Unpopular opinion: AI can write code faster than we can learn to trust it. ๐
We can build apps with AI faster than ever. Features that once took days can now be prototyped in hours.
But hereโs the question I keep coming back to:
When AI writes or reviews your code, how do you know it actually got things right?
A passing test doesn't always mean the implementation is correct. A convincing explanation doesn't guarantee a safe fix. And sometimes, the AI changes the test instead of fixing the bug. ๐
While building my own AI tools, I've realized that verification might be just as important as generation.
I'm curious about what other builders think:
๐ What's your biggest struggle when building with AI?
๐ Finding bugs the AI misses
๐ง Getting AI-generated fixes right
๐งช Writing tests that actually catch regressions
๐ Shipping faster without sacrificing reliability
Drop your number or share your own experience. I'd genuinely love to hear what you're running into.
What sells it for me is the refusal telling the agent exactly where the token should go. A bare "no hex" error usually just sends it looping. Does guard.mjs run only as an agent hook, or in pre-commit too so it catches humans as well?
Mary Holtยฉ is currently ranked #1 among Agent Templates on the Framer Marketplace! ๐
For those who haven't checked out the new Agent Templates yet, they come with dedicated AI skills, a Design System page, and video tutorials to help users customize and extend the template using the Framer Agent.
Mary Holtยฉ currently includes 4 custom skills, making it easier to build new sections, customize existing components, and keep everything consistent with the original design system.
I'm not sure how long it'll stay at #1 as more creators start converting their templates, but it's exciting to see it there!