Figure 1. Each topic is a stone, numbered by layer, with applications at the top, energy at the bottom and repo dives on the right. Move 37 is red, for the club’s name.Rows from top to bottom are applications, models, hyperscalers, chips and energy. On the right are repo dives, build sessions and investing.

Agent 37

A small club to learn AI together, from the agents we use every day down to the models, data centers, chips and energy underneath.

Each session we took a repo or an idea and worked through it together. These are those materials.

These materials are no longer updated, and like most things in AI, some may be out of date. Use them with caution.

If you read three

The record

Grouped by layer, top of the stack first. Numbers match the stones on the board.

Nothing matches that. Try a shorter word.

Applications

Agents, coding tools and the products built on them.

Models

How models are trained, served and scaled.

Hyperscalers

The data centers and networks they run on.

Chips

GPUs, memory, and the fabs that make them.

Energy & materials

Power, copper and the supply chains underneath.

Build sessions

Hands on sessions for setting up and shipping with AI.

Investing

Reading the stack as an investor.

25Investment Frameworks

Repo dives

Open source projects, read closely one repo at a time.

About

Agent 37 started in early 2026. We had been using AI for coding at work for years (heavy Cursor users since 2024), but we didn’t really understand how things worked under the hood, and we didn’t have time to keep up with every new development. So a few colleagues and I started meeting after work to understand AI properly.

We used agents to help us build slides, learn new open source repos and techniques, and guide us through building projects, so we could understand the stack better. We began at the top, with the coding agents we used every day, and worked down one layer at a time through models, data centers, chips and energy. Everything at the top is constrained by what sits below it, and we wanted to learn the whole stack.

For each session we picked a repo, a paper or a question, had AI read the source and draft a deck, and then went through it together. Some topics also have a plain language ELI5 version. Afterwards we would discuss how to apply what we had learned.

The site is no longer actively maintained. We still drop by from time to time, and it might turn into something more in the future.

The name comes from move 37 of AlphaGo’s second game against Lee Sedol in 2016, a move no professional would have played and the one people still talk about.