Your First Unit of Trust Is Borrowed

TL;DR: AI will compress the premium on raw capability. What rises is reputation, relationships, and guarantee eligibility — whether someone will put their name behind yours. Your first unit of trust is rarely earned from zero. It is borrowed. Early career is the conversion process: take small, bounded, survivable commitments; turn borrowed credit into a personal record; graduate from "who introduced you" to "your name is enough."

I am James, CEO of Mercury Technology Solutions. Hong Kong — August 2026

Everyone is asking the wrong early-career question.

They ask: How do I look smarter than the model?

The better question is older than AI:

How do you get licensed to carry more responsibility without staking a risk you cannot survive if it blows up?

That is not a ChatGPT problem. That is a civilization problem. Societies already solved pieces of it. We just forgot the mechanism because exams made trust feel automatic for a few decades.

The reprice: capability down, guarantee up

As models get cheaper, the market premium on pure "I can do the task" declines.

What gets more expensive:

Reputation — a durable signal that you finish, repair, and do not disappear
Relationships — people who have watched you under load
Guarantee eligibility — whether a serious person will stake their credit on your delivery

In other words: the scarce asset is not intelligence. It is underwritten trust.

Young people get crushed by this math first. No assets to collateralize. No multi-year reputation stack. No board-level network. So they try to brute-force proof through capability alone.

Necessary. Not sufficient.

First trust is borrowed, not earned

This is the part polite career advice skips.

Your first real shot often comes from:

who your parents are
who your mentor is
which relative or acquaintance opens a door
which school brand still functions as a weak guarantee

Recruiters staring at "first degree" are doing the same thing with a spreadsheet. The degree is a temporary guarantee instrument. It is also depreciating.

We got addicted to gaokao logic: one objective score, pass the line, identity irrelevant, call it fairness.

That may be a historical intermission, not the default state of human coordination.

Before exam empires, access ran on relationships. After AI absorbs more cognitive work, access will run on relationships again — just not only blood relationships.

More often: chosen ties. People you repeatedly shipped with. People who saw you miss and recover.

Why relationships actually price

A relationship is not "networking drinks."

It is two mechanisms stacked:

1. Information mechanism Someone who worked with you for a year knows more about your reliability than a 45-minute interview ever will.

2. Responsibility mechanism When they introduce you, they risk their own credit. A bad intro is not free. It taxes their name.

So the valuable moment is not "he knows me."

The valuable moment is:

"I know this person. Give them the job. Put my name next to it."

That sentence is a guarantee. Guarantees are how trust scales beyond one-to-one observation.

The Trust Ladder

Ideal path is boring and brutal:

Enter a real collaboration
Take a small commitment with hard edges
Deliver
Get trusted with a slightly larger surface area
If you miss: disclose fast, repair visibly, keep the relationship solvent

At first, a mentor / school / friend lends you credit. Then you convert the loan into a personal track record. Then people stop asking who introduced you. Your name clears.

That is the whole game:

Borrow → perform → convert → compound.

The cold-start problem

Fine. What if your school is average, your network is thin, and nobody important will stake anything on you?

This is where capability still matters — but as a bootstrap, not as the end state.

Intelligence is a permissionless capability. You can study alone. Practice alone. Sit an exam. Earn a certificate. The exam translates private effort into public credit without needing a patron's blessing.

Will AI kill exams?

Maybe as pure skill tests. Maybe not as character and cognitive endurance tests.

If the exam still filters for: can you self-direct, stick through complexity, and show up reliably under evaluation — institutions may keep using it even when models can spit the answers. The point stops being "do you know the content" and becomes "are you the kind of system that finishes hard, boring loops."

But if exams fully devalue, you hit the wall:

Intelligence can be self-taught. Credit cannot.

Trust needs at least two people. It is usually a byproduct of finishing something together.

So permissionless learning has to upgrade.

From permissionless learning to permissionless contribution

Stop only collecting private reps.

Ship public, inspectable work:

internships with named owners and clear deliverables
community work where someone depends on you showing up
open-source contributions with review history
small paid gigs with written scope, deadline, and postmortem

The unit is not "I studied." The unit is "I made a bounded promise in public and closed it."

That is how you mint the first coin of non-borrowed trust when nobody will lend you any.

Small bets, not hero bets

Early career death pattern: swing at a risk that would destroy you if it fails, because you want a fast reputation spike.

Wrong instrument.

You want small, survivable commitments:

| Bad early bet | Better early bet | |---|---| | "Give me the whole product" | "I own this module by Friday" | | Unscoped favor for a big name | Written scope, deadline, acceptance criteria | | Silent struggle until explosion | Early status, early escalation, clean repair | | One theatrical moonshot | Ten boring closed loops |

Each closed loop is a trust tick. Each clean repair after a miss is also a trust tick — sometimes stronger than a perfect run, because reliability under failure is the actual scarce signal.

Stop trying to be chosen as a genius. Start trying to become cheap to underwrite.

This is not only an individual problem

If an industry needs new blood, it must provide permissionless entry ramps.

If a society wants mobility, it needs public infrastructure for first credit — not only elite introductions and decaying diploma stamps.

Exams were one such infrastructure. Portfolios, open contribution graphs, apprenticeship ledgers, bonded referrals, and better work-history rails may become the next.

If exams die and nothing replaces them, you do not get a pure merit utopia. You get a return to closed rooms with prettier vocabulary.

The interesting future is not "back to pure nepotism." It is noisier, more visible, more contribution-based reputation markets — if we bother to build the rails.

What to do this quarter

If you are early-career:

Inventory who can lend you first credit. Mentor, alumni, tight collaborator, former lead. Ask for a scoped intro, not a vague blessing.
Make one promise small enough to survive failure. Write the edge: done means X by date Y.
Publish the receipt. PR, memo, demo, client note — something a third party can inspect.
Practice repair as a skill. Misses that are owned fast compound better than silent perfection myths.
Upgrade learning into contribution. If nobody is watching your private study, put work where review can happen without permission.

If you lead teams:

Stop demanding "2 years experience" as a lazy filter and then complaining nobody junior is good. Build bounded quests that mint first trust on purpose.

The equation

Capability gets you into the room when the door is open. Borrowed trust opens the first doors. Converted trust makes your name a key.

Your first unit of trust is borrowed. Your second must be earned in public. Your third is when other people start borrowing *your* name.

That ladder does not care how eloquent your model is.

It cares whether a human will still stake their reputation on the sentence:

"This person. This job. My name next to it."

Mercury Technology Solutions: Accelerate Digitality.