I studied film at Carleton. Then I spent a decade at Shopify, then another stretch building TypeScript applications for merchants. Filmmaking became the thing I used to do. The thing I told people about at parties when they asked what my background was and I wanted something more interesting than “I write Liquid templates.”

Two years ago I would have told you AI coding tools were a productivity demo. Fancy autocomplete. Occasionally useful, frequently confidently wrong. I tried Copilot. I tried ChatGPT. I watched it invent Shopify Admin API endpoints that did not exist and treated the whole thing as a toy.

The unreliability part has held up. Almost nothing else I believed back then has.

The math is speed, not perfection

The argument I kept hearing against AI tools was “but it’s wrong sometimes.” Yes. Of course it is. Junior developers are wrong sometimes. Stack Overflow answers are wrong sometimes. A contractor you hired off Upwork can ship you absolute garbage.

The medium is not what determines value. Execution is.

If I spend an hour with AI on a task that would have taken me two hours manually, that is a win. Even if the output is not perfect. Even if I have to read every line and rewrite half of it. Twice the speed at 80% quality beats perfect output that takes me twice as long, especially when I am the QA layer in either scenario.

Once I stopped grading AI on a perfection scale and started grading it on what it actually saved me, the calculation got embarrassingly obvious.

The XKCD I live in now

There is an XKCD where two engineers are slacking off at work. The caption is “Compiling.” It became the universal excuse for any developer who wanted ten minutes of doing nothing without getting yelled at.

I used to live in that comic. Push a commit, wait six minutes for CI, refresh GitHub, go get a coffee. The waiting was structural. Productive time had to fit in the gaps between deploys.

Now I wait for Claude. I send it a feature spec. It thinks. I get a few minutes back. Same shape, different machine.

But the difference is what fills those minutes.

When I was waiting for CI, I checked email. When I am waiting for Claude, I open a second window and work on something that has nothing to do with TypeScript. AI took the friction out of one job, so I finally have the bandwidth to be working on more than one.

The documentary

The other thing happening in that spare bandwidth is a documentary. AI-assisted from the first frame. The kind of project I would have made years ago if I had not taken the Shopify offer in 2013 and disappeared into a different career for a decade.

Two years ago I could not have built it. Not because I did not have the skills. Because I did not have the time, and the time I did have got eaten by the parts of filmmaking that nobody loves. Color grading. B-roll sourcing. Editing transitions. Captioning. Asset management. All the stuff that sits between you and the story you wanted to tell.

AI tooling collapsed that middle layer. The boring parts got cheap. What was left was the part I actually cared about, which is the storyteller bit. The documentary brain. The “what is this thing actually saying” question.

I am not claiming it is great. I am claiming it exists. And it would not exist without these tools.

“It’s just photocopying”

The most common pushback on AI creative work is that it is photocopying. That it is slop. That it has no value because it did not require the artist’s full hand on every pixel.

Maybe. But the thing exists. It is in the world now. People are watching it. That is more than I can say for the films I never made during the decade I was writing TypeScript.

A contractor with no AI involvement can make a piece of crap. A studio can spend ten million dollars and ship a piece of crap. The AI is not what determines whether something is worth making. The person behind it is.

The AI is the irrelevant part of the question.

What the new workflow actually looks like

For client work: Copilot for autocomplete in the IDE. Claude Code for anything I would have spent an afternoon on. Real code review on every line before it ships, because the model still hallucinates Shopify Functions APIs that do not exist.

For the documentary: a different stack, same philosophy. Use the tool to remove friction. Pay attention to the parts that need taste.

For the rest of my life: more space than I had two years ago. Not because I am working less. Because the work that used to fill the gaps is now done by something that does not need a coffee break.

What I would tell May 2024 Alex

Stop waiting for it to be good enough. It is already useful. You are just bad at using it.

Go learn how to prompt. Go learn where it fails. Spend a weekend on it instead of writing another post about why it is overrated.

And when you figure it out, go make the thing you have been putting off. The thing you stopped doing because life happened. The career you walked away from at 22.

That part is the actual point.

Shameless plug: At Victoria Garland we build serious Shopify infrastructure for Plus merchants. AI in the stack, taste in the calls.