I love the idea of an AI writing my automation scripts, but LLMs hallucinate all the time. If the AI writes a script with a bad selector, does the whole project just crash and burn?
Do I have to manually debug the AI’s code?
It’s a very valid concern. We knew LLMs would hallucinate, so we built an automatic self-correction loop directly into the chat orchestrator.
If the AI generates a script and our Deno engine tries to run it but hits an error (like a TimeoutError waiting for a non-existent selector), the engine doesn’t just give up. It catches the stack trace, bundles it with the current state of the page, and sends a hidden message back to the LLM saying: “Your code failed with this exact error. Please fix it.”
The AI will analyze its mistake, rewrite the code, and the engine will try again. It can loop like this autonomously (up to a safety limit) until the script succeeds. It’s like having a junior developer who instantly fixes their own bugs.