LLMs get things wrong all the time. We’ve invested heavily in our composite model pipeline[1] that increases correctness, code patching success rates, and security. On the cloud infrastructure side, Vercel can create Functions as easily as it can create client pages, which means we can put code where it belongs. Code that touches user data belongs server-side. These are the foundations that I strongly believe over time will make “automatic coding” more secure than the human-made variety 🙂 [1] https://t.co/nBY12q2SSj