Tuesday, 3 Nov 2026
Subscribe to AIWatcher
AIWatcher
  • Home
  • News

    A British chip startup reaches a $6.5B value before shipping a wafer

    By
    AIWadmin

    Seoul puts $16B behind chips and its own frontier models

    By
    AIWadmin

    Envision pushes 800-volt power straight into a working AI campus

    By
    AIWadmin

    42AI leaves stealth with $62.5M to model how people decide

    By
    AIWadmin

    Tohoku and Aisin stash a whole model beside the arithmetic

    By
    AIWadmin

    Western startups are editing Chinese models until they sound American

    By
    AIWadmin
  • Articles

    Omen AI buys Aquatic Labs to watch what cools the racks

    By
    AIWadmin

    LG CNS hands over a 30-megawatt AI hall in Jakarta

    By
    AIWadmin

    Memory suppliers outprofit five US tech giants for the first time

    By
    AIWadmin

    AgiBot proposes a robot metric built on hours rather than units

    By
    AIWadmin

    Karpathy rebuilds YouTube in two hours and lets the model wire it

    By
    AIWadmin

    Palantir Foundry veterans land $66M to put factories on autopilot

    By
    AIWadmin
  • Spotlight

    Sema4.ai turns repeatable agent work into code to cut token bills

    By
    AIWadmin

    ChatGPT and Codex will draw from one credit pool in December

    By
    AIWadmin

    AMD’s MI355X undercuts Nvidia on one model at the cost of users

    By
    AIWadmin

    Publishers may soon write one rule that every AI crawler reads

    By
    AIWadmin

    Blackstone flips its forecast on Anthropic’s credit grade

    By
    AIWadmin

    A password stealer now hires a model to rank its victims

    By
    AIWadmin
  • About
    • Mission
    • Services
    • Contact
  • Newsletter
  • Shop
    • All Items
    • By Category
      • Hats
      • T-Shirts
    • Cart
  • 🔥
  • Alignment
  • Classification
  • Distillation
  • Explainability
  • Hallucination
  • Legal/Compliance
  • Medical
  • NLM
  • Mobility
  • Research
  • Robotics
  • Safety
  • Startups
  • Prompt
  • Python
  • RAG
  • RLHF
  • Token
  • Vision
Font ResizerAa
AIWatcherAIWatcher
  • Home
  • News
  • Articles
  • Spotlight
  • About
  • Newsletter
  • Shop
Search
  • Home
  • News
  • Articles
  • Spotlight
  • About
    • Mission
    • Services
    • Contact
  • Newsletter
  • Shop
    • All Items
    • By Category
    • Cart
Have an existing account? Sign In
Follow US
© 2022 Foxiz News Network. Ruby Design Company. All Rights Reserved.
News

Karpathy rebuilds YouTube in two hours and lets the model wire it

The engineer published a video site seeded only by Hacker News recommendations, and says he barely touched a terminal to do it.

AIWadmin
Last updated: November 1, 2026 9:02 pm
AIWadmin
ByAIWadmin
Global AI news & information.
Follow:
Share
SHARE

Andrej Karpathy spent a Friday night annoyed at YouTube’s recommendations and a Saturday building a replacement. The result, published October 31, is a video site stocked only with clips that Hacker News commenters have recommended since 2007.

Contents
Three jobs a human still had to doWhat the demo actually measures

The numbers behind it are modest but concrete. Claude Opus 5.5 pulled the site’s comment archive, ran Claude Sonnet over it inside a Claude Code workflow to strip spam, and narrowed roughly 40,000 comments down to those recommending a video. That produced 6,454 videos spanning 150 topics, wrapped in a familiar browsing interface.

Three jobs a human still had to do

Karpathy’s own account of the build is the part worth reading. He wrote that he no longer does anything resembling traditional engineering, and that his manual contribution amounted to creating a Google Cloud API key, booting a virtual private server, and buying the domain.

Everything after that was instruction. He handed over the server address and asked for the code to be deployed, secured, hardened and optimized. The site now runs without user accounts, keeps recommendation state in the browser session, and offers a data export so anyone can hand the corpus to an agent and build their own version.

What the demo actually measures

A two-hour clone is not a business, and Karpathy presents it as a personal fix rather than a product. But it functions as a datapoint on how far a single practitioner can move with a coding agent and no team.

It also surfaces the limits. The work was well-specified, self-contained, and checkable by eye, a shape that coding agents handle far better than sprawling codebases with years of undocumented decisions.

TAGGED:Agentic AIAI codingAndrej KarpathyClaudeopen sourcesoftware engineering
SOURCES:Karpathy
Share This Article
Email Copy Link Print
ByAIWadmin
Follow:
Global AI news & information.
Previous Article Palantir Foundry veterans land $66M to put factories on autopilot
Next Article AgiBot proposes a robot metric built on hours rather than units

You Might Also Like

News

Anthropic rewrites its usage rules to ban abusive chats and election meddling

By
AIWadmin
News

Meta trials robot technicians as AI data centers balloon

By
AIWadmin
News

A Delaware plan would let software run companies with nobody in charge

By
AIWadmin
News

US and China agree to a super intelligence dialogue by November

By
AIWadmin
AIWatcher
Facebook Twitter Youtube Linkedin Rss

Global AI News and Information
AIWatcher is your definitive source for AI updates worldwide, from Silicon Valley to Shanghai.
Our industry coverage keeps you in the loop with the latest news and trends shaping the future of AI.

Quick Links
  • News
  • Articles
  • Spotlight
  • Events
About Us
  • Mission
  • Services
  • Contact
  • Privacy Policy
  • Legal
© 2026 AIWatcher. All Rights Reserved.