Build-a-Thon Saturday Series
Every Saturday. Online. 30 minutes of concept. 3.5 hours of building. Ship before you leave.
Every session ships something real.
The format started at ClawHacks in May — 20 seats, 106 people on the waitlist, 14 builders deployed working AI agents in six hours. Not a demo. Not a tutorial. Something that ran.
The session closes with a deployed URL — not a localhost demo, not a notebook that dies when you close the terminal. A real thing, accessible to anyone, built in 4 hours.
"106 on the waitlist. 20 seats. 14 builders deployed working AI agents in 6 hours — not a demo, something that ran."
ClawHacks — May 2026 session resultsFour hours that actually produce something.
Every session follows this structure. The build sprint is 3 hours. The deploy is non-negotiable.
Just enough theory to build. One mental model, one live example. No slides for their own sake.
The skeleton. Everyone gets the core scaffold working. I stay on call throughout — no one gets stuck alone.
The intelligent layer. The moment the thing starts working. The room changes when the first person gets their agent to respond correctly.
Deploy to a real URL. 3–4 attendees share their screen. Everyone leaves with something they built that runs.
Where this leads. Q&A. The programs and resources for going deeper after the session.
Warm up before the build sprint.
These work in regular Claude.ai right now. Run them before any Saturday session.
I want to build an AI agent that [does X for Y use case]. Help me design the architecture: what tools does it need, what data sources should it access, what are the 3 most likely failure modes, and what's the simplest version I could build and deploy in 4 hours?Run this before any build session. Having a clear architecture before you start saves 90 minutes of debugging later.
I want to build a Claude MCP connector for [API / tool / product]. The API does [describe]. The most useful thing Claude could do with it is [describe the top action]. Walk me through the minimal tool definition for this action — name, description, input schema, and what a correct response looks like.For the July 19 session (Your First MCP Server). Run this the Friday before so you arrive with a clear target.
Build it in 4 hours. Ship it to Claude's directory in 4 weeks.
The July 19 session builds your first MCP server. After that session, you'll have a working connector connected to one API. The next step is getting it listed on Claude's directory — where enterprise users discover tools. I help with that too.
Subject: I just built an MCP server in a 4-hour session. You should list [Product]. Hi [Product] team, I attended a Build-a-Thon session and built a working MCP server connected to an API from scratch in 4 hours. The Claude connector directory is 6 months old. Your category: [check and fill in]. Products listed get surfaced when users ask Claude for tool recommendations. First listing in a sparse category becomes the default. I know the team that does this end-to-end: architeqt.ai — [Your name]
30 minutes to configure. Results visible in week one.
Get notified before each session.
I'll DM you the Luma link before each Saturday session. No calendar invites, no newsletter — just a message when the next one is live.
Running your own Build-a-Thon?
If you run a platform or community and the format makes sense for your audience, I'm open to partnerships. The brief is on my end.