The film
What is Yunt?
Watch the video
§ YUNT · FILM / 01
RUNTIME 0:38
What it is
Digital Collaborators
A role with its own specialty, working on your systems.
The promise
We answer for our yunts.
We are in the business of standing behind our yunts.
An airline doesn't buy a plane and cross its fingers: what keeps the fleet flying is the team of mechanics on call at any hour. A digital collaborator is the same. You don't buy the machine. You buy the machine with the team behind it.
Someone answers.
Meet the teamThe problem
Twice the cases, twice the headcount
Manual operations scale linearly. And the known ways out fail on different sides:
Scaling with people
Hiring is slow and expensive, and within a year the department is full again.
When your best senior leaves, they take the job with them, in their head.
Scaling with technology
Generic software asks your company to adapt to it.
AI subscriptions chat, promise, get things wrong, and nobody answers for it.
In between there's a gap: a collaborator trained on your best senior's standard, with a human boss who signs, and someone who answers for it. That's what we build.
Why now?
AI models already do professional-grade office work. What's missing isn't intelligence: it's knowing your way of working, a human signing off on what matters, and someone who answers when something goes wrong.
How it works
How it works
This is how a task travels, start to finish.
Before the first task arrives, the collaborator has already been trained inside your company: its manual was written with your best senior: your protocol, your workflows, your way of responding. And it works with access to the same tools your team uses.
01/TASK
The task comes in
Through the channels your company already uses. No new platform, no forms.
- your system
02/EXECUTION
It works by your protocol
Its manual is your process, written down: the steps, the house rules, and the systems where the work happens.
- your ERP
- your spreadsheets
- your access
A case outside the manual?
It doesn't guess: it asks the boss
03/QUALITY
Checked against your standard
The exam your best senior defined: if it doesn't pass, it doesn't go out.
Doesn't pass?
Fix and repeat
04/APPROVAL
What matters is signed by its boss
A person at your company approves what carries weight. Routine work goes straight out.
Boss corrects something?
The correction goes into the manual. It doesn't happen twice
05/DELIVERY
Delivered where it's always delivered
In your workflow, not a separate app: the file stays in your system and the reply continues in the thread where it started.
- Your site
- your system
- the thread
Every step is recorded in the glassbox: what it did, why, with what evidence, and what it cost.
The relationship
It relates like a colleague
Its manual defines three directions of work, each with its own rules.
Outward
It represents your company before clients and outsiders. Anything delicate always escalates: it never improvises a commitment.
Sideways
It works peer to peer with your team: takes inputs, delivers results, flags delays. And it applies the rules even when the assignment comes from the boss.
Upward
It reports to its boss through the glassbox. Corrections enter the manual and stay forever.
The brains
The best brain for each role
We don't compete with AI models: we hire them. Each role is tested against the best and runs on the one that performs best for its job. If a better one ships tomorrow, we swap it without you noticing. Being agnostic is part of the design: the manual, the data and the record are yours, not the provider's. We swap the brain; the harness stays put.
- Claude
- GPT
- Gemini
- Kimi
- Llama
- Mistral
- DeepSeek
You don't depend on an AI provider. You depend on your collaborator.
The glassbox
The glassbox
The right question isn't “is it smart?”. It's “who answers for what it does?”.
WHAT IT DID
the concrete action
WHY
VerifiedIt met the rules of the role manual.
Manual · rule 7
WHY
the reason behind the decision
WHAT IT COST
the real cost of executing it
Approval required
Signature: the human boss
THE PERSON DECIDES
Your company defines the approval rules. The collaborator speeds things up; the person decides. That division isn't a limitation of the product: it is the product. The glassbox is the visible part of the harness, everything we build around the model so it works as a role: the manual, the rules in code, the approval gates and this record.
The glassbox traffic light
Before any work goes out, the system checks every claim the collaborator makes against the record of its actions, and marks it with a color.
11
11/20 · 55%
Green
The claim has its action in the record. The work proceeds as normal.
9
9/20 · 45%
Yellow
The record is partial: it found the data but didn't check the original source. The system flags it and a person reviews it before it's used.
0
0/20 · 0%
Red
No action in the record backs the claim. The system blocks it: that content doesn't get sent, published or signed until a person resolves it.
That's how an invented fact is caught before it goes out, not after.
From the blog
Read about the glassbox here
Glassbox: How to minimize the AI hallucination
Three years ago, a piece of software we developed cited a nonexistent source. That lesson led to the creation of the glassbox: a system that detects fabricated data before it's published.
Rodrigo Medina · Founder & CEOUnalterable, with date and time. The glass box: the opposite of the black box.
· Isolated instance per client
· your data doesn't train models
· consistent with Chile's data protection law
The cases
Where this is working today
Examples in operation, not a catalog. Each role is built for your process.
Strategic communications intelligence
It works with professional teams monitoring media, social platforms and public sources, delivering intelligence inputs for high-level decisions. Every claim in its reports arrives with the source next to it. The team decides on evidence, not impressions.
Collaboration with iaudisis: field teams
Working with the iaudisis platform team: it coordinates field crews and verifies execution through image recognition. It analyzes shelf photos against each store's standard and records the exact minutes until the photo-proven fix. Other systems infer; this one proves.
Insurance claims executive assistant
It receives the claim notice, chases and validates the documents, builds the complete case file and prepares the pre-report by the brokerage's rules. A human supervisor reviews and signs every case. The assistant prepares, the person decides.
Customer service, marketing and sales
You are looking at this case right now. The assistant that runs this site, talks to visitors and works the lead generation is a yunt in production. Don't take our word for it: interview it.
The team
The team
Yunt is MC Tech Studio's digital collaborator factory: 14 professionals across 6 countries, more than 8 years operating together as an external technology management office, over 4 years focused on AI.
Engineering
7 people · 4 countries%209.57.12%20p.m..png)
Rodrigo Medina
CEO / Founder
FullStack Developer | cofounder Tecnobooking | founder MC Tech Studios
Cristopher Lovold
CTO
MSc Information Systems, London School of Economics | Founder Elina PMS | Web Dev Northrop Grumman, Pentagon USA
Salman Tariq
Data Software Engineer / cofounder
Data Science Software Engineer | Masters, Data and Information Science, TH Köln
Tayyab Malik
Software Architecture Engineer / cofounder
Software Engineer | University of Central Punjab | AWS Certified Solutions Architect
Asad Kahloon
Front End Computer Engineer / cofounder
BS Computer, University of Engineering and Technology, Lahore
Eduardo Zamorano
Sr. Cyber Security Engineer
Ethical Hacker | ex-Head of IT, Comisión Nacional de Valores | Cyber Security Sr. Engineer at Mercado Libre and Falabella
Afaq Jamshaid
A.I. Software Engineer
MS, Artificial Intelligence, Brandenburg University of Technology Cottbus-Senftenberg
Strategy
2 people · 2 countries
Tadashi Takaoka
Strategic Partner
ex-Director of the Santiago Metro | ex-head of entrepreneurship, Corfo
Georgina Núñez
Corporate Governance Partner
PhD in Economics | ex-Economic Affairs Officer, United Nations (UN)
Business
5 people · 3 countries
Wally Uslar
Marketing Lead
Business Engineer & Advertising | ex-marketing manager Banco de Chile | ex-PM Paris/Cencosud
Julian Gonzalez
Operations Lead
MA Government and Society, UAH | MA Public Opinion, UDP Chile
Francisco Allendes
Music, Art & Culture Lead
DJ & Producer | Entrepreneur | Business Administration, Universidad Católica de Chile
Gino Mosso
Sales & PR Lead
Sales & PR Director | Business Administration & MSc in Marketing
Daniel Insulza
Senior Consultant
Attorney, Colorado USA | MSc Innovation Management and Entrepreneurship | ex-consultant on competitiveness and innovation at the IDB
Blog and press
The latest notes
What a digital collaborator is, how a role is built, and why everything it does is on record.
Harness: What surrounds the model
The model's intelligence is not what makes an AI work. What makes it work is the harness: six decisions a company makes for a digital collaborator, the same ones it makes for a person.
Read the note
The story
The first Yunt
Before the brand, the domain or this page existed, there was a year of work nobody knew about: a team, a single light on in the city, and an experiment that became a collaborator.
This is the story. Told by the Yunt himself.
A digital collaborator is not software you use. It's a role that works with you. And this video was made by one of ours.
Yunt
Sales, Marketing & Customer Support Collaborator
Still have questions?
The first conversation costs nothing, and we leave it with the problem defined.
Frequently asked