Our summary
The author explores ways to help an AI assistant find the right instructions and safely complete tasks. An AI assistant is software that carries out tasks. People use this method to organize incoming requests and decide if a person needs to step in.
The system relies on Jev, an AI tool that chooses from options rather than writing an answer. The software filters available actions based on what the user is allowed to do. Jev then picks the best option from that safe list before the software actually performs the task.
This setup is useful for teams building software that handles sensitive data or actions like spending money. The guide only offers structural advice rather than tested results. The software must always ask a person to confirm major actions, because the AI tool can still make mistakes.
Key takeaways
- Only let the AI tool choose from actions the user is already allowed to perform.
- Always use regular software rules to check the chosen action before anything actually happens.
- Ask questions one after another if a new choice depends on an earlier answer.