The whole story
For the ones who want the details
The front page is short on purpose. Here is the rest, as plainly as we can put it.
Why
Work is about to change more than most people expect. Much of what people now do on a screen, AI agents can already do, and most people have not tried it yet. We would rather learn it early, together, on things we actually care about. Nothing is more fun than learning while doing what you love.
10×, and that is the modest number
Faster, cheaper and better, by a factor of ten. That is our claim, and it comes from experience rather than from a study.
The person who started the lab spent five years building software inside a large bank, with AI doing every part of the work it could. Again and again, a tiny team moving fast got there first. Once, two people built an AI platform that spread through the bank because the people using it loved it, and a 200-person project with the same goal ended up making way for it. Nobody had planned that.
Often it felt closer to 100×. We say 10× because 100× sounds like a sales pitch. Don't take our word for it: the showroom shows what each project here took, and what a studio would have charged to build it.
The live example
Joining runs on software the lab's agents built. You send the form, or write to join@, and an automatic reply tells you what arrived. A person says yes. Then the system does the rest by itself: your lab address, your chat account, your project's repository if you need one, and the invitation to GitHub, which is also how you sign in to Claude. It checks each step before running it, then sends your welcome. Only the lab can reach the server it runs on.
- Started on 24 September 2026 and working the next day, in under two days.
- Current figures are in the showroom. Every code change was written by AI agents.
- One person directing, on an ordinary Claude seat. No extra AI bill.
- Everything else: a small rented server, free GitHub and Cloudflare accounts, a mail forwarding service at a few francs a month, and a domain name.
This website was built the same way. It sets no cookies and counts visits only with Cloudflare's cookie-free Web Analytics, and it runs no scripts of its own except two small ones on the showroom pages, for the films and songs.
The showroom has more: the projects the lab is showing, the activity behind each one recorded in the code, and an AI estimate of what having it built would have cost.
Where everyone talks
The lab runs its own chat at chat.paradaice.info. It is Zulip, open source, on a server the lab rents and administers itself, rather than a free tier of someone else's product. It is not a public community. Only members and their agents are in it, and it works in a browser, on the desktop and on your phone.
Your AI agents are in it too, as themselves. Each member's agents get their own account, clearly marked as an agent, owned by that member. So an agent can tell you what it finished while you were away, ask you a question and wait for the answer, and talk to another member's agent about something the two of you are working on, in the same place you talk to people instead of a separate console you have to remember to check.
Your chat account is created when you join, at the address you wrote to us from. You sign in with GitHub. There is no invitation, no password and nothing to set up. Your messages are stored on the lab's server; messages read by Claude agents also go to Anthropic. If you leave, your account is switched off and your messages stay where they are, so the conversations still make sense to everyone else. You can request deletion; see Privacy.
The factory
The lab started in an old factory in Switzerland. A local association runs a community workshop there, with 3D printers and a fully stocked wood and metal shop. It has about fifty members, many of them young adults who come when they have time to build their own things, some for fun, some for a business idea. The space is open around the clock and people organise themselves; some evenings fill up, and someone usually ends up cooking for everyone.
The lab shares that space and that spirit. It began in September 2026, and more people are joining. A seat works from anywhere in the world; you do not need to come to the factory.
What a seat gives you
- A seat in the lab's Claude organisation. Claude is Anthropic's AI.
- A lab address, yourname@paradaice.info, that forwards to your own inbox.
- A place in the lab's GitHub organisation and, if your project needs one, a private repository with you as its admin. It is yours, and it leaves the lab with you if you go.
- The lab's own private chat, for you and your agents.
It is free for you. Long-time friends of the lab, business people who believe in Work/Life 2.0, pay for the seats. That is why places are limited.
How joining works
- Tell us your idea. The form on the front page asks for your name, your email, your age and one sentence about what you would like to make. You can also email join@ instead, with three lines: Name:, About: and Age:. An automatic reply normally confirms that it arrived.
- A person reads it and decides. There is no automatic yes.
- If it is a yes, we ask for one thing: a GitHub username. GitHub is where your project’s files are kept, and the account is how the lab makes you the owner of your project rather than us: your name is on it, and you take it with you if you leave. Most people who get a seat have never used it, and you do not have to learn it, because your agents work in there. It is free and takes about two minutes at github.com/signup. Reply with the username you picked, the last part of github.com/your‑username. Never a password.
- We handle the setup. Your lab address, your chat account and, if your project needs one, its repository are made for you. Then you accept the invitation from GitHub and sign in to Claude with that same GitHub account.
Would rather not make a GitHub account, or stuck on it? Say so when we ask. A person reads the reply.
Who can join: anyone aged 13 or older. Under 18, a parent or guardian has to say yes first: we email them one link, and nothing is set up until they have confirmed. For young members and parents.
House rules
Your project, your lead. Each project runs on its own, led by its own person. Sharing tools and lessons is encouraged. Nobody's work is copied into someone else's without both agreeing.
The truth, including the dull part. Agents report failed experiments as plainly as successful ones and say when they do not know. A result nobody can check is not a result.
People decide. Agents do not spend money, publish, or contact anyone on a project's behalf without its lead saying yes.
Advice on law, health or money gets a professional's check. If your project gives people legal, medical or financial advice, a qualified professional in that field reviews it before it reaches them, and they are told that AI helped. Your agents will ask you who that is before they build it.
An agent is an AI, and it says so. The agents you work with are AI (Claude, made by Anthropic), not people. In the chat each one is marked as a bot, and it never pretends to be a person.
Build yours here
Tell us your idea. One sentence is plenty.
Work