{ "video": { "id": "3XI90A3ENSg", "title": "How a Remote Company Builds AI Fluency — Em Shreve, Automattic", "duration": 933, "upload_date": null, "channel": "AI Engineer", "source": "AI Engineer", "event_format": "" }, "analysis": { "video_id": "3XI90A3ENSg", "title": "How a Remote Company Builds AI Fluency — Em Shreve, Automattic", "one_liner": "Automattic's bet is that AI fluency comes from protected time, not handed-out tools: it runs two-week, in-person cohorts of about 45 people, split 50% facilitated workshops and 50% real project work, and expects each graduate to teach others.", "summary": "Em Shreve describes how Automattic (WordPress.com, Tumblr; 1,400+ people in 82 countries, fully distributed) runs AI enablement. Groups of about 45 people come together in person for two weeks. Half the time is workshops run by L&D, the AI enablement team and internal hosts; the other half is real work people bring with them, with peer workflow sharing and project demos at the end of each day. Graduates are expected to multiply the reach through guides, local meetups, ride-alongs, shared skills and MCPs, and a help channel. Measured gains keep growing for up to 90 days after the program, and cohorts have shipped tools the whole company now uses.", "key_points": [ "Format: two-week immersive, in-person cohorts. 50% facilitated learning (L&D team, AI enablement team and internal workshop hosts) and 50% real project work: tools, dashboards, new skills or MCPs that participants bring in. There is peer workflow sharing and project demos at the end of each day.", "Three flavours: an in-depth engineering track on applied AI; a cross-functional product, engineering and design track about how the three will work together; and workshops for non-technical groups including finance and the executive leadership team.", "Engineering curriculum progression: basics such as skills and how to create them, using MCPs, spec-driven development, and using AI responsibly and mindfully. Then creating MCPs, extending the harness through the Codex and Claude hook systems, and autonomous agents running routines in the background. It ends with applied topics: applied WordPress AI (AI features in WordPress core) and building your own personal assistant.", "The curriculum has to be updated constantly because the top model changes with each release, so they are 'not teaching six months out of date'.", "An internal MCP server called Context Automattic (also referred to as Context A8C) lets them quickly add MCP toolsets that don't exist yet, with their own governance and observability layers. Cohort projects keep extending it, so capabilities compound across the company.", "Scale: seven cohorts of about 45 people so far this year, aiming for about 600 people by year end, which is roughly two-thirds of engineers plus other functions.", "Multiplier mechanisms: a guides program (a champions program elsewhere), about 10 local AI get-togethers run by graduates around the world, 'AI ride-alongs' where people watch someone in a similar role work, a shared skill and MCP infrastructure, and an always-on help Slack channel staffed by graduates.", "Metrics: intentionality (moving from chatbot use to deliberate agentic use), flow (integrating AI into workflows rather than 'using it for AI's sake') and sharing. They also measure individual lift on skills and custom commands, MCP comfort and context management. Surveys at 72 hours, 30 days and 90 days show people keep growing after the program ends.", "Cohort-built examples: an agentic analytics system for asking questions about a WooCommerce store; a Zendesk connector on Context A8C, now adopted by teams across the company; a cross-tool agent skills directory (they don't prescribe one tool); and a guided admin support tool built with Claude Code by a support person, not an engineer, that highlights parts of the screen during help sessions." ], "takeaways": [ "Give people real, protected time to experiment and learn, and make that time the centre of the program. Handing out tools on its own won't change how people work.", "Spend about half of training time on participants' own real work, so they leave with working skills, MCPs, tools or dashboards.", "Plan to revise the AI curriculum constantly as models and tools change; don't write it once.", "Design for multiplication: turn graduates into guides, meetup hosts, ride-along presenters and help-channel responders. Build shared skill and MCP infrastructure that is tool-agnostic and that later cohorts can build on.", "Measure intentionality, flow, sharing and specific skill lift at several points after training (72 hours, 30 days, 90 days), because gains keep accruing after the program ends." ], "topics": [ "ai-enablement", "upskilling", "mcp", "agent-skills", "remote-work", "champions-programs", "learning-and-development", "agents" ], "tools": [ "Automattic", "WordPress.com", "Tumblr", "WooCommerce", "Context Automattic (internal MCP server)", "MCP", "Codex", "Claude", "Claude Code", "OpenCode", "Zendesk", "Slack" ], "quotes": [ { "text": "so our bet is that you don't just get good quality learning or work out of just handing people the tools.", "at": "00:55", "url": "https://www.youtube.com/watch?v=3XI90A3ENSg&t=55s" }, { "text": "you cannot just write this curriculum once.", "at": "04:49", "url": "https://www.youtube.com/watch?v=3XI90A3ENSg&t=289s" }, { "text": "we're teaching the right thing and we're not teaching six months out of date.", "at": "05:12", "url": "https://www.youtube.com/watch?v=3XI90A3ENSg&t=312s" }, { "text": "you can't just throw a bunch of tools at them and expect their way of working to change.", "at": "14:18", "url": "https://www.youtube.com/watch?v=3XI90A3ENSg&t=858s" } ], "words": 2315 }, "summary_url": "/#3XI90A3ENSg", "transcript": { "html": "/transcripts/3XI90A3ENSg.html", "txt": "/transcripts/3XI90A3ENSg.txt", "vtt": "/transcripts/3XI90A3ENSg.vtt" } }