You don't need to know how to code. You need to understand what you're trying to accomplish, why it matters, and what a successful result looks like.
Start with Codex or Claude Code and a project folder. The downloadable AI Product Starter Kit gives you six files:
AGENTS.md— The shared working agreement: how the AI communicates, what it can do, when it needs permission, how it verifies its work, and when a new thread needs a complete handoff prompt.CLAUDE.md— Imports that agreement for Claude Code. Codex readsAGENTS.mddirectly.PRODUCT.md— Who the product serves, what problem it solves, the outcome you want, and the boundaries that matter.START-HERE.md— Setup guidance and a copy-ready prompt for your first conversation.docs/CURRENT.md— Decisions and authorizations, what actually works, supporting evidence, unresolved questions, and what's next.docs/features/_TEMPLATE.md— The starting point for a feature brief (docs/features/<name>.md): what you're building next, what's included or excluded, and how you'll judge the result. The template itself approves nothing.
You don't have to write these yourself. The working agreement and templates are already supplied. Ask the AI to interview you, examine anything already built, and help complete PRODUCT.md and docs/CURRENT.md for your review.
Once you've agreed on the problem and desired outcome, ask the AI to help you explore possible solutions. Decide what matters, how much effort the work is worth, and which approach best fits.
Then choose the smallest complete, useful part to build. Have the AI document its intended behavior, boundaries, and acceptance criteria in a feature brief. Approve the brief and authorize its implementation before building begins. These can be one explicit decision; neither authorizes later features or publication.
Then work in short cycles:
Build → Test → Demonstrate → Accept or Improve → Repeat.
Accept the work only after you've seen it working yourself, not merely because the AI says its tests passed.
The AI handles the technical work and keeps CURRENT.md updated. You approve changes to the product's purpose, intended behavior, and scope.
Keep the process proportional to the problem. A small, obvious fix doesn't need an elaborate plan. A complex or uncertain feature deserves more thought before implementation.
You decide what matters and whether the result is good enough. The AI figures out how to build it.
The goal isn't to produce more code. It's to make something useful that actually works.
Ready to start?
Download the AI Product Starter Kit — Codex + Claude Code
A ready-to-use project folder with agent instructions, product and progress templates, a feature brief template, and a first-session prompt.
Download it, open the folder in Codex or Claude Code, and follow START-HERE.md. The AI will help you establish the product before building it.
The kit has been tested through real Codex sessions. Claude Code can import the same working agreement, but its full workflow has not received the same behavioral testing.
If you already have an approved plan, the agent follows its priorities, sequence, and boundaries rather than inventing a replacement. After meaningful work, it reports what was verified and recommends whether to continue the thread, start a new one with a copy-ready prompt, or stop and reassess.
For a more structured approach to understanding problems, exploring solutions, evaluating tradeoffs, and planning what to build, see Planning Skills for Agents and Humans.