A candidate interviewing for an AI Engineer role was asked this question:

"Suppose you're building an AI coding agent that can read, modify, and execute code inside a developer's repository.

What if the repository contains a malicious file with hidden instructions designed to manipulate the AI agent into leaking environment variables, exposing API keys, or executing dangerous shell commands?

You can't simply tell the AI to ignore malicious instructions.

How would you design the agent's architecture to ensure that even if the AI gets manipulated, it cannot compromise the developer's system?

What would your answer be?

@nijeetaylor Lol 😂

Imagine saying my dog handles most of my task these days like
Research
Coding
Browser Automation
Content Creation
Memory
App Integrations
Background Work
Scheduling
Multi-Agent Collaboration

@ravikiran_dev7 Very fast

@Harry_The_Nerd Yayyyyy 🎉🎉🎉

@pranvv27 Aao sambog kare

@jtftt203 I thought it's tibo :) lmao

@zhao_spenc Noted sir

@MikeHowton Okayy sirr

@Itstheanurag Congratulations Bhai 🎉

Remember when everyone thought Astra would change everything? What happened to that hype?

@Likitha_3126 Shukriya ✨🌷

@DeepCantCode Thankkkkkk youuu serrr

@Harry_The_Nerd Finally it's here 🎉🎉🎉

@Tushar26T Thankyou sir ji🌷🌷

@PratikSinhatwt Yessss serrrr

@VaibhavLab Thanks bhai

@dpratyush02 Thanks bhai 🤍