The last couple of years have been really fun in the product design and software space. On its surface the product design industry can appear very different in 2026 compared to, say, 10 years ago. Yet, it's very much the same at its core as it always has been.

Sure, production of static mockups is not a thing anymore. Gone are the days of designers being expected to just produce a Figma link. With LLMs as a tool, there is zero reason for designers to not write code, experiment, and push to production. However, that doesn't mean that's the only creation layer left.

The two design layers

The two distinct layers used for creating something are still canvas and code. The auxiliary agent layer is complementary to both (for example: the newly released Figma agent working alongside Claude).

I don't believe that design by prompting will ever replace the traditional, canvas-based method. The two approaches, however, work great with each other, allowing one to start with a prompt, and continue on canvas, then expand into code. However, I don't think either will ever be self-sufficient, even though a lot of tools are promising to solve both in one.

Canvas is super valuable for a quick and dirty exploration. I don't think it will ever go away, because you can never reach the same level of experimentation with another method.

At the same time, I find myself using Figma much less than I used to. It's just so much easier to start a wireframing session in Claude, and only then transition to high-fidelity for specific moments. There's just no point in drawing out every pixel.

The process

More often than not, I find myself starting with Claude when solving a specific problem, whether it's a bigger UX issue, or a small UI tweak. It's great to have something to kick ideas back and forth with. Although, of course, I don't think it would ever replace working with another designer or a product person (nor that it should!)

For example, I find it super useful to put a brain dump and work through a few directions. I prefer Claude on the web for this, just so that I can visualize things quickly, and discard things that just don't make sense.

It's really nice to be able to explore a few ideas quickly, and see some basic wireframes to prove an idea, or completely discard it.

Explicit directions

The other way I like to use AI is to build something that I have a very good idea of. This is where the explicit directions come in. This method becomes super helpful when you know exactly what you want to build.

My preferred method for this is to work with Claude Code right in the terminal. It's a great way to build small features or UI tweaks—I already have the idea of the final product in mind, so the goal is to get there incrementally. Then, you just gotta polish the last 20% by hand.

Of course, this doesn't replace full high-fidelity design, even when hooked up to a proper design system and given directions/context. Usually, you can tell pretty quick when something has been generated with AI all the way. However, it's a great starting point, and allows for a super quick development process. Some things are just much easier to design in code with real data, rather than going to a canvas first.

The blurred lines

With the speed of shipping accelerating, the bottleneck for building software mostly becomes figuring out the small details, as well as deciding what to build. That's where the traditional product management hat comes in. As a designer, you have to have strong product sense, otherwise you will get buried in a pile of decisions that need to be made.

In general, the lines between engineering, product management, and design are much more blurred now. However, it doesn't mean that either role is about to be extinct.

At the end of the day, it's all about jobs to be done. The reality is that there's just much more overlap now between these three disciplines. The best designers have always had a deep understanding of the user problems, as well as the tech stack to translate the solutions to. Now, designers finally have the ability to shape the end product themselves. There's no more need to bother an engineer to adjust some padding on a specific element.

The future of design has never been more exciting—if you're willing to embrace it. There's more work than ever to be done, especially if you consider yourself more of a generalist/builder. You just need to care in order to build something that will stand the test of time.