Alan says its internal AI agent, Hopper, is now used by 80% of the company, including people in engineering, product, design and data. The notable part is not just the adoption figure: Hopper has grown beyond codebase questions into a tool for non-engineers to analyse data and build prototypes.
Watch Desk analysis
What happened
In a post published on 20 November, Alan describes Hopper as a single agent connected to tools, skills and services through an MCP Gateway. The company says that gateway manages permissions, approvals and credentials. Hopper also creates shared sessions, sites and dashboards that persist beyond individual chat windows, with a default of “radical transparency”.
Alan says the agent’s use has spread across four kinds of work, but the post does not break down adoption by team or explain how the 80% figure was measured. Read Alan’s account.
Why it matters
This is a concrete example of an internal coding agent becoming a broader workplace tool. Persistent shared work and access to company systems may make an agent useful beyond its original coding niche, while also making permissions and oversight part of the day-to-day product design.
The 80% figure suggests wide reach inside Alan, not proof of time saved, better work or successful deployment elsewhere. The practical question is what people can safely accomplish with Hopper, and how the company manages the access that makes those tasks possible.
Our read
The interesting shift is from “ask the agent about the code” to “use it to get work done across teams”. Hopper’s shared sessions and managed gateway are the more revealing details: an agent’s usefulness depends not only on its answers, but on what it can access and what colleagues can carry forward.
That is a promising pattern, though adoption is not an impact study. The useful next evidence would be what tasks staff actually use it for and how Alan evaluates the results.
What to watch
- Whether Alan shares examples of Hopper’s use across non-engineering teams.
- How the MCP Gateway handles permissions, approvals and credentials in practice.
- Whether the company reports measured effects on work quality, speed or cost.
Discussion spark: Should workplace AI agents be built around broad access to shared company tools, or should teams have to grant permissions task by task?
Sources and evidence
- We built an internal coding agent. Now 80% of the company uses it (20 November 2026, 08:58 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.