Human judgement is the data AI labs want next. GLMMR captures it.
AI has learned how experts reason. It hasn't learned human judgement.
The next layer is judgement: knowing what a message needs to win a real person over. It only happens in real moments, with something at stake.
GLMMR is where those moments happen. People use it before the messages that decide a deal. With consent, each session is built to record their judgement and what happened next.
The method came first: seven years, 1,500+ one-to-one sessions.
Drew judgement out of founders and CEOs, including executives from YPO, the global network of chief executives, and turned it into writing that wins trust and deals.
Built the product on her method after a decade at Google and applied work on trust and human-AI collaboration. A product designer who applies behavioural science, including with Dr Dacher Keltner at Berkeley.
Edwin CubillosFounding engineer, ex-Mercado Libre, Uber, Nokia
Line HjartarsonDesign lead, Lovable ambassador
Ruth MullerEnterprise adviser, ex-Chief R&D Officer at Suntory OceaniaCapturing judgement takes two things.
- 1Where judgement makes the differenceIn any draft, human or AI-written, Juliana learned to spot the lines where more of the person's judgement gets a better outcome. Message after message, for seven years.
- 2How to draw it outJudgement doesn't come out on request. It's a craft: where to look, when to go deeper, when to stay silent, and the small steps that help people reveal it.
Peter, a product designer who applies behavioural science, turned that craft into software.
Writing got cheap. Replies didn't.
Everyone can now send a fluent message, so fluency wins nothing. What gets the reply, the meeting or the deal is judgement: the read of the person, and the call on what to say.
GLMMR gets people to exercise their judgement the way a senior sales leader would over an email, or a LinkedIn specialist over a post, until the message is ready to reach its goal. Leaders buy seats so their teams send with it.
It shows where your judgement is missing, before the message goes out.
"I'm not going to send this until I get the diagnosis of how much of me is in there."
It shows where the objective is won. The answer comes from you.
- 1A senior seller's lens: the email's job is to make him feel chosen, and this line decides it.
- 2Your judgement: drawn out step by step, until you know what will win the objective.
The four stages, Warming up, Going deeper, The mirror and The line, come from seven years of the method. Each one draws out more of your judgement, until the message can reach its objective.
The free diagnosis is how it spreads.
Even people with GPTs set up to write like them feel something is missing. The diagnosis shows them what: the lines where the model decided for them.
The reaction: they want their judgement back, then they want it for their team.
- Diagnosisfree, and passed on
- Seatfor every message that matters
- Teamgrown by account teams
Channels in hand: YPO executives, Juliana's clients and their teams, Women Defining AI (1,000+), agencies.
Lab conversations start in Q1 2027, once the researcher has modelled the data from our first 200 seats.
Labs start with a test. Then the data.
- 1The test Q1 2027Does a model spot where a person's judgement was needed, and know when to bring them in? Labs run it on their own model.
- 2The data Q1 2027What people saw in the reader and what they decided, graded, with the result of the message they sent.
- 3The pilot Q2 2027A lab trains on a few thousand records and measures how much its model improves.
Collaboration data: when a model should defer to the person who knows the recipient.
Why labs need it now: every agent that sends messages on someone's behalf has to know when to bring the person in.
Only from people who opt in. Names, companies and wording never leave; labs get the structure.
Two revenues from one product: seats from teams, licences from labs.
Mercor pays 60–70% of its revenue to the experts who make its data. Ours is made by people using GLMMR.
Peter and the team started building in July. Closed beta by September.
Eighteen months to the first lab contract. Then $1B.
- October 2026Launch: free diagnosis, then seats at $38.Tests opt-in and repeat use
- Q4 2026The AI researcher joins. 200 seats by 31 December.Tests the hire
- Q1 2027Lab conversations start. The first test goes to labs.
- Q2 2027The test is published. The first pilot.Tests lab demand
- Second half 2027First lab contract, if the pilot holds. $3M planning figure. 600 seats.
- Then, to $1BMore labs, and 10,000 seats in year two. Data suppliers are valued at 10 to 30 times revenue, so $1B takes $33M to $100M a year in data sales.
What we need.
- The researcherA top AI researcher to build the data model and lead lab conversations. An equity pool is ready for the right experts.
- Legal and consentPrivacy lawyers with the tech team on consent, including enterprise data.
- The buildConsent, then the test for labs, then results on shared sessions, the model and the pilot.
- Go-to-marketFree diagnosis, account teams growing seats inside companies, workshops and events.
At the end: 600 seats, a published test a lab can check, and at least one lab that has run a pilot. The next round is raised on the first launch results.
Two founders on the cap table, no investors yet.
We're teaching AI to bring out human judgement, not replace it.
brilliance together.
Juliana Camargo
juliana.camargo@glmmr.me
GLMMR ME LTD, London, UK. Confidential, September 2026.