PROMPT HAIL MARY

One quiet day in April, Donggu and I were sitting around a table at OTP, chatting as we usually do. As always, the conversation drifted toward AI. Our recurring topic at the time was when, exactly, AI might wipe us out. Donggu, who studies AI in graduate school, would arrive each morning with news of the latest features and services appearing at an astonishing pace, like a newspaper boy calling, “Extra, extra!”

After talking about it for a while, we began to wonder how people outside our studio were using AI. What were they handing over to it? How much were they explaining? And how were they judging what came back? Everyone was probably finding their own way, but there weren’t many opportunities to look at that process together. We were all busy enough keeping up with the next new thing. Donggu happened to be vibe coding an imaginary piece of software for communicating with Rocky from Project Hail Mary, the film we had debated so enthusiastically the week before. What if we brought our methods and uncertainties into the open and tried collaborating with this unfamiliar presence called AI? That was how we came to plan OTP’s first AI workshop, Prompt Hail Mary.

Donggu also made the wonderful workshop poster. He told us it took its inspiration from the instructions accompanying the Golden Record. Just as humanity sent the sounds of Earth into space in 1977, we wanted to leave a record of how we were approaching AI at this moment—our methods and our attitudes. We were curious about what we might think when we looked back on it.

On the day, eight people from Seoul and Berlin gathered at the OTP office. There were brand designers, a product designer, a frontend developer, and people working with data and business. Everyone did different work and had a different level of familiarity with AI. Some wondered what they could do beyond generating images and reviewing text. Some found writing detailed prompts exhausting. Others wanted to see how someone else developed an idea.

As always, the first thing that mattered was deciding together what to make and why. We formed four pairs, each bringing someone more familiar with AI together with a newcomer. We thought that different interests and ways of thinking might lead to experiments neither person would have imagined alone. We called the person using AI to build things “Grace,” the person keeping an eye on purpose and direction “Houston,” and the AI helping solve the problem “Rocky.” What to attempt with Rocky—both a tool and a friend—was something the two people had to decide together.

After Donggu’s introduction to AI, particularly large language models, and Gyuyoung’s talk, “Design Is Not a Tree,” we began working in teams. We started with three questions: What does good design mean to us? What problem could we address through that understanding? And how would we collaborate with AI? We decided to take our ideas beyond conversation and turn them into something we could actually try.

Yuri and Gyuyoung wanted to practise interpreting and defining problems in their everyday work. They built “ShadowBox Training,” a tool that gives AI feedback after someone reads a scenario and writes out their definition of the problem. As they tried it themselves, they refined it to carry insights from previous exercises into the next task and offer more varied situations in which to practise their judgment.

Yubin and Hanseul, meanwhile, were curious about how to find that subtle difference that makes a brand its own. They extracted keywords from client conversations, connected words that resonated or clashed, and developed them into metaphors and personas that could describe the brand. Unpacking a process they usually approached intuitively opened up room to explore new combinations with AI.

Joheun and Donggu focused on the order in which ideas are presented to someone else. How could they explain a brand’s character and design direction so that the reasoning behind a proposal came through clearly? They built a “Brand Pitch Workflow” prototype that used project information and branding methodologies to structure a presentation. They gave Claude a brief, then reviewed its response as brand information, suggested methodologies, and a slide outline.

Gyuwon and Chaewon thought about people who had plenty of repetitive work but no idea where to begin with automation. In environments with layers of older systems and established routines, it can be difficult even to see what could change—or to believe you could change it yourself. They built a tracker for recording daily tasks and the time spent on them, then exploring opportunities for automation. They paid particular attention to making sure the recording itself would not become another burden. It also mattered to distinguish work that required human judgment from work someone could attempt with help.

Although everyone started with the same questions, each team headed somewhere different: defining problems, finding a brand’s distinctiveness, making an idea persuasive, or helping someone take a small first step toward change. Seeing what each team had made and hearing their stories gave us a clearer sense of what they had meant by “good design” at the beginning.

The workshop we had planned for five hours stretched beyond ten. (We had a very good time.) When we shared our reflections at the end, the most common response was, “There was less to be afraid of than I thought.” Trying something alongside another person had changed how they approached what had felt vague or daunting on their own. With so much changing so quickly, it seemed to have given us a chance to pause and try things for ourselves. Over the lovely wine Gyuyoung had brought us and Chinese food from Yasanghae, an OTP favorite, the conversation continued late into the night.

Looking back, the ability to work with technology, the perspective needed to define a problem, and the way we made decisions together were closely connected throughout the day. Deciding what to make and adjusting the roles of people and AI were themselves part of the design process. What stays with me is how the same tools led to such different questions and outcomes.

The conversations we had and the questions we left open are a letter to our future selves. Like the Golden Record, we do not know when an answer might arrive. But I wonder what we will be making, and who we will be working with, when we next open this record. I suspect we will return to this old question then, too.

“Technology is the answer, but what was the question?” — Cedric Price