What changes when AI can execute too

When I started using AI in my design process, my first reaction was to try doing the same things faster.

Generating alternatives. Creating interfaces. Writing content. Exploring solutions.

Over time, I realized that was probably the least interesting part of the change.

What really began to change was my role during execution.

Executing faster does not mean deciding better

AI can produce a lot in very little time.

And that is precisely one of the problems.

A vague prompt can generate an interface that looks finished, but is full of decisions I never made.

I have been through that many times.

I would ask for one change and get five others. I would ask for a fix and another part would break. I would ask for a screen and the AI would start reinterpreting the Design System.

The greater the ability to execute, the more important it became to define what should NOT be done.

My prompts started to feel less like prompts

I started writing things like:

Do not redesign. Do not change existing components. Identify the error before fixing it. Change only the files that are necessary. Run the build and validate the result.

I realized I was no longer simply describing an interface.

I was defining context, constraints and criteria for execution.

That changed the way I see AI in my work quite a bit.

The work starts moving up a level of abstraction

If part of execution can be delegated, there is more room for a responsibility that remains mine: deciding.

What problem are we solving?

What needs to exist?

What should not exist?

Which pattern should we reuse?

Where is the experience becoming complex?

Does the result actually solve the problem?

AI can execute a decision quickly. But I still need to know whether that decision should exist in the first place.

There is still plenty that can go wrong

Not everything became simpler.

I have broken builds, generated unnecessary changes and run into issues with previews, deployments, configuration and integrations.

But those mistakes also made me realize that working with AI does not mean giving up control.

In fact, I am learning the opposite.

The more autonomy I give the tool to execute, the clearer I need to be about intent, context and boundaries.

Today, I use AI less to simply “make design faster” and more to expand how much I can explore, test and build from the decisions I make as a designer.