My AI Coding Stack
Everyone has opinions about AI coding tools. Most of those opinions come from people who tried them for an afternoon. I’ve been using these daily for months while building actual products. Here’s where I’ve landed.
Claude Code
Section titled “Claude Code”The one I reach for first.
Claude Code is my default for anything that requires thinking. Refactoring a messy module, designing an API, debugging something subtle — this is where it shines. It holds context well, doesn’t hallucinate as aggressively as alternatives, and the reasoning quality is a clear step above.
The downside? It can be slow when you need quick edits. And the pricing adds up fast if you’re doing heavy iteration. But for anything non-trivial, it’s the best tool I’ve used.
Verdict: My primary coding partner. Worth the cost.
Cursor
Section titled “Cursor”The daily driver IDE.
Cursor is where I actually write code. The inline completions are fast, the chat is good enough for quick questions, and the tab-completion feels like it reads my mind about 60% of the time. The other 40% it confidently writes something wrong, but that’s the game.
What makes Cursor work is the speed. It doesn’t break my flow. I’m not copy-pasting into a browser — the AI is just there, in the editor, where I’m already working.
Verdict: Non-negotiable. This is where the code gets written.
OpenAI Codex
Section titled “OpenAI Codex”The async workhorse.
Codex is interesting because it operates differently. You throw a task at it, go make chai, and come back to a pull request. It’s great for mechanical work — migrations, test generation, boilerplate. I wouldn’t trust it with architecture decisions, but for “convert these 30 endpoints from Express to Hono”? Perfect.
The catch: it works best when your codebase is clean and well-structured. If your code is a mess, Codex will make it a bigger mess. Garbage in, garbage out — but faster.
Verdict: Great for grunt work. Not a thinker.
v0 by Vercel
Section titled “v0 by Vercel”The frontend shortcut.
I’m a backend engineer. CSS gives me anxiety. v0 lets me describe a UI in plain English and get something that actually looks good. It’s not perfect — you’ll spend time tweaking — but it gets me from “blank page” to “80% there” in minutes instead of hours.
The generated code is clean enough to ship. It uses Tailwind and shadcn/ui, which is exactly what I’d pick anyway. For a solo dev building tools fast, this is a genuine superpower.
Verdict: Essential for backend devs who need to ship frontend.
The rapid prototyper.
Bolt is for when I need a working prototype in 20 minutes. It scaffolds full apps from a prompt — frontend, backend, database, the lot. The code quality isn’t production-grade, but that’s not the point. The point is validating an idea before investing a week into it.
I use it at the start of every weekly build to test whether an idea has legs. If the prototype feels promising, I rebuild properly. If it doesn’t, I’ve lost 20 minutes instead of 3 days.
Verdict: Brilliant for idea validation. Don’t ship the output directly.
The stack in practice
Section titled “The stack in practice”Here’s my actual workflow for a weekly build:
- Bolt to prototype the idea in 20 minutes
- v0 to generate the UI components
- Cursor as the IDE for all actual development
- Claude Code for architecture decisions, complex logic, and debugging
- Codex for mechanical tasks — tests, migrations, repetitive refactors
No single tool does everything. The skill is knowing which one to reach for and when to switch.
These tools are evolving fast. What I wrote today might be outdated in a month. I’ll keep updating as my stack changes.