AI Won’t Kill the Design Process. It Will Expose Bad Ones.

Every few months, design gets a new existential crisis.

Right now, it’s AI.

We hear stories about teams barely opening Figma anymore. Designers are prototyping directly in code. Products can be generated from a prompt. “Vibe coding” is becoming part of the design vocabulary. And inevitably, this turns into a much bigger claim:

The traditional design process is dead.

I don’t think it is.

The process has changed enormously. The tools have changed. The speed has changed. The boundary between design and engineering is getting blurrier.

But the fundamental job of design – understanding a problem, exploring possibilities, making decisions, testing them and refining the outcome – hasn’t disappeared.

If anything, AI makes those parts more important.

Not all design work is the same

One thing that gets lost in conversations about AI and design is that we often talk about “design work” as though it’s one type of activity.

It isn’t.

I increasingly think about design as having two broad modes: unstructured work and structured work.

Unstructured work is what happens when the answer isn’t obvious yet.

A new product. A new workflow. A greenfield problem. An unfamiliar interaction model.

You’re moving things around, sketching, making connections, trying something ridiculous, deleting half of it, going backwards and suddenly realising that something you drew twenty minutes ago points towards a better solution.

This kind of work benefits enormously from a canvas.

Sometimes I know roughly what I want before I know how to describe it.

In that situation, drawing something in Figma can be faster than trying to construct the perfect prompt explaining the thing I could have sketched in thirty seconds.

Yes, I can ask an AI tool to generate five directions.

Sometimes that’s useful.

But if the idea is already forming in my head, a rough composition plus a prompt can be far more efficient than repeatedly describing the entire interface in words.

That doesn’t mean AI failed.

It means the canvas was the better input mechanism.

Then there’s structured work.

You already understand the product. The components exist. The patterns are established. You’re adding a smaller feature, extending an existing workflow or changing a relatively contained piece of functionality.

Increasingly, I think a lot of this work can happen directly in code.

You can explore an idea, build it, adjust it and interact with the real thing before you’ve created a polished Figma file.

That’s genuinely transformative.

For certain types of work, “design → prototype → engineering handoff” collapses into something closer to:

think → build → evaluate → refine.

That’s a meaningful change.

But it doesn’t mean every design problem should now begin with a blinking cursor.

The goal isn’t to use AI. The goal is to be faster.

There’s another strange behaviour emerging from the AI boom.

We’ve become so obsessed with whether something can be done using AI that we sometimes forget to ask whether it should be.

Imagine I want to change the visual treatment of a screen.

I prompt the model.

The colour is wrong.

I prompt again.

The contrast is now strange.

I prompt again.

Eventually I get something close to what I wanted.

Technically, I designed it with AI.

But if I could have changed three values in Figma in fifteen seconds, what exactly have I achieved?

Using AI is not automatically efficient.

If it takes three prompts to change a colour, we’ve failed.

The interesting workflow is not replacing every existing tool with a prompt box.

It’s connecting them.

Maybe I explore the look and feel visually in Figma, where manipulating visual properties is extremely fast. Then I use MCP or another bridge to push those decisions back into the working product.

Maybe I begin in code, realise the composition isn’t working, jump into a canvas to explore alternatives, and then bring the result back.

That feels much closer to where design tooling is going.

Not:

Figma or code.

But:

Figma when Figma is faster. Code when code is faster. AI connecting the two.

The measure of success should be how quickly we get to a good decision – not how many traditional tools we managed to avoid.

The Double Diamond didn’t disappear

AI has radically compressed parts of the design process.

Research transcripts can be summarised in minutes.

Large amounts of qualitative feedback can be clustered.

Product data can be explored conversationally.

Competitor patterns can be investigated faster.

Prototype directions can be generated almost instantly.

Functional prototypes that previously required several people can sometimes be created by one designer in an afternoon.

But underneath all of that, I still recognise the Double Diamond.

We still diverge.

We still converge.

We still try to understand the problem before deciding what to build.

We still explore multiple possibilities before committing.

We still test whether our assumptions were correct.

AI hasn’t removed those loops.

It has increased the velocity inside them.

And that distinction matters.

Because if you misunderstand the problem faster, generate the wrong solution faster and ship it faster, congratulations: you’ve simply created a more efficient bad design process.

Speed amplifies whatever process already exists.

Good judgement becomes faster.

Bad judgement does too.

Human judgement matters more, not less

This is probably the part of AI-assisted work that concerns me most.

LLMs are extremely convincing.

They can analyse research and give you a beautifully structured explanation of what customers supposedly want.

They can examine data and confidently identify a “trend”.

They can critique an interface.

They can suggest a product strategy.

And the response can sound excellent.

The problem is that sounding correct and being correct are different things.

LLMs can misunderstand context, over-index on patterns that aren’t meaningful, invent causal relationships or simply make things up.

And they have another characteristic that is particularly dangerous in creative work:

They are extraordinarily agreeable.

Tell an AI your new onboarding concept is brilliant and it will probably explain why it is brilliant.

Tell it ten minutes later that the same concept is fundamentally broken and it can often produce an equally convincing explanation of why you’re right.

That makes human judgement more important.

You still have to ask:

Does this actually make sense?

Is this insight supported by the research?

Is this correlation meaningful?

Is the prototype solving the problem or merely looking impressive?

Should AI automate this decision at all?

What are we missing?

AI can dramatically increase how much information we process and how many things we create.

But somebody still has to decide what deserves to survive.

That is increasingly where the value of a senior designer sits.

Not drawing every rectangle.

Not manually summarising every interview.

Not pushing every pixel.

Making good decisions.

Be careful with the AI experts

There’s an entire industry forming around people’s anxiety about being left behind.

AI courses.

Prompt libraries.

AI design bootcamps.

“Become an AI-native designer in 30 days.”

Secret prompting frameworks.

Threads explaining the 47 tools every designer supposedly needs to master before next Tuesday.

Some of this is useful.

A lot of it is FOMO packaged as expertise.

The uncomfortable truth is that almost everyone is figuring this out at the same time.

These tools are changing monthly. Workflows that seemed revolutionary six months ago can become obsolete after a model update.

Which is why I’d be very sceptical of anyone claiming to have discovered the definitive AI design process.

Particularly if they’re willing to reveal it to you for £499.

Experiment yourself.

Build something.

Break it.

Use Claude. Use Cursor. Use ChatGPT. Connect Figma. Generate an interface from scratch. Try designing the same thing manually. Compare the results.

Ask where AI genuinely saved you time.

Ask where prompting became more cumbersome than simply doing the work.

Ask which parts of your process became dramatically faster and which still require you to sit there and think.

You may discover that the person selling you the course doesn’t understand your workflow any better than you do.

And that’s fine.

There is no secret prompt.

The designer’s job is changing

I don’t think the future designer is someone who heroically refuses to touch Figma.

Nor do I think it’s someone who ignores AI and continues working exactly as they did five years ago.

It’s someone who becomes much more fluid about how ideas become real.

Sometimes that means starting with research.

Sometimes a sketch.

Sometimes a prompt.

Sometimes a prototype.

Sometimes production code.

The artefacts matter less.

The ability to move between them matters more.

And AI removes a huge amount of friction between thinking something and making it tangible.

That is incredibly exciting.

But the danger is confusing reduced friction with reduced need for design.

The easier it becomes to build things, the more important it becomes to know which things are worth building.

That part hasn’t been automated.

And I’m not convinced it ever will be.

AI Won’t Kill the Design Process. It Will Expose Bad Ones. was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.