{
  "id": 9785057,
  "title": "Before You Hit Send: Building a Reply Co-Pilot on TypeSafe's Jev, Deployed on Cloudways Velocity",
  "url": "https://urgent.news/2026/09/25/before-you-hit-send-building-a-reply-co-pilot-on-typesafes-jev",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-25T13:19:21.000Z",
  "source": {
    "name": "HackerNoon",
    "slug": "hackernoon",
    "url": "https://hackernoon.com/before-you-hit-send-building-a-reply-co-pilot-on-typesafes-jev-deployed-on-cloudways-velocity?source=rss"
  },
  "original_language": "en",
  "account": "Before hitting send on a drafted support reply, it's common for it to undergo one round of review by the same person who wrote it, usually in a hurry. Tone issues, unauthorized promises, and replies meant for a manager often surface later during a QA pass, even after the message has been sent. Current AI writing assistants focus on generating text rather than providing structured judgments on existing text. To address this, a fast and structured decision-making system is needed, one that can quickly assess critical aspects of the reply such as tone, potential risky commitments, the need for another review, and the predicted customer satisfaction. TypeSafe AI's Jev, released in September 2026, is specifically designed to tackle this problem. Unlike chat models, Jev operates on typed questions and returns typed, calibrated answers directly, eliminating the need for parsing lengthy text. The system consists of four separate judgments about the reply, each corresponding to one of Jev's three typed primitives: Choice, Score, and Noul, all executed as a single call against a shared state. This approach allows for parallel and independent processing of multiple questions, making the overall response time relatively cheap and efficient. The interface presents the agent with a plain-writing panel for composing the reply and a dark instrument rail displaying the four judgments. The interface also provides additional information like round trip time and Jev's response time. Testing the system with various replies, it correctly identifies tone, commitment issues, the need for review, and predicted customer satisfaction. The model takes into account more than just surface-level risk phrases, demonstrating its effectiveness in providing a comprehensive evaluation of the reply.",
  "summary": "Learn to build a reply co-Pilot on TypeSafe's Jev, deployed on Cloudways Velocity.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}