Designer or tool? When AI design software replaces the hire

Buy the tool when you know what you want and need more of it. Hire someone when you don't yet know what the product should be, or when nobody can say what "good" means here. The software has genuinely taken over production; it has taken over none of the deciding, and most teams shopping for one of these are short of the second thing.

I sell design leadership, so I'm on one side of this. I've tried to be specific about the cases where the tool is the right purchase, because a comparison that always lands on the author's product isn't a comparison.

The question is confused before it's asked

Ask a search engine who the best AI product designers are and you'll mostly get software back — image generators, the AI features inside design tools, research and testing platforms, site builders. Ask for the best AI design tools and you'll get the same list.

That's not a mistake anyone made. It reflects a real ambiguity: for a lot of work that used to require a person, buying a seat now genuinely is an option. The confusion only causes damage when the thing you needed was never production in the first place, and you spend two quarters discovering that the bottleneck moved rather than went away.

What the tools actually do well now

More than the sceptical version of this argument admits, and it's worth being straight about.

They produce a plausible first pass of almost anything, fast. They generate variants at a volume no person can match, which is genuinely useful when you're exploring rather than deciding. They fill out a system once its rules are specified — states, sizes, the tedious completeness that used to eat a week. They handle a large amount of research and testing logistics. And they let one person hold a surface area that used to need a team, which is the change that matters most.

I'm not describing this from the outside. There are 203 pipeline-generated product screens in the case studies on this site, and a convincing one costs me a command. Production is not the scarce thing anymore, and anyone telling you otherwise is selling something.

What doesn't come with a seat

Four things, and they're the ones that were always expensive.

Deciding what should exist. A tool will build whatever you describe, including the wrong product, quickly and to a high standard. Nothing in the software has an opinion about whether the feature should be there.

Defining what "good" means here. Not in general — in this product, for these users, against this constraint. That definition is what lets everyone else work without you, and it's the actual output of design leadership.

Deciding what happens when the model is wrong. In a product built on inference, the interface is largely determined by the failure cases, and a generator producing a beautiful happy path has not touched that question.

Saying no. Every request to add one more variant, one more colour, one more component is a request to avoid a decision. Tools are constitutionally incapable of refusing, which is exactly why systems generated without someone holding the line drift within a quarter.

What's actually scarce

The useful question isn't tool-versus-person. It's which of these you're short of.

Most teams reading a tool comparison are in row three and shopping in row one, because row one is the purchase you can make this afternoon without a difficult conversation.

Tools amplify whatever specification they're given

This is the mechanism underneath all of it. A generator multiplies the clarity you hand it. Given a well-specified system, it produces a great deal of consistent, correct work. Given a vague brief, it produces a great deal of confident, plausible, subtly inconsistent work — which is worse than producing nothing, because now there's volume to review and no principle to review it against.

That's why "we bought the tool and the output got worse" is a real and common outcome. The tool didn't get worse. It faithfully amplified an absence.

The corollary is the actual case for a person: someone whose job is the specification makes the tool dramatically more valuable. Generation should touch tokens and components rather than individual screens, and that division only exists if someone drew it.

When the tool really is the right purchase

Four cases, and I'd say so in a sales call.

You're pre-revenue with taste and no budget. You have opinions about what the product should be and you need to see it. Buy the tool and build it yourself; hiring me at this stage would be a waste of your money and my time.

You need a marketing site, a deck, a landing page — bounded, one-off, with a clear reference. That is a solved problem now and it does not need a designer at all.

You already have a design leader and a system. The tool is straightforward leverage for the team you have, and the decisions are already being made by someone.

The work is genuinely a volume problem. A large backlog of well-specified screens is production, and paying senior rates for it is a mistake I'd talk you out of.

When it isn't

You cannot write down what the product should do. More screens will not fix that, they will just make the confusion higher-fidelity.

The product is built on a model and nobody has designed the wrong-answer cases. The tool will happily produce the version where everything works.

You have three people with three different ideas of what good looks like. Generation multiplies that disagreement rather than resolving it, and you end up with three plausible directions and no way to choose.

Design quality is drifting and you can't say why. That's a systems problem — the vocabulary got loose and nobody said no — and buying more generation accelerates it.

The honest summary: the software has eaten the production half of this job and it isn't giving it back. What it hasn't touched is deciding what to build and what good means, and if that's your gap, more output is the expensive way to avoid the conversation. If it isn't your gap, buy the tool — you don't need me. Here's how I work and what it costs if it is.