Pro tip: Point Jev at your inbox. @jackcheng built a Jev email filter, and the first version flagged every newsletter as urgent. He didn’t want that, so he revised the criteria with his agent: Did a human write this? Is someone I know waiting on me? Would ignoring it cost me anything? When he ran those checks across 90 days of email, older messages that still needed a reply floated back up. To build your own version: 1. Give your coding agent TypeSafe’s official skill: 2. Show it examples you’d select, reject, or hesitate over. Ask it to define what “urgent” means to you. 3. Turn that definition into separate questions Jev can assess. Specify what counts as yes or no and how to combine the results. 4. Test on a small set you’ve already judged, without giving Jev your answers. Investigate disagreements, revise the checks, then test on fresh examples before processing everything. Save the questions and examples that work. You can reuse—and improve—your definition with the next batch. Read the full piece:

How do you put together investigative reporting about AI with AI? @DanShipper used Codex, voice mode, Gemini 4 Argon, and more to put this piece together. Join Dan, @KateLaurieLee, and @JackCheng to hear about how to use AI in creating your own work, and when to leave it out. Bring your questions. RSVP here: