Crazy Idea

When AI Connects Humans, How Will Trust Be Built?

Dalpat · October 2026

Last month I booked a skating course for my kid.

It took me about 48 hours to find a coach.

Not because Dubai doesn’t have coaches. It has plenty. And platforms like Skilldeer already help people discover activities. But I wanted something as seamless as Urban Company or Airbnb, built around coaches: see who’s nearby, compare them, book a class, track progress and message the coach, all in one place.

Four phone screens of a coaching app: finding coaches nearby, a coach profile for inline skating, a course progress tracker, and a family membership plan.

Scroll sideways to see more →

A prototype I made for something I was looking for.

Nothing quite like that existed. So I did what every parent does. I asked friends, Googled, scrolled Instagram, sent WhatsApp messages, waited for replies, and hoped I’d picked someone good.

That small frustration led me to a much bigger question.

The question underneath

I started researching extracurricular activities and spoke to a few founders in this space. What stood out was how much people want to learn outside work and school: swimming, skating, football, drawing, music, and more. And not just kids. Adults want to keep learning too.

Demand is growing. Globally, spending on tutoring, youth sports and after-school programmes is forecast to roughly double by the early 2030s.

Bar chart of the global extracurricular market: private tutoring from $200B to $400B, youth sports from $180B to $360B, after-school programmes from $120B to $240B, recent estimate versus 2030 forecast.

Sources: Global Industry Analysts, Markets N Research, Credence Research. Market-research estimates.

Yet finding the right person still means jumping between profiles, messages and recommendations.

And the further I went, the more one thought kept coming back.

In the age of personal AI

Personal AI is getting very good, very fast. Soon nobody will open ten apps to find a coach. They’ll just ask their AI:

“Find me a skating coach for my daughter, near home, weekend mornings.”

The AI will find someone in seconds. Then comes the real question:

How does the AI know it can trust the person it found?

And underneath that, a bigger one:

How will one human find, trust and exchange value with another human when AI sits in the middle?

A chat with a personal AI that notices a free weekend, suggests bouldering, ballet and mobility classes, and books an intro to bouldering.

Scroll sideways to see more →

Humans could interact with their personal AI through text or voice.

How we’ve always trusted strangers

Until now, trust between strangers has mostly come from two places.

Community. A friend recommends someone. A neighbour knows a good teacher. Referrals and word of mouth.

Agencies and marketplaces. You trust the person at your door because Urban Company trained them, checked them and stands behind them. The trust belongs to the platform, not the person.

Both work. But both were built for humans doing the searching, not AI agents searching for us.

An AI can’t ask your neighbour. And it can’t take every platform’s word as equal. It needs a way to check for itself: is this person real, are they qualified, and have they actually done what they say?

What I found while digging

Others are coming at this from different angles.

RentAHuman (@rentahumanx) is exploring how AI agents can hire humans for tasks. Interesting, but that’s AI hiring a person.

Wajo (@wajo_ai) is building AI agents that find, coordinate with and book services on behalf of humans. Boardy (@boardyai) is an AI “superconnector” that gets to know you through a conversation and introduces you to people in its network.

Then there’s World ID, which proves you’re a real, unique human. When AI can generate any face, profile or review, proving you’re real becomes essential.

But being real isn’t the same as being trustworthy.

A real person isn’t automatically a qualified coach. Or background-checked. Or safe to leave with your kid.

Trust is a stack, not a checkbox

When I sketched out what a trust protocol for human-to-human exchange would need, it came out in layers:

Four stacked blocks labelled Real (identity and KYC), Verified (certifications and checks), Proven (real session history) and Trusted (reviews from real students).
The four layers an AI needs before it can trust a person.

Each layer looks simple on its own. Then real life adds more. Many coaches work through an academy that takes a cut. Some are paid per class, some on a salary, some on a retainer. Gyms run their own bookings and payments. And every country has its own rules: in Dubai, a coach working with kids may need a police Good Conduct Certificate; in the UK, it’s a DBS check.

It isn’t one system, it’s lots of small ones that have to fit together. And when kids are involved, getting trust wrong has real consequences.

How an AI agent actually trusts a human

Here’s the key idea: the AI doesn’t need to trust a stranger. It needs to trust the evidence.

When you ask your AI to find a skating coach, it asks the protocol and gets back proof instead of a profile full of claims:

A personal AI asked to find a skating coach returns a coach card for Ben Carter, linked to five verified checks: real verified person, police clearance, valid certifications, backed by an academy, and 98 sessions confirmed by both sides.

Scroll sideways to see more →

Your AI checks the evidence before it connects you.

Then there’s the money. Your AI pays, but nobody in the middle sits on it. A licensed payment company holds it and releases it only after both sides confirm the class happened. The academy takes its share, the coach is paid on their own terms, and if the coach doesn’t show, you get your money back.

The AI checks, then connects. The humans just show up.

One protocol, many applications

Architecture diagram: apps on top (personal AI agents, marketplaces, businesses and agencies, independent pros) call a Trust API; underneath, a human trust protocol covers identity, checks and credentials, affiliation, service proof, reviews, permissions and live status, with a trust ledger and separate payments; local issuers and rules for the UAE, UK and the next country sit at the bottom.

Scroll sideways to see more →

One trust layer, many apps on top, local rules underneath.

The way I see it, the trust protocol sits underneath and the applications run on top.

Take a coach marketplace. Today, if you build one, you do everything yourself: check every coach’s identity, certificates and background, collect reviews and convince parents it’s safe. It takes months, and none of that trust moves with the coach when they teach somewhere else.

On a shared trust protocol, the coach verifies once. Their identity, qualifications, real session history and genuine reviews are already there. The marketplace just plugs in and focuses on what it does best: great search, easy booking and a good experience for parents.

The same coach can be found by a gym’s booking system or someone’s personal AI, in any country, and the trust comes along every time.

One layer of trust underneath. Many apps on top, wherever one human needs to trust another: coaches, tutors, nannies, carers, and more.

Human-to-human work isn’t going anywhere. AI won’t replace that. What’s changing is how we find each other. More and more, that connection will run through an AI, and it needs trust we can check, not trust we have to guess.

I’ve started sketching what that could look like, but I don’t know the final shape yet.

If you’re building personal AI agents, I’d love to hear how you see it.

Talk to me on X @dalpattapaniya ↗