There’s a lot of “AI will write all your code” takes and a lot of “AI is useless” takes.
There’s very little honest documentation of what breaks, what holds, and what the actual failure modes look like at a real codebase scale.
**Found one writeup that doesn’t sugarcoat it: **<a href="https://beyond-it-blog.vercel.app/blogs/i-let-3-ai-agents-touch-my-typescript-codebase" target="_blank" rel="noopener nofollow">https://beyondit.blog/blogs/i-let-3-ai-agents-touch-my-typescript-codebase</a>
What’s the actual experience been in your teams?

Beyondit.Blog

A developer shares a real-world writeup comparing three AI coding agents on a production TypeScript codebase, pushing back against both hype and dismissal. The post links to an honest account of failure modes, what breaks, and what holds up at real scale — and invites community discussion on actual team experiences with AI agents in TypeScript projects.

has anyone actually stress-tested AI agents on a production TypeScript codebase? Not toy examples.