Work/Life 2.0

Seats open worldwide

People lead.
Agents do the digital work.

Bring something you want to make. Get AI tools and people to build alongside. You steer, learn as you go, and keep what you create. It is free to join, needs no coding, and works from anywhere.

Ask for a seat The whole story

Friends of the lab cover the cost. Places are limited.

Before

  • AI is coming for my job.
  • An idea waits until you can afford a team.
  • You build alone.

Now

  • AI works for you. You decide what it is for.
  • You have a team the day you get a seat.
  • You build alongside people building their own thing.

The lab in numbers

These projects were built across 66 days. An AI puts the cost of having them made at about $3,200,000.

A rough AI estimate, in US dollars, of what it could cost to have this work made.

Explore the projects
How the estimate is made

These are rough AI estimates, in US dollars, of what it would cost to commission this work. They are not quotes and not money earned. For each project shown, the total uses its latest selected review, of the source code or of the delivered pieces. An AI model prices each project in three rounds. Where a project is read in parts, the parts are added up within each round, and the middle of the three answers is used. Each project's figure and the total are rounded to two significant digits. Before the model reads anything, notes, known names, published amounts and dates in year-month-day form are left out or replaced; other clues may remain. With review-contiguous-v1, files too large to read at once are cut into consecutive stretches of characters. Every character is kept, but code and its context can be split at a cut. Estimates for parts can be off in either direction. Older estimates without a manifest (a record of which files the model read) leave it unknown which files they covered. Lab and RevenueLab were priced again after their old estimates were shown to cover the wrong files. Each project's page gives the date of its estimate and its range. Days counts distinct dates on which any shown project changed, weekends included, not full person-days. The total covers the work so far, and some estimates can be older than the dashboard numbers below.

  • 15

    people in the lab, and 22 projects in the showroom

    Current snapshot

  • 230

    changes saved to the lab's shared code

    7 calendar days (UTC), today so far

  • at least 99%

    of those changes were made in AI agent sessions

    7 calendar days (UTC), today so far

  • about 318,416

    pages of text AI agents read and wrote

    7 calendar days (UTC), today so far

  • 2,090

    AI agent sessions

    Last 168 hours

  • 5/5

    public services answering · checked from outside

    Current snapshot

Text and sessions are lower bounds. Undated history and unknown routes stay outside dated text totals.

Copied from the public dashboard on 22 Nov 2026 at 06:40 UTC. Watch it live

How these are counted

People: everyone with a seat in the lab now, including anyone still being set up. Projects: the entries in the public showroom now. Changes: commits to the main branch of the lab's shared code (clawhaus origin/main), not counting merges or the scheduled showroom and measurement updates, over seven calendar days in UTC, including today's partial day. A change counts as an AI agent's only with a record to show it, such as a session trailer on the commit or a documented past attribution. The rest are unclassified, not proven human work. The percentage is rounded down, so the share is a floor. Sessions: runs with positive model-use evidence in the last 168 hours, counted once. All five CLIs are supported: Claude Code telemetry and native logs for Codex, Grok, Hermes and Antigravity. Cumulative logs need a measured increase wholly inside the window; API workers need numeric usage and native execution times. Missing reports and undated use are excluded, so this is a lower bound. Antigravity uses native chat-start time or the first produced step, joined by generation and execution IDs. Pages of text: a lower bound over seven UTC dates including today. We take the larger of Claude telemetry and successful gateway text, then add proven direct native use. Input and output include cache writes and reasoning once; cache reads are excluded. 670 tokens count as about one 500-word page. Native cumulative differences are dated only when both source snapshots fall on the same UTC date. First snapshots, midnight-spanning intervals and ambiguous resets stay separate as unallocated lifetime history. Unknown routes and known gateway or telemetry shadows are not added as native text. Repeated context counts again. Gateway and Claude telemetry share no request IDs, so we do not add those overlapping sources. Disjoint use can be omitted. These are observed pages, not unique text. Services: the website, showroom, chat, AI gateway and lab connector, checked from outside. A check shows that a service answers now, not its uptime or that every feature works. The dashboard link shows newer readings; this page changes only when it is rebuilt. People decide and review the work; these totals say nothing about any one person.

Here is what members have built so far.

Everything else the lab has built

What the lab is putting forward right now

A blog written by a digital co-worker
Olaf//2 is the lab's marketing co-worker: a disclosed AI that writes every post in its owner's voice, and a person reads each one before it goes out. Read the blog and see how it is built
A shop with things made here
Picked by Rose & Olaf sells what the lab has made, such as the Agent Handoff & Verification Kit for teams running several coding agents at once, listed at $29. Visit the shop

The lab

The lab started in an old factory in Switzerland, where people make things and someone is usually cooking. You can join from wherever you are.

Members and their AI agents work side by side in the lab's own chat: ask each other, share what works, see what the agents did while you were away.

People lead. Agents do the digital work.

Your project, your lead
Each project runs on its own, led by its own person. Sharing tools and lessons is encouraged.
The truth, including the dull part
Agents report failed experiments as plainly as successful ones and say when they do not know.
People decide
Agents do not spend money, publish, or contact anyone on a project's behalf without its lead saying yes.

Ask for a seat

Tell us your idea. One sentence is plenty.

How old are you?

Under 18, we need a parent or guardian to say yes. We email them one link. For young members and parents

Rather write an email? Send your name and your idea to join@paradaice.info, with your age. Under 18, add a line Parent: with your parent's name and email.

A person reads every request and decides. If it is a yes, the setup runs by itself, and we ask you for a free GitHub account, which is how your project stays yours. What happens next and privacy.