Every home built right. Every job, ground truth.
Labs is our moonshot. Every job we do teaches us how the physical world really works. We're turning that into proof on our own jobs first, and then into robots that help small businesses: programmed, assembled, installed and maintained by people who actually know the work. It starts small and honest: every shower we build gets a 24-hour flood test, and starting this fall we record every result.
This page is our plan, in the open. We update it as we go. Last updated September 27, 2026.
Why a renovation company has a lab
This is our moonshot. We build bathrooms, and on every job we learn how homes and the physical world actually work. Labs puts that knowledge to work: first as proof on our own jobs, then by programming, assembling, installing and maintaining robots that help small businesses, starting with our own. Robots are going to need a general contractor. We intend to be one.
The problem
A shower pan either holds water or it doesn't. Once tile goes on, nobody can see which. Most waterproofing is judged by eye, then buried behind the finish.
The label
We already flood-test every shower: plug the drain, fill the pan, wait 24 hours, measure. It is hard to get right and easy to check, which makes it a perfect pass-or-fail record.
Our advantage
Real jobs and real trade judgment, the two things research labs don't have. 550+ completed jobs in the last two years and 100+ five-star reviews from Seattle homeowners.
The Singularity has plenty of authors now, Alex Wissner-Gross wrote, but
“what it has always needed is a general contractor.”
Dr. Alex Wissner-Gross, co-author of Solve Everything, in “Solve Everything,” The Innermost Loop, February 14, 2026. Read the essay
We're a general contractor. This is us taking him at his word.
One trunk, five branches
Everything in Labs grows from a single habit: an honest record of every job.
Proof
- A flood test on record for every shower
- The homeowner sees the result at closeout
- Both sides agree the job is done before the final invoice
Data
- Same fields, same photos, same file names on every job
- Consent first, always optional
- No client names or addresses, ever
Tools
- The crew talks instead of typing: voice-note job reports
- AI drafts, a person approves
- One weekly number for every AI workflow
Robots
- Build and program our first robot arm, taught by demonstration
- Design, 3D-print and machine our own parts in-house: Bambu P1S today, CNC next
- Assemble, install and maintain robots for small businesses
- A robot and a tradesperson on the same job
Community
- 20-minute working sessions with robotics people
- Our own flood-test challenge, when the data is ready
- Seattle robotics and AI meetups
What we're tracking
Honest numbers, including the zeros. Updated September 27, 2026.
- DoneSep 16DuetRight Labs started
- DoneSep 24-25Learned from Moonshots LIVE: XPRIZE, Google X, the Organizational Singularity
- DoneSep 26Capture spec, version 0.1
- DoneSep 27Plan, tests and kill criteria set, and this page
- NextOctoberFirst recorded flood test on a real shower
- NextOctoberBench test pans, including planted defects
- NextOctoberFirst robot arm assembled, programmed and taught one real task
- NextNovemberSit down with five small businesses about the robot they'd hand a task to
- NextNovemberVoice-note job reports: one crew lead, one week
- NextDecember 1Keep-or-kill review, decided on the numbers
The 90-day plan
September 28 to December 27, 2026. Small on purpose: our first attempts at big AI programs stalled, so this one is built around a few habits real people own.
Clear the decks
- Follow up with the people we met at Moonshots LIVE
- Pick the first pilot shower
- Lock the capture fields
Train the monkey
- Every shower-pan build gets its flood test on record: two gauge photos and a pass-or-fail line
- Test pans on the bench, some built with planted defects
- A consent option drafted and reviewed
- Assemble and program our first robot arm, and teach it one real task by demonstration
Measure and remove typing
- Voice-note job reports in any language; a person approves each one
- A weekly Labs number
- A blind review of 10 past jobs: do our records hold enough to judge the work?
- Sit down with five small businesses: which task would they hand a robot, and what would they pay for us to install and maintain it?
Decide with numbers
- December 1: keep going, or drop back to a simple quality habit
- If we keep going: a first internal dataset, our challenge rules, and a first robot pilot with a small business
Guardrails: under $500 in cash this quarter, no new software subscriptions, and experiments never slow a paying job.
How a flood test becomes a label
Membrane cured, drain plugged, water filled to a marked depth, a ruler taped in as a gauge, and a cup of water in the same room to measure evaporation. We take photos at the start and again after at least 24 hours, plus moisture readings outside the shower.
At least 24 hours, both gauge photos, the water dropped no more than the evaporation cup (plus 1/16 inch), and everything outside the shower stayed dry.
The water dropped too far, or moisture showed up outside the shower. We record where, fix it, and test again. Failures are the most valuable records we keep.
Under 24 hours, a missing photo, a leaking plug, or disturbed water. That isn't a result, so we retest.
Real showers should pass, so we also build small test pans with deliberate defects. That's how we learn what failure looks like before it happens in someone's home.
We ran the Google X test on ourselves
At Moonshots LIVE, Astro Teller, who leads X, Google's moonshot factory, explained how X decides what to pursue and how quickly it kills what isn't working. Here's how Labs holds up.
A huge problem
Home repair is slow, labor-bound and hard to verify. Water failures hide behind tile until they rot a floor.
A radical solution
A robot crew for every small business, installed, programmed and maintained by people who know physical work. On our own jobs, a robot and a tradesperson side by side, and a family gets its bathroom back in days, not months.
A technology with a prayer
Robot models that learn a task from a demonstration, and open-source robot arms that cost a few hundred dollars, are arriving fast. Our edge is the physical know-how to make them work in real places, and keep them working.
Monkey first
Teller's rule: do the hardest part first. We have two. Can a small contractor record every job consistently, at almost no cost to the crew? And can we get one robot to do one useful task reliably, first for us, then for one real small business that would pay for it?
We stop, and keep only the quality habit, if any of these is true on December 1
- A shower-pan build since October 1 went unrecorded
- Recording takes more than 15 minutes of crew time
- After five asks, no robotics or AI practitioner will look at a sample record
- Labs work delays a client's quote or project
- Our first robot can't do its one task reliably in our own shop, and we can't see how to fix it
We double down if all of these are true
- Every shower-pan build is recorded
- At least two outside practitioners reviewed a sample and told us what they'd want
- At least one homeowner saw the proof and it showed up in a review, a referral or a story
- At least one small business asks us for a paid robot pilot
Verdict: it's a moonshot, so we run it like one. We're audacious about where it's going, honest about where it is today, and quick to learn from every test. Astro Teller calls that audacity and humility in equal measure.
Our XPRIZE-style ladder
No XPRIZE is open to a company our size today, and none targets home repair. So we climb in steps, each one earned by the step before it.
Ask the question
Could a recorded 24-hour flood test be the finish line for robots doing real work in unmodified homes, and how would you stop anyone from staging a win?
Learn how new prizes launch
Several new prizes are being announced this fall. We're watching how they're designed and funded.
Submit a Verified Repair idea
When XPRIZE reopens its public idea window: robots, or robot-assisted crews, complete real repairs in occupied homes, checked by objective tests such as the flood test.
Run the DuetRight 24-Hour Challenge
A small community prize: predict whether a shower will pass its flood test from photos taken before it. We'll run it once we have enough labeled tests, realistically late 2027.
Help set the standard
Publish our wet-area job record format, and offer our jobs as a real-world test site. Prizes often create the standard a whole market adopts, and that's the part we care about most.
What we brought home
Stan and Angelo spent two days at Moonshots LIVE in Los Angeles. These are the ideas we're putting to work, in our words, with what each one changes here.
The data set matters more than the model.
We lock our record fields before the first test, so every job is comparable.
Pick targets that are hard to solve and easy to check.
The flood test is our label and our proof to the homeowner.
Train the monkey before you build the pedestal.
Records come before dashboards, datasets or robots.
Kill ideas early on first-principles economics.
Our stop signals are written down before we start.
Keep the human accountable above the loop.
AI drafts, and a person approves anything a client sees.
Don't lose your people. They carry the knowledge that isn't written down.
Labs captures our crew's craft. It doesn't replace it.
Don't build what others already build well.
We don't reinvent robot hardware. We program it, assemble it, install it and keep it running.
Say your one sentence out loud.
Every home built right. Every job, ground truth.
The rules we hold ourselves to
- Consent first. Labs filming is optional, and saying no never touches a homeowner's project.
- No client names or addresses in any Labs record.
- Our claims match the record. If we haven't measured it, we don't say it.
- Experiments never slow a paying job.
- AI drafts, and a person approves anything a client sees.
- We capture our crew's craft. We don't replace it.
Work in robotics, AI or building science?
We'd like to compare notes. Ask us for a 20-minute working session, and tell us what you'd want to see in a real job record.
Frameworks referenced on this page: XPRIZE's prize-design principles and Astro Teller's moonshot test as presented at Moonshots LIVE 2026; The Organizational Singularity by Salim Ismail and contributors; Solve Everything by Alex Wissner-Gross and Peter Diamandis. DuetRight isn't affiliated with XPRIZE, X or Moonshots.