Design is now the bottleneck, not code.

My cofounder has been whipping up demos for design tools of the future and his one complain with me is I'm slow. He needs UI updated, consistent with how we show it to the world and regular UX audits to ensure the flow is simple (because as we all know by now, AI design is a maximalist).

'Design is fast, code is slow' - that's been the assumption for most product teams for decades. We could create mockups in Figma while Devs spent weeks converting those static frames into a working software. The bottleneck was always implementation.

That's no longer true.

Software prototyping is now faster than Figma prototyping. Wiring up a fake version takes longer than building the real thing (I didn't imagine I'd ever say this). All of my future predictions of design have been around UX and the interface itself but now the practice itself is changing, like most things today.

"Something has changed where the bottleneck is now almost solely design. Less time coding = more time perfecting." - Jordan Singer, founder of Mainframe.

Geoffrey Litt from Notion describes what this looks like in practice: AI can ask him 50 interview questions about a spec in rapid succession, reaching such clarity that it can then one-shot a big feature. Prototypes that would have taken days now take hours. "The main bottleneck," he writes, "is my own decision making and judgment."

The constraint has now shifted upstream; where implementation is cheap and knowing what to build is expensive.

Close to the code, without coding

I read something recently that captured exactly what I've been experiencing:

"We changed our whole design process overnight. The weird part is: I didn't want this change. I liked designing slowly. I liked the craft.
I liked caring about details. Designing shadows to perfection. But the
reality is: engineering got faster, and I became too slow. And when
more people started using Figma Make, the concepts were good… but the
output didn't have the polish. It didn't feel right. So I downloaded
Cursor, connected it to our Git, and started prompting designs based
on our actual components. WOW – that was mind-blowing."

This is the path I'm on now. Not learning to code. Learning to work close to the code.

I now write and maintain markdown files - (1) the decisions (2) the reasoning (3) the vibe; what should this feel like, why that, what are its constraints etc. For Aiverse members, we've started providing all of our AI patterns as actionable .md files. In fact, whenever Rahul (my cofounder) is working on a new AI feature, he literally just copy and pastes it into Claude or Cursor and the UX of the prototype becomes better.

For example this is the before vs after Semantic search. From getting a basic input field to AI adding suggested prompts, a step-by-step processing step and confidence score when we pasted our Onboarding AI patterns.

1. First AI-generated draft:

2. After giving the coding agent our AI patterns .md file:

It added the suggested prompts, drop-down suggestions the processing steps and more explanation for transparency.

3. Picking the style of another website (using our AI style picker) and applying it to the app:

The transformation is crazy! All this without me actively being present.

That's my new role now. Maintaining and curating these .md files for faster implementation of good UX. Essentially enabling non-designers to design.

We also have a secret weapon that we use internally, our /playground, to get closer to the code. It's a design canvas built on top of Cursor or Claude. Essentially, it allows me to code software the same way I design in Figma. Design is the new code. I can add any of my prototypes on the canvas, create variations of any component using natural language and deploy to the actual app with one click. I can explore more concepts, test out different prompts thus making my decision making faster.

This I believe is the real shift taking place in design work. It is moving from crafting individual screens to crafting the system that generates screens. When AI can produce infinite variations, the value isn't in making one perfect mockup. It's in defining the constraints, the principles, the components that make all the outputs feel right.

This also means design decisions can't stay bottlenecked in a design team. The judgement of knowing which of the 50 possible directions is correct needs to be distributed.

Whether it's product managers evaluating prototypes or engineers choosing between AI-generated options, everyone is making design decisions, because design is just decisions that help users navigate intuitively.

So start building these set of files for distribution, your design context or as we internally refer to it as "Design Intelligence".

And to end this with my favourite quote (lately), "either go with the times or go with the times".