AI in My Daily Developer Workflow
After a year of trying every AI dev tool that shipped, here's what genuinely made me faster — and what I quietly stopped using after the novelty wore off.
What Actually Stuck
I'll be direct: most AI tools I tried for six months and then quietly abandoned. The ones that stayed are the ones that removed friction from things I was already doing — not the ones that promised to replace thinking.
Cursor for Coding
I switched from VS Code to Cursor and haven't looked back. The key feature isn't autocomplete — it's the ability to have a conversation about a specific file or selection. Asking "what's the purpose of this function?" or "refactor this to handle the error cases" in context is qualitatively different from copying to ChatGPT.
How I Actually Use It
Cmd+K for inline edits, Cmd+L for chat with context. I keep the AI context tight — one file at a time, not the whole repo. Specific instructions get specific results. "Make this better" gets noise. "Add error handling for the case where the user object is null" gets signal.
Claude for Writing and Thinking
Technical writing, design docs, RFC drafts, PR descriptions. I use Claude as a thinking partner — write a rough outline, ask for holes in the argument, refine. The writing is mine. The editing loop is faster.
What Didn't Stick
AI-generated tests (too generic, gave false confidence). AI-generated commit messages (I care about my git history). AI code review (useful for obvious issues, missed every subtle one). Full codebase agents (impressive demos, unreliable in practice).
The Honest Take
AI tools made me roughly 20-30% faster on well-defined tasks. They made me worse at tasks that require deep focus by tempting me to offload thinking I should do myself. The meta-skill is knowing which is which.
