How startup designer uses Generative AI
Hello, I’m Hal. I am primarily in charge of product design at GMO Flatt Security, a startup in the cybersecurity domain. At Flatt, I’m involved in a broad range of tasks beyond design, such as development, product management, and customer support. In this article, I’d like to share some practical examples of how a designer working at a startup leverages generative AI.
*These use cases will continue to be updated in the future; I plan to add more whenever I come up with new ideas.
Introduction
One characteristic of a startup is that you have to focus on what’s valuable to the business—and to society—as you work with limited resources. You need to produce results in a short period, yet resources are often tight in every area: development, design, and even on the business side.
Of course, hiring more talent is one way to address this. However, even after you’ve decided to hire someone, it can take six months to a year or more before new members actually start contributing. In my own case, it took about eight months from my first interview to my actual start date. But in a startup, a few months can be extremely precious.
In such an environment, it becomes crucial to make full use of each individual’s skills and quickly enhance the value of the product or service. I always keep in mind the importance of “boosting my work speed and output quality,” and while that applies to more than just AI usage, in this article I will focus on concrete examples of how I utilize generative AI.
1. Understanding the Domain
Before joining Flatt, I spent over eight years designing so-called B2B products, including tax and accounting software for tax accountants. However, the specialized terms and mechanisms in cybersecurity were a completely different world for me. After I joined, new words and concepts popped up one after another, and at first, it was difficult to keep up—especially with all the English abbreviations.
That’s where generative AI—specifically ChatGPT—came in handy. Whenever I encountered an unfamiliar term, I’d ask something like “Could you explain what ___ is in a way a fifth-grader could understand?” That gave me a rough overview. From there, I’d read official documentation or our internal wiki, and ask questions to engineers to deepen my knowledge. By following these steps, the barrier to understanding new specialized terms dropped significantly.
Of course, the answers ChatGPT provides aren’t always 100% accurate. However, they’re extremely helpful as a first reference point for my initial understanding. I always make sure to confirm what I’ve learned with experts, but it’s a huge benefit to go from “I have no clue what this is” to “I can kind of imagine what this is” much faster.
For more details, check out the blog post linked below.
How to understand cybersecurity domain
2. Getting AI to Create a Rough Draft
In Japan, there’s a saying: “たたき台を作る人が一番えらい”, which means "the person who creates the initial rough draft is the real Most Valuable Person (MVP)". This is one of my personal mottos at work. Even if the quality is low at first, creating a rough draft makes it easier for others to give feedback, which then guides the process of refining it. In my previous job, I was mindful of how quickly I could share an initial rough draft with the team. I continue the same approach in my current company.
For example, if you need to write user stories, use cases, or a PRD, you might start off with no input at all—meaning you don’t even know what you don’t know. In that state, you can’t really plan out your research tasks.
In the past, I might have spent a week or two reading blogs and books to gain the necessary background knowledge. But now, working at a startup, there’s never just one project I need to learn about. And even if it’s only one project, I don’t have the luxury of taking one or two weeks just to gain domain knowledge.
So I have generative AI prepare the rough drafts for user personas and user stories. It can generate something that seems “about right” at lightning speed, which is incredibly helpful for getting an overall picture of what I’m working on. When combined with the domain-understanding method I described earlier, it helps me quickly grasp the overall scope of tasks.
Using the knowledge I’ve gained, I talk to our in-house domain experts and users, refine my understanding, and revise the initial user personas or user stories. If a project is small and the internal domain experts have a very clear picture of what needs to be done, I’ve sometimes been able to finish the entire UI design process in under half a day. By optimizing this portion of my time, I can fully devote myself to more complex issues that require deep thought, such as thorough user research—even in a fast-paced startup environment.
Creating rough drafts is also useful for proposals and project management. I might roughly outline an idea in ChatGPT, have it produce a plan, and in an instant, I get something that at least looks like a proposal. Of course, sometimes the result makes me think, “What is this?” But that emotion—“This is not what I meant!”—actually helps activate my brain to shape it into the ideal vision I have in mind.
In my case, I take the rough draft from generative AI and give it my own feedback, then have the AI revise it again, and check it myself once more. I do this loop a few times before sharing it with domain experts or engineers.
By rapidly cycling through these “personal draft loops,” I can prepare a reasonably high-quality idea even before gathering feedback from others.
3. Task Management
I used to manage tasks using Notion or Jira. But lately, I’ve changed my workflow by using Cursor for task management.
At the start of each week, I open Cursor and write down the goals I’d like to achieve. Then, I instruct Cursor accordingly—for example, “Propose a list of tasks for this week that align with these goals.”
The suggested tasks are, of course, rough, so I have to refine them myself to improve accuracy. But delegating “roughly creating a task list” to AI saves a lot of my mental bandwidth. I fix whatever’s off, and sometimes I have it create a day-by-day or hour-by-hour draft schedule, which I then finalize manually.
As a result, it’s become easier for me to see what I should be focusing on right now, and I can prioritize tasks more smoothly.
4. Engineering
Since joining a cybersecurity startup, I’ve been gaining more engineering knowledge on a daily basis. I’ve started helping with front-end UI implementation and managing our design system. However, because I never formally studied programming in depth, I used to rely heavily on our engineers whenever an error occurred—they’d fix it before I fully understood the root cause.
Now, though, I learn alongside generative AI. For instance, when I encounter an error where code isn’t working as intended, I’ll say, “Here’s what I’m trying to do with this code—why might I be getting an error?” Generative AI often gives a fairly relevant response. It’s not guaranteed to be entirely correct, but it helps me quickly pinpoint the potential cause of the error.
Beyond just getting things to work faster, I really value how much this approach helps me deepen my understanding. In conversations with engineers, I can now discuss not only design but also code, so our communication moves faster and we have fewer misunderstandings about specifications.
5. Building a Prototype for Implementation
In the past, even if I created an intricate interaction design in Figma, there were times when it seemed too difficult to implement given our resource constraints. Often, we would release a simpler functional version first and plan to implement the more advanced features later. But with the help of generative AI, this flow has changed a bit.
Specifically, before I even share my design mock with engineers, I’ll create a prototype (v0) of the interaction I want. If it’s so complex that I can’t easily create a quick v0, then maybe the UI design itself isn’t ideal. Conversely, if it looks complex but I’m able to quickly build a simple working prototype with v0, I can share both the code and the Figma UI with engineers, and we can proceed with development more efficiently.
Note: This overlaps somewhat with Section 4 (Engineering), but I wanted to highlight that even if you’re not personally involved in development, generative AI can still help streamline the development process.
6. Voice Input
Our development team communicates entirely in English. I’ve spent about three years working on global teams, so I can handle listening and speaking without any real issues for work. However, when it comes to writing chat messages or meeting notes, I still get bogged down by grammar concerns, which slows me down.
That’s where Superwhisper comes in. Initially, I started using it to make typing into Cursor easier. However, since it lets me edit the prompt freely, I thought, “Why not just have it output everything in English?” And it worked surprisingly well.
Prompt
Please translate the content I have spoken in Japanese into English before printing, if you cannot hear what I say, just print a blank. be simple, just print what you translated. which means you don’t have to say “Since I can hear the, here is the translated content:”
You could also use this to convert casual expressions into more business-friendly language, or register variables as shortcuts. It’s easy to imagine many possible uses.
7. Writing Blog Posts
This very article is more than 80% drafted by o1-pro. If I were to publish it as-is, I think it’s already at a certain level of quality. But because I’ve only written a rough prompt, the text still has a bit of an “AI flavor” to it. That said, even with a rough prompt, about 80% of the text it generates can be used directly. With a little editing, I can publish an article. It’s a huge help because I can finish one article in around 30 minutes.
Even if you’re new to generative AI, getting it to draft a blog post is one of the easiest ways to experience immediate benefits in your work.
Conclusion
For a designer working at a startup, “How can we learn quickly, shape ideas fast, and maximize value with limited resources?” is a major challenge. By leveraging generative AI, I’ve been able to gain the following benefits:
- More efficient domain understanding
- Faster creation of initial drafts
- More efficient task management
- Deeper participation in engineering
- Streamlined development processes
- Greater productivity through voice input
- Increased volume of blog posts
Generative AI has become a reliable partner for enabling each person to work at maximum speed and creativity. Please consider testing these ideas in your own work—and I’d love to hear about your experiences if you have any interesting use cases!
Thank you for reading to the end. If you’re interested in boosting cybersecurity together or in product design enhanced by generative AI, feel free to reach out. I look forward to connecting over a cup of coffee!