Walk-up AI workstations — 24/7 lab

Dedicated office terminals (3–5 to start) preloaded with the standard agent stack — Claude Code, Codex, Cursor, the company's shared AGENTS.md, common MCPs, and a curated bookmark of skills. Open 24/7 to any employee. Goal: collapse the distance from "I have an AI idea" to "I'm trying it in 30 seconds."

The biggest adoption gap is friction. People who don't have the tools installed don't experiment, and people who almost-have-them-installed never finish. A walk-up lab makes adoption a 30-second decision. Used as a lever to push adoption from ~10% to 50%+ in target functions.

1. Buy 3–5 powerful workstations + good monitors + comfortable chairs. Place near coffee / lunch area.
2. Pre-install: Claude Code, Codex CLI, Cursor, the company's MCP stack, the shared AGENTS.md cloned, the skills repo cloned, a dictation tool, Raycast (if Mac).
3. Use a shared service account with quota limits; auto-reset session state between users (or auto-create per-user scratch dirs).
4. Print a one-page "what you can do here in 5 minutes" handout with three example prompts.
5. Run weekly "drop-in hours" — one champion sits at the lab for an hour, anyone can come ask.
6. Measure: number of distinct users per week; number of skills first-tried at the lab.

• If you let stations rot (outdated tools, broken MCP keys) they become anti-marketing. Assign one owner.
• Don't gate by reservation. Walk-up is the point.