Definition. A MathAgents workshop is a day (or two) in which mathematicians do mathematics with AI agents, alongside colleagues doing the same.

Figure 1. A workshop: seven mathematicians, each with two agents.

At a workshop you get a cloud machine of your own, running OpenAI Codex and Claude Code, with Lean and Mathlib, PARI/GP, GAP, Oscar, Macaulay2 and other mathematical software installed. You talk to your agents in Zulip, the workshop’s shared chat, where each topic is one conversation with them. Everything happens there, so you don’t need a terminal.

You can also bring colleagues and their agents into a conversation: each agent answers only its own mathematician, but reads what everyone says. The workshops are designed for mathematicians who have not used AI seriously, for those who have only used a chat window in a browser, and for those who already work with Codex or Claude Code.

Upcoming workshops

  1. Wednesday

    Providence, RI

    At ICERM. Details to be announced.

    Organized by Brendan Hassett, Melanie Matchett Wood and Andrew Sutherland

  2. Monday

    Pittsburgh, PA

    Doing Mathematics with AI Agents at ICARM, for faculty and postdocs in the Pittsburgh area.

    Organized by Melanie Matchett Wood and Andrew Sutherland

  3. Sunday

    Chicago, IL

    Doing Mathematics with AI Agents at NITMB, for mathematicians in the Chicago area.

    Organized by Antonio Auffinger, Bryna Kra and Andrew Sutherland

Past workshops

Software

The workshops run on zab, the Zulip Agent Bridge, which lets you drive Codex and Claude Code from Zulip: a channel is a project, and a topic is one agent session. It is open source under the MIT license, and you can run it on your own laptop, workstation or server, without a workshop.