AI is a power tool, not a carpenter

It’s 2024 and no meeting is safe from the question. Client kickoffs, team lunches, a wedding reception one of us attended in June: sooner or later somebody asks what we think about AI, and they ask it the way people ask about weather that might be a storm.

The industry has sorted itself into two congregations. One is holding a funeral, certain that design, copywriting, and development have eighteen months to live. The other is holding a revival, certain that anyone not rebuilding their entire company around these tools deserves what’s coming. Both congregations are loud, and each is selling something, even when the something is just its own certainty.

We keep thinking about the nail gun.

When pneumatic nailers spread through job sites, a framing task that took a carpenter most of a day started taking a couple of hours. Circular saws did the same thing to rip cuts a generation earlier, and routers did it to edge work, and nobody now argues that any of it was cheating. Here’s what those tools never did: they never decided where the wall goes. A nail gun has no opinion about whether the deck suits the house, whether the doorway should be three feet wider, whether this whole addition is a mistake the family will regret. It compresses the distance between a decision and its execution. The decision still has to come from somewhere.

Carpenters did not disappear. Bad carpentry didn’t either, and this is the part both congregations skip. Give a careless framer a nail gun and you get careless framing at triple speed, crooked studs installed with astonishing efficiency. The tool amplifies the hand holding it. It has never once improved the hand.

The view from the shop

That’s roughly where we’ve landed with the current wave, as working practitioners rather than commentators. These models are genuinely startling. We use them most weeks now: first drafts of boilerplate code, throwaway variations of a layout, summarizing a fifty-page discovery document into something a human can hold. The time savings are real and we’d be posturing to deny them. What the tools have not once done is know what the project is for. They generate options with total confidence and total indifference, and confident indifference is exactly what you want from a saw and exactly what you can’t accept from a builder.

Somebody still has to look at the client’s actual business and decide what’s worth making. Whether the brand should sound like this, whether the feature everyone’s excited about solves a problem anyone has, whether the honest answer is a smaller site, a shorter video, no app at all. Those calls were the expensive part of our work before 2023, and if anything they’re more expensive now, because when execution gets cheap the whole project becomes the decisions, and a wrong decision executed instantly is still wrong. It’s just wrong sooner.

Two sermons

To the funeral congregation we’d say: the people who lost work when power tools arrived were mostly the ones whose only offer was manual repetition. The ones who understood structures, who could walk a site and see what the drawings missed, got more valuable, not less, because their judgment now moved at tool speed. There’s a version of that sorting coming for us, and pretending otherwise helps nobody.

The revival tent gets a different sermon. We’ve now reviewed a fair amount of machine-generated work product, and the tell is never quality in the technical sense: it’s that nobody decided anything. Every choice sits at the median. The copy is grammatical and could belong to any company on earth, which in a market where everyone owns the same tools is another way of saying it belongs to no one.

And to both we’d admit the obvious: it’s September of 2024 and nobody knows where this lands, including us. The tools are improving at a pace that makes every confident essay, presumably this one included, a hostage to the next release. We’re writing this down partly so we can check our own thinking in a few years, the way you keep a photo of a framing job to see whether the joints held.

What we’re not doing is panicking, and what we’re also not doing is handing the shop over. New tools have been arriving at Colony for fourteen years. Each one changed how we build. None of them changed why anyone hires a builder, which is that a structure has to hold weight, fit its site, and serve the people inside it, and no tool has ever cared about any of that.

A nail gun in a drawer builds nothing. Neither does one firing at random.

The interesting work is still deciding where the nails go.