saw this piece on using 'ai orchestration' instead of just basic workflows. it's basically saying we should move away from fixed rules and let the agent decide if an email needs a reply or not.
>it could tell the difference between an email that needed a reply and one that didn'tsounds fine until you realize how much cleanup you'll have to do when it hallucinates a response to a client. i don't see how this works for anything
mission critical without someone babysitting every single action. anyone actually trusting these things with customer-facing tasks yet?
link:
https://zapier.com/blog/ai-agents-for-business