I walked into the office this morning and told my team, who was working on retouching images for a catalog we produce, that “AI is coming for your repetitive tasks!” It got a laugh. But underneath the joke is a serious matter for creatives — and it’s the same conversation we’re about to have.
There are two conversations happening about AI in this industry right now, and almost nobody is having the right one.
The first conversation is fear: “Is AI going to replace developers? Designers? Agencies like us?” That question gets asked in every Slack channel, every LinkedIn post, every client meeting where someone half-jokingly says “couldn’t ChatGPT just build this?”
The second conversation — the one that actually matters — is about speed, and who’s using AI to get faster without getting worse.
That’s the conversation we’re having at Dirt.
The fear is pointed at the wrong thing
AI isn’t going to replace you. It’s not smart enough, it’s not accountable, and it doesn’t know your client’s brand guidelines, their PM’s pet peeves, or why the last agency’s build broke on iOS Safari. What it can do is compress two weeks of work into two days for someone who already knows what they’re doing.
That’s the actual threat. Not the tool — the adoption gap. We’ve seen this movie before. Photoshop didn’t kill illustrators; it killed the illustrators who wouldn’t learn it. WordPress didn’t kill web shops; it killed the ones still hand-coding every page from scratch in 2015 while everyone else shipped in a third of the time. AI is the same shift, just faster and louder.
The people who lose aren’t the ones “replaced by AI.” They’re the ones who kept working the old way while someone equally skilled — or less skilled but faster — ate their timeline and their price point.
What “AI slop” actually looks like
Here’s the part most companies won’t say out loud: most people using AI right now are bad at it. Not because they’re unintelligent — because they don’t have the background to know when the output is wrong.
They ask for a component, get something back that looks right, and ship it. They don’t know it violates their own theme’s CSS specificity. They don’t know the copy reads like every other AI-generated landing page because it is every other AI-generated landing page. They don’t know the “clean” code has a bug that only shows up on one browser, because they’ve never had to debug someone else’s code by hand at 1am and develop the instinct for where things break.
That’s AI slop. It’s not a style, it’s a tell. It’s what happens when the person operating the tool doesn’t have the craft to catch what the tool gets wrong.
What using AI well actually looks like
At Dirt, AI isn’t the builder. It’s leverage on top of people who already know how to build.
In practice, that means:
- Scaffolding, not shipping. AI gets us a first-pass component or layout fast. Then we go through it the same way we’d review a junior dev’s pull request — because that’s functionally what it is.
- Debugging with it, not trusting it blindly. AI is excellent at generating plausible-looking code and mediocre at knowing your specific theme, your specific stack, your specific client’s quirks. We know where those gaps are because we’ve been doing this long enough to know where the bodies are buried in Elementor and Salient.
- Speed on the parts that don’t need taste. First drafts of copy, boilerplate structure, repetitive JSON/PHP scaffolding, batch image retouching for a catalog — hand that to AI. Save the human hours for the parts that actually require judgment: what the client means versus what they said, what the brand should feel like, what will actually convert.
That’s the joke I made this morning, and also the point. Retouching a hundred catalog images one by one is repetitive work that AI can absolutely take off someone’s plate. But someone still has to know what “on-brand” looks like, catch the one image where the AI clipped a shadow wrong, and decide which shots are worth a human’s eye in the first place. The task gets automated. The judgment doesn’t.
The output looks like ours, not like a template, because we’re still the ones deciding what’s good.
Where AI still can’t save you
No amount of AI fixes:
- Client trust, built over years, not prompts
- Taste — knowing when a design is technically fine but actually wrong
- Institutional knowledge — knowing which override will quietly break which plugin
- Understanding what a client actually wants when what they typed and what they meant are two different things
This is the stuff that doesn’t show up in a prompt. It shows up in experience. That’s not a nostalgic take — it’s the actual moat right now, and it’s shrinking for anyone who doesn’t reinforce it.
The real skill shift
The old value proposition was “I can build X.” The new one is “I can direct and vet X being built, faster than you can, without it looking like everyone else’s AI output.”
That’s not a lesser skill. It’s arguably a harder one — it requires knowing enough to catch what’s wrong in something that looks right, which is a much higher bar than building it slowly from scratch and catching your own mistakes as you go.
The takeaway
Stop asking if AI is going to replace you. Start asking if you’re using it well enough that it couldn’t. The gap that matters isn’t between companies that use AI and companies that don’t — everyone uses AI now. It’s between companies that know how to direct it, and companies quietly shipping slop and hoping nobody notices.
We know which side we’re on.