The Honest State of AI Dev Tools
I’ve shipped two SaaS products and countless prototypes using AI-assisted development over the past year. Here’s what actually moved the needle.
Tier 1: Daily Drivers
Cursor
The single biggest productivity multiplier. Tab autocomplete, Cmd+K inline edits, and the chat sidebar with full codebase context make it irreplaceable. I write about 40% less boilerplate than I did in VS Code.
Best for: Ongoing feature development on existing codebases.
ChatGPT-4o
Still the best thinking partner for architecture decisions. I paste my schema or system design and ask “what are the failure modes here?” — it consistently surfaces edge cases I’d miss.
Best for: System design, debugging complex state, writing SQL queries.
Tier 2: Situational
Lovable AI
Remarkable for spinning up a full-stack app from a text prompt. In under an hour you can have a working React + Supabase app with auth, a dashboard, and basic CRUD. The generated code is clean enough to hand to a real engineer.
Best for: Investor demos, hackathon MVPs, validating ideas before committing.
Bolt (by StackBlitz)
Similar to Lovable but runs entirely in the browser. Faster iteration loop for pure front-end prototypes. The in-browser execution is a superpower for sharing with non-technical stakeholders.
Best for: Front-end prototypes, component libraries, design reviews.
What I Avoid
- GitHub Copilot — Cursor is strictly better at this point
- AI for tests — I still write unit tests manually; AI-generated tests are often tautological
- AI for DB migrations — too risky; always review schema changes yourself
The Mental Model That Helps
Think of AI tools on a spectrum from exploration (Bolt, Lovable) to execution (Cursor). Match the tool to the phase of the project.