First Edit

  • Led engineer working on technical design

  • Worked in a cross-functional team of editors, product managers, designers, and engineers

    I led the technical design for "First Edit," an AI editorial assistant built into our CMS that helps journalists refine drafts before submission. Working with our product team, Senior Editor of AI Strategy, and Copy Desk Chief, I researched and tested different approaches to prepare the project for engineering development.

    After building and validating prototypes, this project was put on pause due to layoffs and restructuring of our Creator Tools team.

    First Edit has four functional areas:

    1. Proper Noun Verification

    An AI agent using RAG to verify proper nouns against our newsroom's taxonomy database of names and terms. It flags errors and suggests corrections based on previously verified and published content, which helps prevent the AI from inventing false information.

    2. Content Length Guidance

    To help reporters meet assigned story lengths (T-shirt sizes), embed a collaboratively developed, content-length style guide directly into our prompts. This guide, created with newsroom editors, codifies concise writing tactics and identifies verbose patterns, ensuring LLM recommendations align with Post standards while preserving editorial control.

    3. Inline Link Recommendations

    For this feature, use an LLM agent to identify the best passages for links, paired with our existing vector store to find matching stories. This approach uses tools we already have while letting us focus on making the link selection smarter.

    4. Spelling, Grammar, and Style Guide Adherence

    For this component, we decided to use our existing Tansa integration rather than building our own solution. This choice recognizes when to build vs. buy, especially in fast-changing areas where external tools might improve faster than what we could build internally. By making Tansa-as-you-type the default, we gave users immediate help while saving our time for features that don't already exist.