Allow

There is a button in Claude. It says Allow.¹

Click it, and Claude reaches into Supabase and runs a SQL query to build my first database table. Columns, types, a structure for the ideas I wanted my website to hold. Nothing exotic. An engineer would have clicked it before finishing the sentence that explained it.

I sat there. Cursor hovering.

Decades in advertising and strategy. A Midjourney alpha tester before most people had heard the word. The author of a book about thinking critically in the age of AI. And I was frozen in front of a dialog box like a man reading the terms and conditions on a parachute.

The hover lasted maybe ten seconds. It felt longer. Everything I've learned in the month since starts there.

The Long Blade

Everyone loves a hockey stick. Nobody looks at the blade.

Mine is six years long. It started in an alpha version of Midjourney, when a prompt was still mostly something your phone did to remind you of a dentist appointment. My early prompts were simple. Then they weren't. I started writing like a director briefing a cinematographer: the lens, the light, the hour, the weather, where the camera sits and what it refuses to show. Nobody told me to. The images just got better every time I described them more precisely.

When ChatGPT arrived, the habit came with me. I was already putting in more detail than the average bear. On a chart of my capabilities, those years would look flat. Some prompts, some images, some decent decks.

They weren't flat. They were compounding. I was practicing one skill, describing exactly what I wanted, without knowing what it would eventually be for.

Figma Weave changed its altitude. Weave works in nodes, discrete units of instruction wired together into a system. A camera move is one unit. A lighting rule is another. Nobody called it this at the time, but I was writing system prompts². Description had become decomposition.

From there, the step to agents, instruction sets and skills was shorter than it looked. Then, within the past month, the curve went vertical. I stopped finishing at the strategy deck or the static design layer and started saying: okay, that's the strategy, now build the app³.

I didn't learn five things in six years. I learned one thing five times, each time a little higher up.⁴

The Garden Wall

The first real shift came before the database, when I rebuilt the journal on ChristopherSmith.sg. It began as "wouldn't it be interesting if," the kind of thought I'd normally sketch in a deck and hand to somebody else. This time it came alive on a screen, and the question changed from "could this exist" to "now let's tune it."

The second shift was the button. Turning the site into a working web application meant moving between Claude, a GitHub repo, Vercel and Supabase. A stack I had never seen, in a vocabulary I didn't speak. I had never worked with a structured database in my life.

Engineers reading that sentence are already smirking. You built a table? Stop the fucking presses.

Fair. They're right. They're also beside the point.

Show the same sentence to a CMO or a creative director and they lean in, because they know the wall I mean. Every organization I've worked in kept one: creative on one side, engineering on the other, a ticketing system in between. The wall was never really about difficulty. It was about vocabulary, and it held because both sides found it useful. Engineers kept the mystique. Creatives kept the excuse.

A database is a decision about what matters, how it's grouped and how each piece relates to the rest. Strategists do that for a living. A brand framework is a schema.⁵ A positioning matrix is a table with better fonts. StoryEngine1 is a structure for stories, rows and relationships in everything but syntax. I had been thinking in databases for decades. I just never had the words, and nobody on the other side of the wall mentioned that's what I was doing.

Except the engineers aren't only guarding mystique. Knowing what a structure should hold is not the same as knowing how it fails, scales or gets breached. A creative who strolls through the gate without that humility ships something that breaks in public. The wall had one honest function: it kept people from shipping what they didn't understand. Take the wall down and that job doesn't disappear. It becomes yours.⁶

Verbs Before Sentences

Nothing broke.⁷ Not the journal, not the web app, not the event app that came after.

That wasn't luck, and I'm not that talented. I went slowly.

I never handed Claude a laundry list.⁸ One table. One feature. One skill pulled out of a GitHub repo. Then I looked at what had actually happened before asking for the next thing. When something was ambiguous, I let it stay ambiguous for a step and trusted the next step to clear it, rather than demanding the whole picture up front.

It's how you learn a language. Enough verbs to say something. Then adjectives. Then sentences that hold together. Nobody becomes fluent by swallowing the dictionary in a weekend, and the people who try don't remember much of it on Monday.

The instinct with AI runs the other way. Ask for everything. The whole app, the full campaign, transformation by Q3. The machine obliges, produces something sprawling and half-wrong, and the person on the other end decides the technology isn't ready. The flood creates the failure. The failure creates the fear. The fear becomes a policy.

My curve went vertical because each step was small enough to understand. Slowness wasn't the price of the hockey stick. It was the engine.

Reading the Fine Print

By the time I built AfterShow, an event content app with real data moving through it, the Allow button had changed.

I clicked it more often. I also spent longer reading what it wanted me to allow.

We usually describe confidence as the end of hesitation. You stop worrying and just go. That isn't what happened. My hesitation in that first hover came from not knowing what I was looking at. Confidence came from knowing⁹, and knowing made the review real instead of ceremonial. I read the query because I could finally understand the query.

There are three ways to meet an Allow button.¹⁰ Freeze, because you don't understand it. Rubber-stamp, because you've stopped trying to. Or read it, because every small step before it taught you enough to judge this one.

Every organization is about to be handed a great many of these buttons. Agents that want to send the email, update the record, run the query, approve the invoice, talk to the customer. Not one button. Thousands, every day, at every desk.

Most companies will freeze, or appoint a committee to freeze on their behalf. Others will rubber-stamp and call it efficiency. Very few will build the third habit, and no transformation program sells it. It gets built one table at a time, by people who went slowly enough to understand what they were allowing.

Your company is about to click Allow ten thousand times a day. Somebody had better know what it's agreeing to.¹¹