Our summary
The author shares saved instructions that help an AI assistant understand exactly what a person wants. Instead of guessing or asking dozens of questions about a vague task, the software checks existing project files or customer records to find the missing details itself.
The instructions sort the request into four clear steps. The assistant searches available records for facts, asks a human only when a choice cannot be undone, and stops completely if a system breaks. Finally, it saves a clear summary file that other tools can read.
This is a design lesson for people organizing AI tasks, not a finished program. The current version relies on basic text instructions. Advanced connections to tools like Jev, an AI that chooses from options rather than writing answers, are only planned for the future.
Key takeaways
- Guide the software to check available records before it asks a person for missing information.
- Keep a clear record of why the AI assistant moved a task from one step to another.
- Test the software with graded examples before trusting its estimates about how certain it is.