Our summary
This project provides instructions for building AI assistants that handle tasks like sorting incoming messages or browsing websites. People use this approach to make their software run faster and cost less by splitting the workload between two different types of AI.
The software uses a fast tool called Jev to pick from a set list of options, like choosing which button to click next. It leaves complex jobs, like writing text or planning the next steps, to a standard AI assistant. A shared connection links both tools to the task.
This guide is useful for people testing different AI tools on the exact same tasks. The speed and cost results happened in the author's specific tests. The software requires a private access code to run, and it cannot handle typing tasks without a standard AI assistant.
Key takeaways
- Use a fast AI tool for multiple-choice steps and a regular AI assistant for writing text.
- Test different AI tools fairly by recording and replaying them on the exact same tasks.
- A shared connection lets you swap out the AI tools without rewriting the whole program.