Finding space for craft in the age of speed
发布时间：2026-08-20 ｜ 浏览：5
We use cookies to deliver personalized ads and experiences, and to analyze traffic on our site. Read our cookie policy for more information.
By Bobby Marko
How new AI-powered capabilities are expanding what designers can create
The conversations about AI all seem to point to one answer: speed. Ship faster. Iterate quicker. Do more with less. And there's truth to that, product development is accelerating, and design is adapting to keep pace.
But I keep wondering about a different question. What if, alongside going faster, there's also an opportunity to go deeper? To focus on craft and delight? That's what I've been exploring at Pinterest, and what I've found has changed how I think about what's possible.
Better work, not just faster work
While some are proclaiming the design process dead¹, at Pinterest, we see following a process (discover, define, design, and deliver to our users) as more important than ever. In a world where execution is only getting faster, design's role in helping the organization decide what to build and when to launch with the right amount of craftsmanship becomes critical. That's how we create experiences that truly work for the people who use Pinterest.
The design process isn't dead. It's more important than ever. AI just gives us new ways to do it well.
For us, the real motivation for using AI in design isn't about productivity hacks. It's about enabling work that was previously too complex, too expensive, or just not possible for most designers. AI supports deeper thinking at every stage, from discovery through delivery.
Here's what that's looked like across our team.
Discover: An AI thinking partner in brainstorming
Our team was kicking off a project exploring ideas for an AI-powered assistant experience. We brought together a dozen people from across the organization for a large brainstorming session, and by the end, we had over 200 ideas on a shared whiteboard.
In the past, synthesizing that volume of input meant hours of manual sorting. Grouping sticky notes by gut instinct, debating categories, and losing good ideas in the shuffle. This time, we had an AI agent in the room with us. We'd brought it into the session from the start, so by the time we were done brainstorming, it had the same context we did: the whiteboard, the conversation, the project goals.
When it came time to organize, we asked the agent to suggest groupings. In the time it took us to produce one set manually, the agent had produced three completely different frameworks we hadn't considered. We put all four versions side by side and actually preferred the agent's over our own. It had surfaced patterns we'd missed.
The agent made us realize the most valuable pattern was a strategic one. We had pulled in people from across the organization for this brainstorm, and each team was focused on their own corner of the product. But the agent's groupings cut across those silos. They showed us that investing in shared capabilities like new tools and new ways to generate UI could unlock experiences across the entire platform, not just within one product area. It was obvious once we saw it, but none of us had gotten there on our own.
From there, we worked with the agent to create a scoring rubric, and it built a custom dashboard that evaluated all 200+ ideas against our criteria. That let us quickly zero in on which ideas were worth going deeper on.
As the project moved forward, the agent kept getting more useful because it carried all of this context with it. When we moved into storyboarding and early prototyping, it could help us build artifacts that actually reflected our ideas without us having to re-explain everything from scratch.
The agent didn't replace our thinking. It extended it. We still brought the judgment and the taste. But it helped us see across boundaries we didn't know we were working within.
Design: Designing with real data, not assumptions
Yaxin, a Senior Product Designer on our team, was working on a project that required her designs to work across a wide variety of product imagery. In the past, this meant manually pulling together a handful of example images and designing against those. You'd make decisions based on limited data, then hope for the best until the technical solution was built and you could finally see how things looked in practice.
With a data analytics agent, Yaxin could query our datasets directly and pull in actual product Pins that met specific criteria for her project. Thousands of them. From there, she used agentic coding tools to build a custom preview tool that let her test her designs against all of them at once.
"Before, I'd design against maybe five or ten example images," Yaxin told me. "Now I can see immediately how a design holds up across the full range of what people actually submit."
That changed how she designed. Instead of working from a handful of examples and hoping things held up at scale, Yaxin could see immediately what was working and what wasn't. She'd tweak a background treatment, refresh the preview and know right away if it fell apart on certain types of imagery. The tool wasn't just checking her work after the fact. It was shaping her decisions as she went.
Q, a Senior Staff Product Designer, had a different design challenge. The project involved reconciling two overlapping but separate product features, a strategic question that's hard to discuss in the abstract. People can have very different interpretations of what the experience should feel like, and without something tangible, conversations tend to go in circles.
Using an agentic Integrated Development Environment (IDE), Q built working prototypes in a few days. These weren't simple mockups. They were functional enough for leadership to interact with and give real feedback.
"I thought I'd be building for weeks," Q told me. "Instead, I had something in front of leadership in a few days, and we made decisions in that meeting that had been stuck for months."
Yaxin and Q's projects looked very different, but the throughline is the same: AI let them explore more broadly and make better decisions earlier. When you can test ideas against real content or put a working prototype in front of stakeholders in days instead of weeks, you catch problems sooner and build conviction faster.
Develop: Prototypes that actually work
Here's where things got personal for me. Classic clickable prototypes didn't apply to what I was working on, an AI-powered agentic experience where the model's capabilities and voice are as much of the product as the UI containing it. You can't get meaningful feedback on something like that from a static screen. Users need to actually interact with the model.
So I built a prototype that integrates with our actual models. Using an agentic IDE, I had a working experience up in only a few prompts. I added voice and tone dials so we could test different approaches with users and get clear direction on the model's personality before a single line of production code was written. And because we built it ourselves, we could obsess over the details that make an AI interaction feel right. How will the model respond? What personality comes through in the language? The small moments that only show up when you can iterate on the full experience, not just the visual layer.
But the story didn't end with one designer and one prototype. We added the ability for ten designers to collaboratively edit the same prototype in real time. Think of the multiplayer experience that makes collaborative design tools so great, but applied to functional prototypes calling actual APIs and language models. Our biggest priority was zero setup. No configuration, no environment wrangling. Designers could just open the prototype and start bringing their ideas to life through words and images. When you take that friction away, what people come up with is inspiring.
The prototype had access to over twenty different tools, pulling in real data and connecting to real services, so we could actually feel what our product would be like with all the capabilities we were building toward. In a matter of hours, ten designers brought that vision to life together.
You have to use something to know if it feels right, and you just can't get that from static images or clickable prototypes. We're also finding that these prototypes help our ideas get traction much faster. I've had several instances where a team saw something a designer explored through a couple of prompts and immediately moved to implement it. When people can experience an idea rather than just look at it, decisions happen faster and conviction comes more easily.
Deliver: From logging bugs to fixing them
In the past, design's recourse to improve the final product was logging bugs and advocating for their prioritization. You'd see a spacing issue, a color that wasn't quite right, a transition that felt off, and you'd add it to the backlog and hope it eventually got picked up.
Wes is a designer on our team who had never written a line of code. No familiarity with the codebase. No engineering background. But he saw a spacing issue that had been bothering him for months, and with the help of an agentic IDE, he was able to fix it himself and send a pull request to the engineering team.
"I'd never written a line of code at the company," Wes told me. "But in a matter of minutes, I was able to fix this long-standing error myself and send it to the team. That felt like magic."
This isn't about designers becoming engineers. It's about removing the friction between seeing a problem and fixing it. Design is now empowered to pick up fit and finish issues that never got prioritized and just take care of them.
The result is clearer handoffs, prototypes that show behavior instead of just layouts, and stronger alignment between design, product and engineering. Fewer assumptions make it out to Pinners.
Designing what wasn't possible before
What has AI changed for designers at Pinterest? We're spending more time on judgment and taste, and less time on the repetitive tasks that used to fill our days. Day-to-day, that means more space to obsess over the details that create delight. The small interactions that make an experience feel considered. The moments of surprise that make people want to come back. AI is supporting craft, not replacing it.
The mindset shift is real. Design is more hands-on in execution now. In the past, we'd advocate for polish through bug reports and prioritization discussions. Now we can add that polish ourselves.
The question I started with was whether speed is the only answer. For us on the Pinterest Design team, the answer is no. AI has given us deeper insight into how designs work at scale. Better decisions because we can explore more broadly. Higher-quality outcomes because we can iterate on the full experience, not just pieces of it. And more space to focus on the craft and quality that serves our mission of bringing everyone the inspiration to create a life they love.
What could you design if you had more room to go deep?
Here's to designing what wasn't possible before. With intention, craft and care.
¹ “The design process is dead. Here’s what’s replacing it. | Jenny Wen (head of design at Claude)”, Lenny's Podcast, United States, March 2026.
Artificial intelligence
Product design
No Related Posts found