Over the past year, I've shipped more than 10 products as a solo developer: macOS apps, Chrome extensions, web apps, mobile apps, and games. That would have been impossible without AI-assisted development. Here's exactly how the workflow works.
The Stack
My primary tools for AI-assisted development:
- Claude Code — My primary development partner. I use it for architecture decisions, complex refactoring, debugging, and writing entire modules. It's especially powerful for multi-file changes where you need to keep context across a whole codebase.
- Cursor — Great for in-editor completions and quick inline edits. I use it when I'm deep in a file and need suggestions as I type.
- GitHub Copilot — Tab completions for boilerplate and repetitive patterns. Useful but not where the magic happens.
The Workflow
Every product follows roughly the same pattern:
1. Architecture First
I describe what I want to build in plain English. The AI helps me think through the architecture: what frameworks to use, how to structure the project, what the data model looks like. This conversation usually takes 10-15 minutes and saves hours of wrong turns later.
2. Scaffold
The AI generates the project structure, config files, and boilerplate. For a React app, that's the Vite config, Tailwind setup, routing, and component stubs. For a macOS app, it's the Xcode project structure, Info.plist, and SwiftUI views. This used to take a day. Now it takes minutes.
3. Iterate
This is where I spend 80% of my time. I build features one at a time, testing as I go. The AI handles the implementation details while I focus on product decisions: What should this button do? Where should this data come from? What's the best UX for this flow?
4. Debug
When something breaks, I paste the error and relevant code. The AI usually identifies the issue immediately. Cross-platform bugs (like macOS-specific permissions or Chrome extension manifest issues) that used to take hours of Stack Overflow searching now get resolved in minutes.
5. Ship
Deployment configs, GitHub Actions, release notes — the AI handles all the tedious parts of shipping. I focus on testing the final product and making sure the user experience is right.
What AI Is Good At
- Boilerplate and project setup
- Implementing well-defined features
- Cross-platform compatibility fixes
- Refactoring and code cleanup
- Writing tests
- Documentation and README files
- Debugging with error messages
What AI Is Bad At
- Product decisions (what to build, for whom)
- UX design (what feels right, what's intuitive)
- Understanding user context and real-world constraints
- Knowing when to stop adding features
- Taste — the difference between "works" and "feels great"
Real Project Timelines
Here's how long some of my projects actually took from idea to launch:
| Project | Type | Time to Ship |
|---|---|---|
| MeetingNotes Extension | Chrome Extension | 2 days |
| LowPower Automator | macOS Menu Bar | 1 day |
| MenuTemp | macOS Menu Bar | 4 hours |
| FileGenius | macOS App | 1 day |
| Flappy Axolotl | Cross-platform Game | 1 day |
| Salmon Run | Web Game | 3 hours |
| AI Usage Tracker | macOS SwiftUI | 2 days |
| Tri-Cities Utility Monitor | React Web App | 3 days |
These are calendar days from first commit to public release.
The Bottom Line
AI doesn't replace the developer. It replaces the tedious parts of development: the boilerplate, the syntax lookup, the config file debugging. What's left is the creative work — deciding what to build, how it should feel, and when it's good enough to ship.
The developers who will thrive are the ones who can clearly describe what they want, evaluate the output critically, and make good product decisions. The code is becoming the easy part.
Want to see the products built with this workflow?
Browse all products →