How Non-Technical People Should Use AI to Code
You're staring at Claude or ChatGPT trying to build an app. You describe what you want. The AI spits out code. You paste it somewhere and it breaks. You ask the AI to fix it and now everything else is broken. Three hours later you're further from a working app than when you started.
Nobody talks about this part of coding with AI. The problem isn’t your skill level. It’s how you’re using AI. You treat it like one developer when you should be running it like a team.
An orchestra, not an instrument
After building an award winning AI tool without writing a single line of code, I learned something counterintuitive. The most successful builders don't use just one AI. They use two. When organizations build software they don't just have developers banging out code. They have product managers who decide what to build, architects who design how it fits together, and developers who implement it. You need that same structure.
How to set it up
First, your Planning AI is your CTO, your architect, and your product manager. Talk to it in plain language about what you want to build. Have a yap session. Contradict yourself. Change your mind three times. That's fine. The Planning AI's job isn't to code. It's to turn your messy human thoughts into a precise technical specification. Tell it about user flows, features, and the problems you want to solve. It will ask questions like "What happens when a user clicks this?" "Where does this data come from?" "What should happen if this fails?"
When your Planning AI understands the project, it creates a prompt that details the technical specification for your Coding AI. Then switch to your Coding AI and paste in that crafted prompt. Let the Coding AI execute. When it's done, don't just test and move on. Bring the results back to your Planning AI. That back-and-forth is where the work gets good. The Planning AI reviews what the Coding AI built, spots issues you wouldn't notice, and creates new prompts to fix them. You're not debugging, you're orchestrating.
Why this matters
This setup isn't just organizational. It's about cognitive load. Asking one AI to plan and code is too much for a model's limited context window. They might manage, but neither task gets their best work. AI coding tools are powerful but finicky. They tend to rewrite your entire codebase when you only want to change one feature. Not because they're bad, but because they try to interpret your intent while also executing it.
Separating planning from execution protects you from that fragility. Your Planning AI keeps the vision and remembers the architecture. Your Coding AI executes specific, well-defined tasks. When something breaks — and it will — the Planning AI knows enough of the overall structure to create a fix that doesn't destroy everything else.
This also solves the context problem. Every AI has a context window — how much it can remember at once. Use one AI for everything and that window fills up fast. Features get forgotten. The structure gets lost. When your Planning AI holds the high-level view and sends focused prompts to the Coding AI, you can build larger, more complex applications.
what this really means
You're not learning to code. You're learning to be a technical executive. You're making architectural decisions, managing trade-offs, and deciding what to build and in what order. The actual coding is implementation. Real CTOs don't write much code and neither should you.
That feels wrong to a lot of people. There's a sense that if you're using AI to code you should understand every line. That's like saying if you use Google Translate you should learn the language. You don't have to, that's the point.
You do need to think systematically about software. What are the core features? What's the user flow? What can go wrong? Those are product questions, and you're probably better at them than you think.
When you get stuck, don't ask your AI to "fix it." That's too vague. Take the error to a different AI, maybe a reasoning model, and ask it to explain what's happening. Get second opinions. It's like getting multiple contractors to quote a renovation — each will see different problems and different solutions.
The principle is simple and profound. Stop trying to be a developer with AI assistance. Start being an executive with AI employees. Make that shift and AI coding stops being frustrating and starts being powerful. The code was never the hard part. The thinking was.
THE PLANNING AI PROMPT TEMPLATE
You are my technical architect and product strategist. I'm building an application and need you to help me think through the architecture and create detailed prompts for a separate Coding AI (Claude Code).
PROJECT OVERVIEW
[Describe what you want to build in 2–3 sentences]
USER NEED
[What problem does this solve? Who is it for?]
CORE FEATURES (in order of importance)
- [First feature]
- [Second feature]
- [Third feature]
CONSTRAINTS
- Technical: [e.g., "needs to work on mobile," "must integrate with X"]
- Resource: [e.g., "using free tools only," "needs to deploy in 1 week"]
- User: [e.g., "users are non-technical," "needs to be accessible"]
YOUR TASKS
- Ask me clarifying questions about user flows, edge cases, and technical requirements until you fully understand the project.
- Identify potential problems I haven't considered.
- Once we've refined the vision, create a detailed technical specification.
- Generate a precise prompt I can give to a Coding AI to build the first version.
Your AI Development Session Checklist
Click a section to expand it, and use the checkboxes as you work through your session.
Before You Start
Planning Phase
Coding Phase
Review Phase
Before Ending Your Session
Red Flags, Stop and Reassess
If you check any of these, pause the session and reassess your scope, prompts, or architecture.
This post covers the basics, but there’s a lot more: which tools work best for different types of projects, ensuring the integrity of your tech stack, scaling, version control, testing framework, etc. Contact us at info@theconfleuncial.com if you need a thought partner or have more questions about this.