What Used to
Take a Team.

If a skater bought one Personal Video Feedback, ONE should already know.

Their account should simply show that they have one submission available.

If they bought two, they should have two.

If an unused purchase is refunded or cancelled, the entitlement should disappear.

When they submit, ONE should use exactly one entitlement.

No receipts.

No order numbers.

No proving that they paid.

The system should already know.

That one decision changed the scale of the project.

Shopify is excellent at knowing what someone bought.

It was not designed to manage the entire workflow I was beginning to imagine.

So ONE needed somewhere to remember its own information.

We built a backend.

Then a database.

Now ONE can maintain records for things such as purchased entitlements, submissions, submission status, completed feedback and customer history.

But if ONE was going to hold customer-specific information, another problem appeared.

The customer's browser does not tell ONE who they are and expect us to trust it.

Shopify signs their identity.

ONE verifies it.

Only then does the backend return information belonging to that customer.

That distinction matters.

A convenient system that exposes someone else's information is not a good system.

Knowing who the customer is solved only half the problem.

ONE also needed to know what happened in Shopify.

When somebody pays for Personal Video Feedback, ONE needs to create an entitlement.

When something is refunded, ONE needs to know.

When an order is cancelled, ONE needs to know.

So we connected Shopify's commerce events to ONE.

We tested this against the actual production store.

A controlled $10 order was created manually.

No card was charged.

Shopify sent the real paid-order event.

ONE received it.

ONE verified it.

ONE created exactly one Personal Video Feedback entitlement for the correct customer.

The order was then cancelled and refunded.

Shopify sent the cancellation and refund events.

ONE received them.

The entitlement became unusable.

That was the first time the entire commerce side existed outside a test environment.

Once the commerce engine worked, we returned to the customer.

They needed a form.

Two fields.

YouTube video link.

What would you like feedback on?

Then:

That was all the customer needed to see.

But underneath that button we needed to guarantee something important.

Pressing Submit twice should not consume two purchases.

A slow connection should not accidentally create two submissions.

If the response gets lost after ONE successfully receives the submission, retrying should not use another entitlement.

Initially we tried generating the secure submission identifier inside Shopify's customer-account application.

It failed.

The extension runtime did not expose the cryptographic capability we expected.

So we changed the architecture.

ONE's backend now generates a secure, short-lived submission attempt.

The customer submits using that attempt.

If something goes wrong and the request needs to be retried, ONE can recognize it as the same attempt.

One submission.

One entitlement.

No accidental duplicate.

The customer sees none of this.

They press a button.

That is how it should be.

Once a skater submits their video, somebody has to respond.

So we built a private ONE Academy admin system.

It is protected separately from the customer account.

A new submission appears as pending.

I can see the skating video and what the skater wants help with.

I can provide written feedback.

The system can also support a response-video link when demonstrating something makes more sense.

Then I mark the feedback ready.

The customer's ONE Academy updates.

Under Your Academy they now see:

Their submission has become part of their history.

That matters to me.

A lot of coaching disappears.

You have a lesson.

Someone gives you useful information.

The call ends.

A week later, some of it is forgotten.

I want ONE Academy to become somewhere a skater's work can accumulate.

Not meaningless gamification.

Actual learning.

Actual coaching.

Actual feedback.

Actual assessments.

Actual progress.

Eventually there was only one test that mattered.

Could a real customer actually use it?

We temporarily activated the new ONE Account configuration.

I signed into a real customer account.

ONE recognized me securely.

A test entitlement was added.

ONE Academy immediately showed:

I opened Personal Video Feedback.

Entered the video.

Explained what I wanted feedback on.

Pressed Submit.

ONE's backend created a secure submission attempt.

The entitlement was consumed.

The submission was saved.

The screen changed to:

I returned to ONE Academy.

Without refreshing the page, the submission was already inside Your Academy.

It showed:

Then the admin side.

The submission was waiting.

Feedback was added.

It was marked ready.

Back in the customer account:

The feedback was there.

The entire loop worked.

Then we deleted the test data.

It would be easy to tell this as:

I had an idea.

AI wrote some code.

It worked.

That is not what happened.

We repeatedly hit real technical constraints.

Over and over.

The interesting part was not that AI magically knew the answer.

It did not.

The difference was how quickly we could move through that loop without losing the context of everything we had already learned.

It is easy to get distracted by the technology.

The technology is not the point.

The skater got something better.

They can increasingly come to one place.

From the customer's side, it looks like a clean account page.

Underneath it now sits:

By the end, the backend alone had more than 80 automated tests, with additional tests around the customer application.

All of that so the customer can press:

And have it work.

This is where the day became difficult for me to ignore.

Imagine commissioning the system after it already exists.

You would potentially need expertise across Shopify development, frontend development, backend development, database architecture, authentication, infrastructure, security, user experience, testing and ecommerce operations.

Then someone has to coordinate all of it.

A strong independent developer could potentially cover much of that.

An agency could assemble the different skills.

But this is not a small Shopify theme customization.

A conventional build would reasonably be measured in weeks or months rather than hours.

Depending on who built it and how much discovery was required, an agency replacement cost could reasonably run into tens of thousands of pounds.

Perhaps £25,000.

Perhaps £60,000.

Potentially more with a larger specialist agency.

Those are not valuations of what ONE built.

They are estimates of what it could reasonably cost to commission comparable work conventionally.

And there is another problem with even making that comparison.

I did not know what it was yet.

This morning I did not have a specification saying:

Build an authenticated Shopify Customer Account application connected to a Cloudflare backend with an entitlement ledger, webhook retry lifecycle, encrypted OAuth credentials, protected administration and an asynchronous coaching workflow.

I was trying to make the customer experience better.

One question led to another.

Each answer exposed the next useful question.

Historically, the cost of following those questions would have become the limiting factor very quickly.

At some point I would have needed to stop.

Write a specification.

Find developers.

Get quotes.

Find capital.

Decide whether the idea justified spending tens of thousands before knowing where it would ultimately lead.

AI changed that equation.

There is another important part of the process.

I was not asking a chatbot to make an app.

Different systems increasingly played different roles.

At different points, one system could inspect or challenge what another had proposed.

And the real store, real browser, real backend and real customer remained the final source of evidence.

The interesting capability was not one enormous prompt.

It was the loop.

Then repeat.

That is very different from asking AI to write something for you.

This may be the part with the biggest long-term consequence.

We did not build all of that infrastructure solely for a $10 Video Feedback product.

We now have reusable building blocks.

That means the next service does not begin where this one began.

Assessments can reuse much of the same infrastructure.

A customer purchases an assessment.

ONE knows they are entitled to it.

They submit what is required.

A human assesses it.

The result becomes part of Your Academy.

Coaching can increasingly connect into the same customer history.

Certificates can eventually represent actual completed requirements.

Other ONE services can use the same identity and entitlement systems.

That is where the economics become even more interesting.