I’ve been using Photoshop for almost two decades. I teach it. I write about it on this blog every week. And I want to be honest with you about something that’s been bothering me for a while, because I don’t think enough people in our craft are saying it out loud.

The AI tools that shipped in Photoshop over the last two years are genuinely useful. Generative fill, generative expand, neural filters, Sky Replacement, content-aware masking. They save time. I use them. You probably use them.

They are also built on a foundation that I think we should all be paying more attention to. And the long-term cost of casually adopting them is starting to show up in places we don’t usually look.

What These Tools Are Actually Doing

When you click Generative Fill in Photoshop, you’re asking a model that has been trained on hundreds of millions of images. Most of those images were created by working artists, photographers, illustrators, and designers. Almost none of them gave informed consent to have their work used as training data, because the question wasn’t really asked when the models were being built.

This isn’t a conspiracy theory. Adobe is open about training Firefly on the Adobe Stock library plus public-domain content. That’s the cleanest training pipeline in the major AI image space, and it still doesn’t ask the stock contributors to opt in any meaningful way. The other major models (Midjourney, Stable Diffusion, DALL-E) were trained on substantially less curated datasets that included plenty of work whose original creators have publicly objected.

This matters because every time you generate AI content in Photoshop, you’re benefiting from the unpaid creative labor of people who didn’t know they were contributing.

The Effect On The Craft

I want to talk about something else though, because the ethical conversation has been had a hundred times and I don’t think I’m going to move the needle on it.

What I want to talk about is what these tools are quietly doing to the craft of Photoshop work, and specifically to the people learning Photoshop right now.

A few weeks ago, I was helping a younger designer with a complex composite. The brief was a hero image for a campaign, three subjects against a fabricated background, complex hair selection on one of them. The kind of job that, ten years ago, would have been an afternoon of careful work with the Pen tool, refined edges, multiple masks, and a real understanding of how channels separate.

She didn’t know how to do any of that. Not because she was lazy or inexperienced. She had several years of Photoshop work behind her, mostly on smaller jobs. She just had never needed to learn the underlying techniques because the AI tools had handled them in everything she’d worked on. The first time she hit a job where the AI subject-selection produced a wrong result, she had no fallback.

This is not a hypothetical problem. It’s already here. The retouchers and compositors I know who came up in the past five years often cannot do the work the AI does for them. When the tool gets it wrong on an edge case, they’re stuck, because they never built the muscle memory for the manual version.

The craft is being hollowed out underneath us.

Why Originality Suffers

The other thing I notice, and this one’s harder to put into words, is that the work being produced with heavy AI involvement is starting to look the same.

Open any “AI-enhanced” landing page or campaign right now and run your eye across it. The skies have the same composition. The bokeh has the same character. The skin smoothing has the same look. The color grading has the same tonal range. AI models are trained on what’s popular, which means their outputs trend toward the visual center of the dataset. When everyone uses the same models, everyone’s work starts to converge on the same aesthetic.

This is bad for designers and bad for the brands they work for. A campaign that looks like every other campaign is invisible. A portrait that has the same skin treatment as the rest of the AI-touched portraits in the feed doesn’t stand out. The competitive advantage of being a skilled Photoshop operator was always partly that you could make work that looked like nothing else. That advantage is being commoditized.

What I Recommend

I’m not telling you to stop using AI tools. They’re embedded in Photoshop now and they’re not going away. But I am telling you to be intentional about how and when you reach for them.

Don’t use them on the parts of the job that distinguish your work. If you’re known for a particular kind of compositing or color treatment, do that part by hand. Use AI for the boring fills and the repetitive masking, where it saves time without affecting the work you’re being hired for.

Keep the underlying skills sharp. Practice manual selection. Practice frequency separation. Practice channel-based masking. If you can only do the work the AI does it for you, you’re vulnerable to every edge case the AI fails on, and there are more of them than the marketing suggests.

Be skeptical of “AI-first” workflow advice. Most of it is written by people who are paid to be excited about AI tooling. The honest advice usually comes from working designers who use the tools quietly and selectively, not the loud voices selling the future.

Pay artists. If you commission references for your work, commission them from real artists. If you license stock, license from photographers who got paid for the shoot. The cleanest way to push back on the upstream training-data problem is to keep paying for human creative work even when AI-generated alternatives exist.

The craft will survive this. But the people who make it through with their skills intact are going to be the ones who refused to outsource the thinking part, even when the tool made the thinking optional.