JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Python walkthrough

Turning everyday requests into specific software actions

The author tests a method to translate everyday language into specific software actions. This approach asks an AI assistant multiple-choice questions to pick the right predefined task and fill in the required details.

Original by TypeSafe AIAgent workflows

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“Function calling” by TypeSafe AI. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

This guide explores a way to help an AI assistant understand user requests, like asking to compare stock prices. Instead of letting the software guess what to do, this method forces the AI to answer specific multiple-choice questions to find the correct predefined software action.

The method asks the AI tool which specific software action matches the user's request and which exact details to include. The software then checks the tool's confidence estimate before acting. The AI answers all the questions at once, but the program only uses the details for the chosen task.

This approach is useful for developers managing a fixed list of tasks, but it has strict limits. The AI only selects from provided options and cannot authorize actions. The main software must still check permissions and validate any open-ended details before the program actually runs.

Key takeaways

  1. Turn the requirements for your software actions into clear multiple-choice questions for the AI.
  2. Ask the AI all the questions at once, but only use answers for the chosen task.
  3. Keep safety checks in your main software instead of trusting the AI to authorize actions.

Jev is an AI tool that chooses from options rather than writing an answer. Any open-ended details still need their own separate validation to work safely.

TypeSafe cookbook · Source reviewed

Read the original guide Opens the author’s site in a new tab.