Our summary
Claude Code, Codex and Cursor are AI helpers that write software for people. They can build apps that ask Jev questions, but Jev cannot take over the AI inside them. Flavio Copes explains why these helpers often use Jev wrongly: they repeat habits learned from older code written for chatbots.
First, he adds TypeSafe's skill, a short set of instructions the helper reads whenever a task involves Jev. Next, he lets the helper read TypeSafe's own guide pages, then asks it to describe a plan before it writes anything. A small notes file in the project keeps reminding the helper of the rules.
The skill leaves out some exact details, so a person still needs to check the helper's work. Copes suggests checking that related questions go to Jev together, that sums and dates are worked out elsewhere, that a person approves risky actions, and that the app's secret password for Jev is never shown to visitors.
Key takeaways
- Jev goes inside the app a coding helper builds; it does not replace the helper's own AI.
- Give the helper TypeSafe's instructions and ask it for a plan before it writes code.
- Check the work: related questions sent together, sums and dates done elsewhere, a person for risky actions, and a hidden password.