Agents are interns with infinite stamina
Use Cursor, Claude Code, and GitHub Copilot to draft code, tests, and refactors. You still review. You still own production. An agent that "finishes" a feature you do not understand is a future outage.
Write standing orders: coding standards, folder map, and forbidden actions (no force-push, no deleting billing code). Agents follow instructions better than vibes.
Keep secrets secret
Never paste production keys into random chats. Use environment variables. Rotate anything that might have leaked. Paranoia is a professional skill.
Prefer small pull requests. Giant agent diffs are how bugs throw costume parties.
Automate toil, not taste
Let robots write boilerplate. You decide product taste, pricing, and which customer you refuse. That division of labor is how you stay the ruler instead of the intern.
Running tip: once a week, ship something without AI assistance. Keep the muscle. Tools change. Judgment compounds.
Field notes
Field note: robots are excellent at drafts and terrible at consequences. Keep humans on the hook for anything that moves money, deletes data, or emails customers. That boundary is the difference between leverage and liability.
When an agent surprises you with a clever approach, ask it to explain tradeoffs before you merge. Cleverness without tradeoffs is how outages get stylish.
Operator checklist
Operator checklist: write standing orders for coding agents — folders, style, forbidden actions. Keep secrets in env vars, never in prompts you paste casually.
Require review on anything touching auth, billing, or customer email. Merge small diffs. Fear large clever commits.
Ship one small change this week without AI so your taste stays calibrated. Tools should amplify judgment, not replace it.
How to run it this week
Require tests for agent-written code that touches money or auth. Fear is appropriate there.
Keep a human changelog of what agents changed. Blame needs an address.
Budget for tool spend like payroll. Infinite AI tabs can bankrupt a thin margin business.
What good looks like
What good looks like: standing orders for agents, small reviewed diffs, and secrets kept out of prompts. Leverage with a spine.
Common failure mode: merging clever code you cannot explain. Outages love mystery.
Series bridge
Series bridge: this chapter sits between measuring what matters and keeping the lights on. If you skipped here, go back one part and finish the checklist first. The playbook only compounds when each layer is real — research under offer, offer under site, site under product, product under distribution.
Keep a single Notion page titled Kingdom OS and paste the outcome of this part at the bottom before you leave. Date it. Future you will skim twenty dated notes faster than twenty open browser tabs. That page becomes the operating system the last chapter asks you to run daily.