I tested Claude Code vs AntiGravity head-to-head this week. Same prompts, side by side, three real tasks.
Here's what actually happened:
Test 1: Mobile App Design (using Sleek) Both agents were given a URL and told to design a mobile app. Both completely ignored the actual website and just guessed what the app should look like. Neither wrote a decent prompt for Sleek. Identical level of lazy.
Draw.
Test 2: Landing Page AntiGravity auto-opened Chrome to preview its work in real time — genuinely cool. Gemini's motion graphics were slicker too. But both got the pricing wrong. Neither researched the site. Both produced something mid compared to what you'd get with a proper prompt.
Draw.
Test 3: Real Backend Code This is where it split. FastAPI + Supabase auth — both finished in under 3 minutes. But Claude Code built more endpoints, proper user schemas, working auth. AntiGravity gave me a skeleton with no user profile, network errors, and zero working buttons.
Claude Code wins by a mile.
The real takeaway: The tool didn't matter as much as the prompt. Both agents were terrible at writing their own prompts. The best results I got all week were when I gave proper context and used skills correctly. Your prompts are still doing 80% of the work.
3 things to take from this:
Don't let agents write prompts for other tools — do it yourself
Use skills (like Sleek for mobile design) but feed them real context, not just a URL
For actual working code, Claude Code is the one to use right now
I filmed the whole thing side by side → HERE
Speak soon Luke
