Bolt.new

I tested Bolt.new (StackBlitz) by prompting one-shot prototypes for three different specs: a SaaS pricing page with toggle, a kanban board with drag-and-drop, and a small dashboard with chart and table. The cohort I had in mind was non-developers and PMs who want to spin up a working visual prototype in minutes for stakeholder review, distinct from Lovable which leans more full-stack web app builder with auth and database.

The time-to-first-rendered-prototype is genuinely impressive. The pricing page request returned a working React+Tailwind page in about 90 seconds with the toggle wired up and visually decent. The full app runs in WebContainer in the browser, so I never had to download anything or set up a local stack to see the result, which is a real productivity unlock for anyone who would otherwise be blocked at "npm install" friction. Real-time editing in the chat-driven flow is fast: I asked for a copy change on the hero, and 8 seconds later the preview reloaded with the new text. One-click deploy to Netlify worked first try and gave me a shareable URL.

What broke down was anything past the initial scaffold. The kanban board worked visually but the drag-and-drop logic had race conditions where dropping mid-animation duplicated cards. When I asked Bolt to fix the bug, it rewrote the entire DnD component twice without resolving it, and the third iteration introduced a new TypeScript error that took me 10 minutes to manually patch. State management beyond simple useState gets unreliable: the dashboard chart filtering broke whenever I added a new filter dimension, and Bolt could not consistently remember the data shape across edits. The token cost also adds up quickly on iterative debugging sessions where the model is essentially regenerating large files every prompt.

Bolt is the right pick if you need a one-shot visual prototype for stakeholder review or a static-leaning marketing site that can be deployed in 10 minutes. Skip it for production apps with auth, database persistence, or complex business logic (Lovable handles that better with Supabase integration), iterative debugging beyond 2-3 rounds (token cost spirals), or anything where the working code needs to graduate into a long-term codebase a real engineering team will maintain.