# How a Remote Company Builds AI Fluency Em Shreve, Automattic | AI Engineer | 15:33 Source: https://www.youtube.com/watch?v=3XI90A3ENSg Channel: AI Engineer (https://www.youtube.com/@aiDotEngineer). Summarised by AIE Talks. Page: https://aietalks.com/talks/how-a-remote-company-builds-ai-fluency Published: 2026-10-08 Tags: agent-skills, agents, mcp, team-adoption ## TL;DR - Automattic gives employees two weeks of dedicated, in-person time that combines workshops with real project work. - The program trains graduates to spread practical knowledge through guides, local meetups, ride-alongs, shared skills, and an always-on help channel. - Automattic measures intentionality, workflow integration, sharing, skills, custom commands, MCP comfort, and context management over surveys taken up to 90 days later. ## Summary Em Shreve explains how Automattic helps a distributed workforce learn to use AI. The company's main bet is that access to tools is insufficient. People need protected time to experiment, learn, and apply AI to work they already need to complete. Its two-week programs divide time between facilitated workshops and real project work, with daily demos and peer sharing. The programs cover engineers, cross-functional groups, finance, executive leadership, and other non-technical teams. The engineering curriculum moves from basic skills and responsible use through MCPs, spec-driven development, hooks, autonomous agents, and applied WordPress AI. It changes often because the tools and models change. Graduates then extend the program through guides, regional meetups, AI ride-alongs, shared skills, and a Slack help channel. Shreve also describes projects built during the programs, including an agentic WooCommerce analytics system, a Zendesk connector, a shared skills directory, and a guided admin support tool built by a support teammate. ## Key ideas ### Dedicated learning time gives people a way to change their work [00:37](https://www.youtube.com/watch?v=3XI90A3ENSg&t=37s) Automattic's central bet is that handing people AI tools does not produce good learning or better work by itself. The company brings participants together for two-week, immersive programs where they can experiment and apply what they learn. Half of the time goes to facilitated learning led by the learning and development and AI enablement teams. The other half goes to real work, such as building a tool or dashboard for daily tasks, or creating skills and MCPs that people can use after the program. ### The programs combine practical work with peer teaching [01:46](https://www.youtube.com/watch?v=3XI90A3ENSg&t=106s) Participants bring real work into the program instead of completing isolated exercises. They share their workflows so colleagues can see what is possible, and the group holds project demos at the end of each day. Automattic runs versions of the program for engineers, cross-functional groups made up of product, engineering, and design, and non-technical teams such as finance and executive leadership. The format lets people discuss how their roles may change together while they build something useful. ### The engineering curriculum moves from basic skills to autonomous systems [03:25](https://www.youtube.com/watch?v=3XI90A3ENSg&t=205s) The engineering track starts with basic skills, how to create them, MCPs, spec-driven development, and responsible AI use. It then covers creating MCPs for new functionality, using the hook systems in Codex and Claude, and running autonomous agents and background routines. The program ends with applied work, including WordPress AI features and creating a personal assistant. Shreve describes this as a progression from foundations to making new tools for themselves. ### The curriculum has to change as the tools change [04:31](https://www.youtube.com/watch?v=3XI90A3ENSg&t=271s) Automattic cannot write the curriculum once and leave it alone. AI tools and leading models change from day to day and week to week, so the team updates the material continually. Shreve says this prevents the programs from teaching practices that are already six months out of date. Automattic also has an internal MCP server called Context Automattic, which lets the company add tooling that may not exist elsewhere and add its own governance and observability layers. ### Graduates extend the program across a distributed company [06:19](https://www.youtube.com/watch?v=3XI90A3ENSg&t=379s) Automattic wants each participant to become a multiplier who can help colleagues the central team cannot reach. Shreve says the company has run seven cohorts of about 45 people and aims to put around 600 people through the programs by the end of the year, including about two-thirds of its engineers. Graduates can join a guides program, run local AI meetups, and teach colleagues in their regions. Automattic has held about 10 of these local events. ### Seeing a peer work makes AI practices easier to adopt [08:09](https://www.youtube.com/watch?v=3XI90A3ENSg&t=489s) AI ride-alongs show participants how someone in a similar role uses a tool in real work. Shreve says this is easier to grasp than discussing concepts in the abstract. Automattic is also building shared infrastructure for skills and MCPs, encouraging people to publish work that others can extend. An always-on Slack channel gives newer users a place to ask questions, with graduates helping answer them throughout the day. ### Automattic measures continued use after the program ends [09:29](https://www.youtube.com/watch?v=3XI90A3ENSg&t=569s) The program tracks whether people apply AI more intentionally, integrate it into their workflow in ways that make sense, and share what they learn. It also measures changes in skills, custom commands, comfort with MCPs, and context management. Surveys happen 72 hours, 30 days, and 90 days after the program. Shreve says participants continue gaining confidence and experience during that period as they have more time to try the techniques in their own work. ### Two-week projects include tools built outside engineering [11:44](https://www.youtube.com/watch?v=3XI90A3ENSg&t=704s) Participants have built an agentic analytics system for asking questions about a WooCommerce store, a Zendesk connector adopted by teams across Automattic, and a shared skills directory that works across Codex, Claude, OpenCode, and other tools. Another project, guided admin support, lets someone lead a help session by highlighting parts of a dashboard. A support teammate built it after becoming more comfortable through the program, rather than an engineer. ## Notable quotes - "You need to give them real time to learn and experiment." (00:37) - "We really try to walk people from the foundations up into the ability to start creating these new things for themselves." (04:31) - "We also want each person we send through to be a multiplier within the company and extend our reach to the folks that we can't get to yet." (06:19) - "You can't just throw a bunch of tools at them and expect their way of working to change." (14:07) ## Tools & references mentioned - Automattic - WordPress.com - Tumblr - WooCommerce - Context Automattic - MCPs - Codex - Claude - OpenCode - Zendesk ## Who should watch - You are responsible for AI adoption in a distributed company and need a program that gives people time to practice on real work. - Your organization has bought AI tools, but employees still use them as simple chatbots or do not know how to fit them into daily workflows. - You want examples of how to measure learning after training and how graduates can help colleagues outside the initial cohort. ## Related talks - [Dispatch from the Future: Building an AI-Native Company](https://aietalks.com/talks/dispatch-from-the-future-building-an-ai-native-company) (Dan Shipper, Every, 17:58) - [500 People Vibe-Coded for 30 Days. I Was One of Them.](https://aietalks.com/talks/500-people-vibe-coded-for-30-days-i-was-one-of-them) (Sanja Grbic, Automattic, 17:49) - [Rise of the AI Architect](https://aietalks.com/talks/rise-of-the-ai-architect) (Clay Bavor, Sierra & Alessio Fanelli, Latent Space, 18:55) - [Agents for Everything Else](https://aietalks.com/talks/agents-for-everything-else) (swyx, AI Engineer, 14:10) - [Mentoring the Machine](https://aietalks.com/talks/mentoring-the-machine) (Eric Hou, Augment Code, 19:06)