---
title: "Comment by Julian Neagu on \"I Sent an AI to Order My Coffee. The Restaurant's AI Answered.\" | Hashnode"
description: "the biggest takeaway for me is that prompts cant protect the final action if the model can speak the effect directly then thats where the real guardrail needs t"
url: "https://hashnode.com/posts/i-sent-an-ai-to-order-my-coffee-the-restaurant-s-ai-answered/6a9faa3a125cdbdd134ad862/comment/6aa297a2da0e6b494718d77c"
---

[Discussion](https://hashnode.com/posts/i-sent-an-ai-to-order-my-coffee-the-restaurant-s-ai-answered/6a9faa3a125cdbdd134ad862)

Comment [I Sent an AI to Order My Coffee. The Restaurant's AI Answered.](https://hashnode.com/posts/i-sent-an-ai-to-order-my-coffee-the-restaurant-s-ai-answered/6a9faa3a125cdbdd134ad862)

[Julian Neagu](https://hashnode.com/@julianneagulab)

[Sep 10](https://hashnode.com/posts/i-sent-an-ai-to-order-my-coffee-the-restaurant-s-ai-answered/6a9faa3a125cdbdd134ad862/comment/6aa297a2da0e6b494718d77c)

the biggest takeaway for me is that prompts cant protect the final action if the model can speak the effect directly then thats where the real guardrail needs to live