{
  "video": {
    "id": "UcYoMg-8-L8",
    "title": "500 people vibe-coded for 30 days. I was one of them. - Sanja Grbic, Automattic",
    "duration": 1069,
    "upload_date": null,
    "channel": "AI Engineer",
    "source": "AI Engineer",
    "event_format": "Online"
  },
  "analysis": {
    "video_id": "UcYoMg-8-L8",
    "title": "500 people vibe-coded for 30 days. I was one of them. - Sanja Grbic, Automattic",
    "one_liner": "A product designer's account of Automattic's \"Radical Speed Month\" — ~500 people pausing the roadmap for 30 days and starting ~794 projects in two-person teams — and how shipping three of them with Claude Code moved her from designer to design engineer and made the prototype, not Figma, the main design artifact.",
    "summary": "Sanja Grbic, a product designer of 5 years on Automattic's Jetpack design team, describes a company-wide experiment: pause roadmap work for 30 days, pair with a partner, and ship something real, with AI welcome but not required. About a third of the ~1,400-person company — roughly 500 people — took part in the first round and started around 794 projects; she built three. Her argument is that the speed didn't come from AI alone but from the surrounding conditions: a two-week role-specific AI enablement course, an internal MCP server exposing decades of documented company knowledge, security/dev-environment docs good enough for a non-engineer, and above all the agency to own a whole project without negotiation and handover. The organizational lesson she draws is that in large companies the bottleneck is human behaviour, and engineers' highest-leverage move becomes enabling and teaching others rather than writing more code themselves.",
    "key_points": [
      "Radical Speed Month: 30 days, roadmap work paused, two-person teams with full autonomy to build and ship; AI encouraged but not mandatory. First round was about a third of Automattic (~500 people) starting ~794 projects.",
      "Prerequisites that made it work were already in place: a two-week immersive AI enablement course designed per role for every employee, systems/ops security and documentation good enough that a non-engineer could spin up a working dev environment, and \"context AC\" — an internal MCP server giving AI tools access to the fully-distributed, asynchronous company's decades of written knowledge.",
      "Project 1 — a board game session manager built in 2 hours by a group of four (two designers not comfortable in code, an engineer, a product lead) during the in-person training: create/view/join/manage sessions, start and stop a session, see available games and spots, and chat. Built with Claude Code only; Nano Banana produced the 16-bit illustration of their actual office. The engineer set the repo up in GitHub and taught the group basic commands and how Git actually works.",
      "The speaker had reviewed PRs and played with Claude Code but had never built and shipped anything in code at work before this month.",
      "Project 2 — a design system status tracker, ~2.5 weeks. Discovery was three people (her, an engineer, a design ops person); the engineer raised performance, maintenance and feasibility doubts, so she built the full idea as a proof of concept alone. It pulled links from the GitHub repo, Storybook and Figma, sorted by status, tagged by library name, had working search and live component previews. Prototype in one week; another week and a half to rebuild with fixes, set up hosting, and deploy it both as an internal tool and as a public-facing site with curated data.",
      "Project 3 — in 6 days, with a fellow designer: a fully working iOS chat proof of concept for WooCommerce merchants, including wordpress.com authentication, a Jetpack connection, a site widget that inherits the site's theming, and an AI agent that scans the site for information, answers visitors in real time, and discerns whether it can answer a question or not.",
      "Process inversion: the prototype became the main artifact. Instead of starting in Figma and working to high fidelity, she planned, built the product, and only then went back into Figma for a mood board and UI fine-tuning — or to hand Claude Code an image for things she couldn't explain in words.",
      "On project 3 they started exploration as a Claude Code project folder where all chats and ideas were recorded into the file system; she says this significantly sped up both collaboration and the later build, and she has kept the practice in daily work.",
      "AI velocity looks very different in a 1–3 person zero-to-one team versus a large organisation with 5–20 year old software and thousands of people; the speaker's framing is specifically about pushing code to production in a well-established large system with serious onboarding and security practices."
    ],
    "takeaways": [
      "If you're the strongest person on a mixed-ability team, spend your time enabling: set up the repo, teach Git and the basic commands, and unblock others — that impact can exceed the engineering you'd have done yourself.",
      "Don't just hand out AI tools. Pair access with role-specific training, find your enablers and champions, carve out explicit space for experimentation on something with no risk attached, and give people the agency to make the decisions — changing tools in a big org means changing the human behaviour behind them.",
      "Try inverting the design artifact order: plan, build the working prototype, then return to a visual tool for mood-boarding and fine-tuning, using images only for what you can't explain to the coding agent in text.",
      "Keep exploration in a coding-agent project folder so chats, ideas and decisions land in the file system — it speeds up both collaboration and the eventual build.",
      "Make your design system's state legible to AI tools as well as humans; a tracker that pulls from the repo, Storybook and Figma and sorts by status is buildable as a proof of concept in weeks, by a designer.",
      "If you're in a large organisation and can't run this company-wide, run it at your team level."
    ],
    "topics": [
      "vibe-coding",
      "design-engineering",
      "ai-enablement",
      "org-change",
      "prototyping",
      "design-systems",
      "mcp",
      "agents"
    ],
    "tools": [
      "Claude Code",
      "Nano Banana",
      "Figma",
      "GitHub",
      "Storybook",
      "MCP",
      "Automattic",
      "WordPress.com",
      "Jetpack",
      "WooCommerce",
      "Tumblr",
      "Beeper",
      "Sketch",
      "Adobe suite"
    ],
    "quotes": [
      {
        "text": "We had about a third of the company participating in the first round. That's around 500 people that started around 794 projects in these 30 days.",
        "at": "04:40",
        "url": "https://www.youtube.com/watch?v=UcYoMg-8-L8&t=280s"
      },
      {
        "text": "If you're an engineer, the impact that you have when you enable others may be far greater than the impact of doing more engineering yourself.",
        "at": "08:54",
        "url": "https://www.youtube.com/watch?v=UcYoMg-8-L8&t=534s"
      },
      {
        "text": "This is of course enabled by AI, but most of all, it was enabled by the fact that I could own the entire process.",
        "at": "12:44",
        "url": "https://www.youtube.com/watch?v=UcYoMg-8-L8&t=764s"
      },
      {
        "text": "Changing processes or tools in large organizations requires shifting the human behavior behind them.",
        "at": "17:22",
        "url": "https://www.youtube.com/watch?v=UcYoMg-8-L8&t=1042s"
      }
    ],
    "words": 3080
  },
  "summary_url": "/#UcYoMg-8-L8",
  "transcript": {
    "html": "/transcripts/UcYoMg-8-L8.html",
    "txt": "/transcripts/UcYoMg-8-L8.txt",
    "vtt": "/transcripts/UcYoMg-8-L8.vtt"
  }
}