The Designer-Builder: How AI is Collapsing Creative Boundaries
From Google Translate to Slack: Pendar Yousefi on the Future of AI and Design
Pendar and I met a few years ago, and we’ve been having coffee conversations since then. Recently, we met virtually for the Stories from the Field series and had a great time talking about the changing nature of design, agentic flows, and his perspective on taste.
Without further ado, enjoy.
Kursat: Can you tell us about yourself and your journey as a designer? What brought you to where you are today?
Pendar: I grew up in a family of engineers and artists. Art classes were a serious commitment in our house, and I have vivid memories of flipping through my parents’ art books as a kid.
When it came time for university, I initially followed my father’s path into engineering. But during graduate school at the University of Toronto, I found myself drawn back to visual work. I took on jobs as a graphic and web designer for university departments and served as the illustration editor for the student newspaper. By the time I graduated, it was obvious where my heart was—and luckily, I had built a portfolio to prove it.
UX was just emerging as a formal discipline then, and many graphic designers were transitioning over. After stints at a few startups, I joined Google in 2013, where I spent 12 years working on Google Translate and early AI efforts. Last year, I moved to Slack, where I now help design how people work alongside AI every day.
Looking back, Translate was about using technology to break down human-to-human communication barriers. Today, I’m designing for human-to-human and human-to-AI interaction. In both cases, the core challenge is identical: making something incredibly powerful feel natural.
Kursat: How has AI changed the way you actually work, day-to-day?
Pendar: It’s still evolving, but there are two big ways AI has changed how I work.
1. As a thinking partner
Early in a project, I use AI the same way I’d use a design sprint or a whiteboard session with a PM. It’s a rapid way to explore ideas, challenge assumptions, and stress-test directions before bringing them to the broader team.
One of my favorite AI skills is a custom version of “grill-me.” Instead of giving me answers, it aggressively questions the problem space and my proposed solution. Working through its prompts forces me to make dozens of micro-decisions I might have otherwise skipped, leaving me with a much sharper point of view.
2. As a rapid prototyping engine
I built a Claude Code-based prototyping kit that lets me spin up interactive prototypes in minutes. Everything lives in one GitHub repository and shares the same setup, so I can reuse components across projects instead of rebuilding them every time and create new prototypes simply by adding new folders. If I need a component or interaction from an older prototype, I can ask Claude to bring it into the new one.
I also added a lightweight commenting system that automatically converts feedback on a prototype into GitHub issues. It started because my prototypes were moving faster than my Figma files, and our content designers needed a better way to collaborate on the latest work. I’ve since shared the kit with the rest of the team, and quite a few designers and PMs are using it for their own projects now. Some have expanded on it and made it better, and now I’m pulling those improvements back into my own kit. I really like that kind of remix culture that’s emerging around AI coding.
The line between “designer” and “builder” has effectively collapsed for me. I can concept an idea, build the tooling behind it, and refine the execution without waiting on engineering capacity.
Kursat: Where has AI surprised you in all these new ways of working?
Pendar: The positive surprise: Watching AI solve genuinely hard, highly constrained problems out of nowhere.
For example, at Slack we have a strong culture of sharing designs through short video walkthroughs. I wanted a way to turn my prototypes into polished demo videos, so I asked Claude to build a tool that takes my natural language video descriptions, generates a Playwright script, and creates dynamic videos with pans and zooms that I can control. I gave it a constraint that it couldn’t require a separate login or auth key, and it somehow figured out how to reuse my existing authenticated terminal session even though it runs in a browser. I definitely didn’t expect that.
The disappointment: AI still struggles with design intent and precision.
Even if you give AI the actual design files, the prototype it generates often doesn’t quite match. It’s generally bad at one-shot generation. My prototyping kit has an agent loop that takes screenshots of its own work, compares them against the target, and keeps refining the implementation until it’s much closer. There are a few other tricks too, but getting AI to respect design quality is still one of the biggest challenges—and one of the biggest opportunities.
Kursat: When it comes to interaction design patterns or new paradigms — you’re defining a lot of new workflows at Slack — what are you seeing emerge? What’s working from the established patterns, and what isn’t?
Pendar: When LLMs blew up, everyone debated whether chat would replace traditional UIs or if we’d revert to task-specific interfaces. We were having those exact debates at Google Translate.
We’re still in the transition phase. Chat is clearly here to stay, but people are developing text fatigue. When we spend our lives consuming visual, micro-interactive feeds, asking someone to read three paragraphs of dense text to complete every task is a UX failure.
What excites me is the move beyond text responses to interfaces that are generated on the fly. Instead of returning read-only text and widgets, we need to make them visual and interactive.
I’m a little surprised we haven’t seen more of this from the frontier labs. It feels like one of the biggest opportunities for design and AI. In a few years, designers will no longer just design fixed interfaces. We’re increasingly going to be designing systems that can generate the right interface for a particular moment.
Kursat: What’s your perspective on treating AI as a teammate as a construct — and where do you see the boundary of that thinking?
Pendar: I already treat AI like a teammate in a lot of my work. I’ll brainstorm with it, ask it to challenge an idea, or hand it coding tasks. For me, that shift has already happened.
For broader adoption, I think two things need to happen. First, AI has to earn people’s trust. That means being transparent about what it’s doing and why. Second, it also has to become much better at doing more and taking reliable actions on our behalf.
There’s only so many summaries and daily briefings I need. What I really want is for AI to quietly take care of the repetitive parts of my life without me having to ask every time. The real opportunity is moving beyond prompt-and-response. AI should increasingly know what’s routine, step in when it’s helpful, and let me stay in control.
A good example is a monthly chore I currently do. I have a bill split system for our tenants, so every month I have to collect the bills for that property from different providers’ websites, calculate their share, update a spreadsheet, download the PDFs, and save them to a Dropbox. AI can already figure out the workflow. The hard part is the last mile—logging into multiple sites, dealing with authentication, and actually completing those actions reliably. That’s where I think today’s AI still falls short.
I don’t think the future is asking AI to do more things. I think it’s asking AI less often because it already knows what needs doing.
Kursat: Can you walk us through the agentic side — how you design these flows? Last time we met, you were showing a workflow you’d defined for an agent.
Pendar: One mental model I find useful is to think about automation in two buckets: on-demand and ongoing.
On-demand automation is when you hand an agent a task in the moment: research something, build a prototype, analyze a document, or write code. Prompting, Skills, and one-off agent interactions all fall into this category.
The second is ongoing automation. These are the repetitive things that happen on a schedule or when something triggers them: every morning, whenever an email arrives, or when a file changes. That’s where agents start feeling less like assistants and more like coworkers. They’re quietly doing work in the background.
This is actually the area I spend most of my time on at Slack. I work on both our node-based Workflow Builder and our prompt-based automation experiences, so I think about these tradeoffs every day.
Today there are two main ways of building automations. One is visual, node-based workflows, where you connect together different actions and integrations. The other is prompt-based, where you describe the workflow in natural language and give the agent access to the tools it needs.
People often frame those as competing approaches, but I don’t think they are. Some people naturally think in flowcharts, while others think in language. So I think both paradigms have their place.
The part I’m most excited about, though, is what comes next. Most people don’t wake up wanting to build workflows. They just want repetitive work to disappear. Over time, I think AI will observe how you work, notice patterns, suggest automations, and eventually build many of them for you. The interface for automation becomes less about constructing workflows and more about reviewing and approving what AI proposes.
Kursat: What advice would you give designers navigating the mix of excitement and unease around AI today?
Pendar: First, if you’re not already on a team that’s embraced AI, it can be hard to find the time to experiment. My advice is to carve out a little time anyway, and approach it with curiosity rather than pressure. Start by asking, “What’s something that’s annoying, repetitive, or time-consuming in my work or personal life?” Then see if AI can help you brainstorm and build a solution.
That’s how I’ve learned almost everything I know about AI. I didn’t set out to become an AI expert—I just kept building small tools to scratch my own itch: a prototyping kit, a better way to record demos. Many of those started as personal experiments and ended up being useful to my whole team.
Second, I’d encourage every designer to keep pursuing creative work outside their day job. Draw, write, make music, build furniture - whatever it is. I personally love illustrating and making children’s books. That unconstrained act of creating goes a long way in helping you develop your sense of taste, which is ultimately what will set you apart from not just other people but from AI as well.
Kursat: I couldn’t agree more about having creative side projects to develop your taste. Thank you so much, Pendar — it was great to have you, and as usual, it was full of insights.
And that’s a wrap for this issue. Until next time, take good care of yourself and your loved ones.
Pendar!