---
title: "Discussion on \"Why We Treat the LLM as an Untrusted Planner\" | Hashnode"
description: "Discussion on \"Why We Treat the LLM as an Untrusted Planner\". Giving AI access to your business software is easy. Making sure it uses that access responsibly is the real challenge.\nImagine giving a new employee access to your company's database on their first da"
url: "https://hashnode.com/posts/why-we-treat-the-llm-as-an-untrusted-planner/6ad506cfe79da00754d01ea8"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Why We Treat the LLM as an Untrusted Planner](https://caterfli.hashnode.dev/why-we-treat-the-llm-as-an-untrusted-planner?utm_source=hashnode&utm_medium=feed)

Giving AI access to your business software is easy. Making sure it uses that access responsibly is the real challenge. Imagine giving a new employee access to your company's database on their first da

Oct 18 · 17 min read · caterfli.hashnode.dev  
Read the article

-   [#llm-security](https://hashnode.com/tag/llm-security)
-   [#ai-agent-safety](https://hashnode.com/tag/ai-agent-safety)
-   [#ai-tool-calling](https://hashnode.com/tag/ai-tool-calling)
-   [#ai-agent-permissions](https://hashnode.com/tag/ai-agent-permissions)
-   [#caterfli](https://hashnode.com/tag/caterfli)

## Written by

[Caterfli](https://hashnode.com/@caterfli)

@caterfli · New York City, USA

  
[View profile](https://hashnode.com/@caterfli)

Reinvent the software you already own

Meet Caterfli — your AI colleague for the systems you already use. 🚀 Caterfli connects to your databases, APIs, and MCP servers, learns what they can do, and lets your team operate them through simple conversations. Create orders, generate reports, answer questions, and automate daily tasks without replacing your existing software or rebuilding anything. 🔒 Real permissions. Human approval for destructive actions. No complicated migration. Simply connect a system, review your AI agent, and get started. We're launching today and excited to learn from teams solving real-world problems with AI. 💬 \*\*If you could give an AI agent access to one internal system your team uses every day, what would you let it handle first?\*\* Share your ideas in the comments!

## Responses

Markdown supported · ⌘↵ to post

No responses yet.