.com/how-i-work.
Building something like this? How I work covers the rates, the availability and what I take on.