Don't just roll out tools — clear two full weeks, half workshops and half real work, and even non-engineers come back with tools people use
How a Remote Company Builds AI Fluency — Em Shreve, Automattic — AI Engineer · Read the original ↗
This page is a translation of the Korean original. Read the Korean original →
What you get
- Ways of working that tool rollouts alone never changed turn into real project deliverables
- Graduates run guides, local meetups and an always-on help channel, spreading learning across a distributed company
- Outputs accumulate in an internal MCP server and a shared skills directory, raising the starting line for the next cohort
When to use it
Use the same format not only for advanced engineering programs but also for cross-functional courses that mix product, engineering and design, and for workshops with non-technical groups such as finance and executives. In a distributed company, extend the reach with regional meetups and an always-on Slack help channel.
How to do it
- Run a two-week in-person immersion program: 50% facilitated learning led by internal workshop facilitators, 50% real work projects that participants bring with them.
- Start the curriculum with basics such as building skills, using MCP and spec-driven development, then move step by step to building MCP servers, the codex/claude hook systems, autonomous agents and applied topics.
- Models and tools change every week, so do not write the curriculum once and leave it; keep updating it so you are never teaching six-month-old material.
- Run your own internal MCP server and connect tools that lack native MCP or need a governance and observability layer, so capability accumulates.
- Turn graduates into multipliers who run a guide (champion) program, local meetups, AI ride-alongs, a shared skills and MCP repository, and an always-on Slack help channel.
What to borrow
The idea of designing training not as a one-hour session but as 'bring real work and finish it within the program' is worth copying right away. Two more points can go into your own training design: keep the curriculum updated, and track people 30 and 90 days later, not just at graduation, to see real change. It also helps to build first the structures, such as an internal MCP server and a shared skills directory, that let outputs carry over to the next cohort.
Links
Background — what the original covers
Automattic, which runs WordPress.com and Tumblr, is a fully remote company of more than 1,400 people spread across 82 countries. Em Shreve, who leads AI enablement, has been running immersion programs that bring people together in person for two full weeks, with seven cohorts of about 45 people this year alone. The talk documents the curriculum, the internal MCP server, the mechanisms for spreading adoption after graduation, and the metrics.