# Screensharing Kevin Rose's AI Workflow/New App Source: https://www.youtube.com/watch?v=QPAy9R9V1rA Recap page: https://rapidrecap.app/video/QPAy9R9V1rA Generated: 2026-02-02T19:33:33.529+00:00 --- ## Quick Overview Kevin Rose showcased his personal AI workflow, which he runs on cloud code and Vercel, detailing the development of a new, unreleased product called "Nylon," an AI-powered news aggregator designed to find novelty and signal in fast-moving tech news by using vector embeddings and custom editorial scoring engines like the "gravity engine." **Key Points:** - Kevin Rose is building a new product called "Nylon," which functions as an AI news aggregator focused on finding signal in fast-moving AI topics, explicitly stating it is "not meant to be a competitor" to platforms like TechMeme. - The workflow relies heavily on ingesting data from 63 RSS sources and processing articles using external services like iFramely for metadata, Firecrawl for crawling, and Gemini as a fallback for rich data extraction. - Rose emphasizes the importance of vector embeddings stored in Postgres for nuanced information retrieval, noting that this allows distinguishing between similar concepts, unlike old keyword search, stating: "The beautiful thing about what we have today with our understanding of linguistics and around using vector embeds and algorithms on top of that is that you can say there is a difference even though they're both have the same type of keywords but there is a huge difference between Apple sues Google and Google sues Apple." - For reliable execution and retries when integrating AI APIs, Rose uses trigger.dev for durable, cloud-based TypeScript functions, noting that Vercel's new beta Workflows offer a free, integrated alternative. - He developed a custom scoring mechanism called the "gravity engine" which assigns editorial scores based on dimensions like industry impact (90% for the Nvidia/Coreweave story example), novelty (important for finding early trends), and builder relevance. - Rose believes the future developer skill will be refinement, stating: "the future engineer and the future developer, the future product builder here, it's not going to be uh what you build as much as what you don't build." - Rose revealed that he has aphantasia, an inability to form mental images, which explains past struggles with code syntax retention, and now AI fills these deficiencies, allowing him to focus on creativity and core concepts. **Context:** The video features an interview and screenshare session with iconic entrepreneur Kevin Rose, where he takes the host through his current artificial intelligence development process and unveils a new personal project named "Nylon." This project is an attempt to replicate and enhance the functionality of news aggregators like TechMeme, focusing specifically on surfacing important, novel information within the rapidly evolving technology and AI space, using a custom stack built around modern cloud tools. ## Detailed Analysis Kevin Rose presents his extensive AI workflow, built using tools like cloud code and Vercel, to create a personalized news aggregation system called Nylon. The system ingests data from 63 diverse RSS feeds, processes articles by using a sequence of external APIs (iFramely, Firecrawl, Gemini) to select the 'winner' for the best metadata and content, and stores rich data including vector embeddings in a Postgres database. Rose champions vector embeddings for capturing linguistic nuance beyond simple keyword matching, allowing the system to build meaningful clusters of related stories, such as the 47 stories clustered around the Nvidia/Coreweave investment. To ensure durability and handle API failures, he utilizes trigger.dev for managing asynchronous tasks, which provides automatic retries for critical steps like generating a vector-based TLDR. A central feature is the 'gravity engine,' an editorial scoring system that evaluates clusters based on subjective criteria like industry impact, novelty, and technical depth, helping him prioritize trends he believes will become significant over time, even if they seem silly initially. Rose frames building in the AI age around iteration, stating failure is essential, and notes that AI fills technical gaps, particularly for him as someone with aphantasia, democratizing the ability for solo builders to create complex software. ### Kevin Rose's AI Philosophy - Building is about refinement, not creation; "what you don't build" is the real skill - Failure is the best part of learning, enabling more shots on goal - Focus on building personal software that solves individual needs. ### The 'Nylon' Aggregator Architecture - Ingests 63 RSS sources into Postgres - Uses iFramely, Firecrawl, and Gemini to enrich data, selecting the 'winner' for best metadata - Employs vector embeddings for nuanced clustering of articles. ### Workflow Durability and Tooling - Uses trigger.dev for reliable execution and free retries on failed AI tasks - Mentions Vercel's AI gateway for easy model swapping - Notes that Firecrawl offers stealth modes for difficult-to-scrape sources. ### The Gravity Engine Scoring - Custom editorial score based on multi-dimensional analysis - Evaluates stories on Industry Impact (e.g., 90% for Nvidia/Coreweave story) and Intellectual Gravity (Novelty) - Can detect patterns suggesting AI-reworked paid sponsorships based on vector similarity and timing. ### Product Development Process - Builds feature by feature based on gut instinct and addressing what platforms like TechMeme miss - Uses AI as a sparring partner for choosing between competing algorithms (e.g., selecting Cluster V2 over V1) - The goal is to cut 90% of features to find the 10% that is truly useful. ### Personal Revelation and AI Impact - Rose discovered he has aphantasia, explaining past coding difficulties, which AI now overcomes by filling technical deficiencies - He advocates for building things users want, even if they crash under load, because fixing scaling issues is easier than finding demand.