How AI Reshapes the UX Design Pace

Kristen Liu at UXDX Community: AI Speed, Real World Design. Video: https://youtu.be/Lpi6gFDHmq0

Readable transcript: edited from the recording's captions for readability (fillers and false starts removed, punctuation and section headings added). Wording is the speaker's own. Timestamps are positions in the video. Names marked [?] could not be verified against the audio.

Introduction: will AI replace us?

[00:00:07] Host: That brings us to our second session of this evening. Speaking will be Kristen Liu, who's a UX designer at Amazon, and Kristen will be talking about how AI reshapes the UX design pace. I'd like to welcome Kristen to the stage. Over to you, Kristen.

[00:00:25] Kristen: Thank you. Thank you for having me, and great discussion and great content, Evan [?], because first of all my question to this community, to this audience, would be: do you think AI will replace us? Thank you for joining at this time, everyone, especially if you are coming from the East Coast. It's freezing outside, but let's talk about the hot topic of AI.

[00:00:53] I've been working in UX design for a couple of years, and last year I had a project starting from scratch. It was a complex enterprise project. I worked with the PM and engineer; we started from brainstorming together, all the way to presenting a live prototype to our customer, within eight days. That project made me think, oh, maybe it's time to change. Maybe it's time to discuss this with our community. So I kept talking with my internal colleagues: what do you think AI can do for us? Do you have that kind of concern? I also joined a couple of different communities to try to get this answer. So here I am today. I wanted to talk with a broader audience: what do you think AI can bring to us, and how can AI reshape the UI/UX design pace? That's the title of our content today.

[00:02:04] A little bit of the agenda for today. The first question, just like you mentioned: what do you think about AI? Is AI going to replace us? Next, we want to talk about how the expectations have already changed from all the stakeholders, what the old workflow is that we have, and whether your old workflow fits the new expectations. So we want to cover a little bit about the old workflow, what the new workflow is, and where AI can help us. And in the last part I wanted to talk a little bit about speed versus quality, and also leave some time for you if you have questions.

[00:02:52] First and foremost, the question: what do you think about AI? We are in this age of AI. Everybody is talking about AI. Do you think AI is going to replace us? We had this discussion within our team. We also hosted different workshops to try to get this answer, trying to hear from other designers within my company. We heard two different voices. Some people said, and it's kind of negative, yes, AI is going to replace us. Why they think that, I'm going to explain a little bit. But some of my colleagues say no, AI is never going to replace us, just like Evan thought, because we still need a human in the loop for all of the designs, even when AI is trying to help us. So what do you think? You can leave your questions or your thoughts for our community, and we can talk about them later at the end.

Expectations have already changed

[00:04:08] No matter whether you think AI is going to replace us or help us, the expectation, if you've observed, has already changed. The expectation is coming from the stakeholders, and by stakeholders I mean the PMs and the engineers. The expectation is coming from all the companies. You can observe all these big tech companies. They want you to make the design quick, quick, quick, fast, and deliver all your artifacts fast, fast, fast.

[00:04:48] And if you observe, before that we had the different roles. We have PMs, we have designers, we have engineers, and everybody was working on their own artifacts. The PM is drafting their documents, UX is delivering their design artifacts, and the engineer is coding. But now you can observe each different role wearing different hats. The PM is also drafting some mockups, and the UX designers also need to understand requirements, since not every team has a PM. And the engineers can just use some of the AI tools to draft some mockups.

[00:05:36] So there is a question: can everybody be a designer using some AI tools? For example, we have a lot of AI tools that can generate some of the visuals, like Figma Make. I'm not sure about your company, but I've observed some PMs, after drafting a document, trying to have some visuals to help other stakeholders understand what the requirements are, to visualize it, even to have some mockups. So the expectation has already changed, no matter if AI is going to replace us. What do you think: is AI going to replace us or not?

The old, siloed workflow

[00:06:21] That means the old collaboration workflow is changing. As I mentioned, in the past the collaboration between PM, UX, engineers and stakeholders was siloed. We could spend weeks or months, based on the size of your project, on each of the steps. The PM could draft the requirements, like product requirements or business requirements, for weeks or months. After that we'd review with our customer's representative, and also internally with all our stakeholders. After we review, decide and align that yes, that's the feature and that's the product we're going to build, then we'll move to the UX design phase.

[00:07:09] We will go through the whole process of our design. We'll do the user research, then draft the low-fi design and iterate it based on all the feedback, all the way to hi-fi design. After we usability test even the hi-fi design, then we will hand over to our engineering team, and the engineering team will maybe also spend months or weeks, again based on the size of the project, to launch the final product or feature. As I mentioned, this is siloed for almost each of the steps. PM, UX and engineering could spend weeks or months on each of the steps, which makes this collaboration workflow take months or weeks to deliver a project.

[00:08:07] The design process that we have is based on the double diamond design process that everybody knows as a designer. When we reach our phase, we will discover the problem, define the problem, then design, and also iterate based on the customer feedback and the stakeholders' feedback, all the way to delivering our designs. This is the traditional design process we use on a daily basis.

The new, collaborative workflow

[00:08:38] But in the AI age, based on the project my team and I were working on together, it has completely changed, and the expectation of the stakeholders has also changed. How it changed, I'm going to use a real case example to show. It changed from siloed phases between PM, UX and engineering to a more deeply collaborative workflow. The PM, UX and engineer could work very closely from the very beginning, all the way to the delivery of the project. At the very beginning we might not even have requirements. It's completely starting from scratch: brainstorming together, drafting the requirements together, brainstorming about the design together, and then delivering to our engineers.

[00:09:38] I'm going to use a real example that we had last year, in the last quarter. We got requirements from our customer. It's an internal project, an internal enterprise project, and usually a project like this takes three months, because it's an enterprise project and very complex. You will spend maybe weeks learning all the terminology. For example, what is GRC, governance, risk and control? What is account reconciliation? All of these accounting terminologies and enterprise terminologies.

[00:10:23] After that we'd go through the traditional process: the PM drafting requirements and reviewing with the stakeholders and customers, and then handing over to design. It depends on what we have from the PM. Do we have personas and user research out of this, to decide whether we want to go through the full circle of research into our design phase, or whether we start from the design phase? What I wanted to say is it would usually take a while to deliver a project; this project specifically would take three months.

[00:11:02] But for that project we were competing with another party, which means we didn't have that time. The requirement from our leadership, the expectation from our stakeholders, was: can we have a proof of concept of the product in one week, so that we can win the pitch against the third-party tool? So my PM, engineers and I thought together about how we could make this happen with these AI tools.

[00:11:37] What we did is we got together on a Friday afternoon and dragged our customer into a conference room, and we talked with our customer to gather the feedback. After that we spent some time, because my engineering team and I were completely new to that project. We spent our entire afternoon understanding all the terminology and all the domain knowledge of that enterprise project.

[00:12:08] Then each day in the following week, what we did is in the morning we would brainstorm together, using some AI tools to draft the requirements and iterate the requirements together with our PM. Of course, the PM is the lead on the document drafting. After that, at noon, I was using some AI tools to explore the design ideas, and after we had some mockups and ideations we'd sync with our customer in the afternoon and then hand over to our engineering team. They also used some AI tools for coding and developing. Then we had daily syncs with our customer to do the iteration every single day, and on the following Friday we presented to our customer's leadership, and finally we won the pitch against the third-party tool.

[00:13:10] So what we did, and where we used AI: our PM used AI to draft the requirements. My PM also used Figma Make, not just drafting the requirements but also having some visualization, because it's an enterprise and complex tool, to try to articulate what the tool in his mind is, to tell the engineer and to tell me what he thought the tool should look like. Of course, it's not the final design; that's not the purpose of it. It's just to help us understand, because we had a very tight timeline.

[00:13:58] After that I would do my exploration using Figma Make, and then I'd sync with my customer and hand over to coding. My engineering team also used some AI tools to pull all the designs, all the design mockups, through an AI tool into the code. At the end, when we pitched and presented to our customer's leadership, it was a live prototype. So we finished a project within eight days, starting from scratch, that would usually be a three-month project.

Where AI helps: discovery and research

[00:14:41] So where did AI help us, and where can AI help us, in my experience, based on the design process that we have? I used AI in discovery and research. I recently used AI for one of my usability tests with my customer, using AI to draft the user testing plan. Then, after conducting the user research, I'd throw the user research meeting notes and minutes into AI to help me draft a summary and also the usability study and usability testing report. After that I would review it with the stakeholders. This was a healthy increase in productivity and efficiency.

[00:15:35] In the discovery and research phase, AI can help us with interview synthesis, survey analysis, pattern detection and competitive reviews. But it still needs us to ask the right questions. Even if we have AI to help us draft the usability testing plan, what are the right questions for us to ask? And also the context and emotion: when we conduct the usability testing, we need to observe our customers. That's not something AI can help us with from the text perspective.

Where AI helps: design, testing and iteration

[00:16:13] Next, when we move to the design phase, where can AI help us? I asked this question to some of my colleagues: do you use AI, or do you think AI can help you design something? I got a whole bunch of different responses. Some said, oh, I don't think AI right now is mature enough for me to just generate some design and then use it directly, especially when we have a design system and the AI tool is not integrated with our own design system. So it's hard to use all the designs that AI generates for us.

[00:16:56] But one interesting response I got from my colleague was: yes, I think AI can help us, as long as I can provide the right prompts to AI, and give long enough prompts to AI. But the thing is, I need more time to give the AI correct prompts than I need to design it myself. So maybe right now AI is not mature enough for me to use, but maybe in the future AI can help us. So I got a whole range of different responses from my colleagues.

[00:17:36] What AI can help me with for now, at least, is I can use some AI tools to do some exploration for the early stage of the design. But as I mentioned, most of the designs are more generic. Especially for complex enterprise products, they're not very specific to your requirements and domain.

[00:18:06] After the design phase, the next phase is testing and iteration. During this phase, how AI can help us: I actually use AI to do heuristic evaluation, which is just throwing some mockups at AI and saying, okay, can you give me some feedback and an audit for the screen? It's actually pretty good. It can also give me faster feedback cycles. But I still need to take all the feedback I have and draft the heuristic evaluation docs for my project.

Speed versus quality

[00:18:47] The last part I wanted to talk about is speed versus quality. Remember the question we had: do you think everyone can be a designer with AI tools? Especially with some of the internal tools, my colleagues, the PMs, are already using AI to generate some of the design without consultation, and they just use it and build the product. Some teams generate some mockups and use their designer as a consulting team to do some consultation. But I think customer-facing products will be better; they still need a UX designer to have a great product and UX.

[00:19:35] But still, faster shipping does not equal better UX. As I mentioned, as of now at least, when AI tools generate designs, they cannot customize them to the very specific requirements that we have. For example, with one of the dashboards I asked AI to give me, no matter whether it's an entertainment product dashboard or an accounting product dashboard, it looks the same. And the logic behind each of the UX decisions is lost from the AI tool. So what we can say right now is that AI is there to accelerate our thinking, not to replace it.

[00:20:28] All right, that's about it from me. If you have any questions, just leave your question. Thank you.

Q&A

[00:20:36] Host: Excellent. Thank you very much, Kristen. It was really interesting. I love it: it's augmenting our intelligence, not replacing us. So it's a nice positive bit to end on, rather than the doom and gloom that sometimes can come up. I'd love to know your thoughts, because in that example, I thought it was great how you showed the compression. It would usually take three months, and then you were given eight days, or one week as you said. Having gone through that, do you think it's better to go with that really accelerated timeline but then give yourself another little bit of time to iterate and improve? And do you think that's a quicker and better way of working versus doing more of that traditional upfront research and work? What's your opinion after having gone through it?

[00:21:29] Kristen: That's a great question. In that situation we had to come up with something, because we were pitching our idea to our customer so that they could decide to go either with us or with the third-party tool. So we had to come up with something. In some cases we face a similar situation: we have to come up with something for the design, starting from scratch with the PM or engineer, to deliver something to our customer to earn the trust. But in the traditional way, after we've pitched our product to our customer and won the pitch, we need to come back later to see if this design is the best design we have. Maybe we'll go through some of the process that we have in the traditional way, and also listen to our customers and do the more detailed design. But AI at least gives us an opportunity to quickly come up with something in a very short time. That's what I want to say.

[00:22:44] Host: Excellent. So it supports it. But I'm going to put you on the spot now. If you were given the choice on a project, did you enjoy that accelerated timeline, or are you more of the belief that it's a higher-quality output if you put that work in?

[00:23:02] Kristen: I would say it really depends on the project, and also, like I mentioned, the expectation of the stakeholders is changing right now. I talked with someone on the team, and they're saying, okay, UX is kind of slow, because the engineering team right now is using AI tools right after the requirements, and then I can build something right after and show it to my customer. I cannot wait for months for you to come up with some mockups. So I guess we don't have a lot of choice. This is a new age for us. But as I mentioned, it really depends. For some more generic projects, maybe we can use that. But for a complex one, we'd probably have to go back after we show the concept idea and go into the iteration part to learn more from our customer.

[00:24:01] Host: Brilliant. I'd also like to touch on something you mentioned at one point. It's this challenge with experienced people: like you said, one said, okay, I don't think it's quite ready, and another person said it takes me longer to get the output I want with prompting and going back and forth versus just doing it myself.

[00:24:21] Kristen: Yeah.

[00:24:23] Host: It is a challenge with the tools. I'm looking more on the dev side than the design side, but I've noticed that shift, and I've noticed a lot of people on the dev side saying it was around December that a new model was released, and that was the shift for them. I've noticed that myself, where it's actually a lot quicker for the AI to do it than me. Do you think that point is coming soon for design, or do you think we're still quite a bit away from it?

[00:24:56] Kristen: That's a great question. That's actually why I'm here; I wanted to hear everybody's thoughts. As I mentioned, I talked with my internal team, and we have different thoughts. Some may be very negative, and some are very positive, thinking that even with AI we still need humans in the loop for the design process. What I think, as of now (I cannot say for the future), is that the AI tools are not mature enough, especially if your requirements are very complex. The logic behind the product is not there yet. Maybe they can come up with some beautiful mockups using some general design systems, but the logic behind the screen and the information architecture of the product are not there yet. So we'll see.

[00:26:07] Host: I think it was in Evan's presentation where he mentioned Magic Patterns. I don't know if you've experimented with it. I haven't tested it, but they market a feature where you can load in your components, your system, and it will use that. Have you been experimenting with those types of things?

[00:26:27] Kristen: Yes, we tried some. Just like the response from my colleagues, we spent a lot of time trying to integrate our own design system with that AI tool. We also have some internal tools from our company; we're not allowed to use external ones. But it takes quite a lot of time and effort as of now. So, just like the response from my colleagues, I'd rather do it myself, because giving the AI prompts takes me more effort and much longer.

[00:27:07] Host: So it's great, as you mentioned, for really rapidly trying different ideas and rapid prototyping, to help you shape your thoughts, but then it's ultimately up to you to go and make it better aligned with your product, more coherent, all those kinds of items.

[00:27:29] Kristen: Yeah.

[00:27:31] Host: Brilliant. Well, I think that leads us to time. I really enjoyed both of the talks today. It's kind of scary how quickly this field is moving, but it's great to see how different people are approaching it and adopting it. So I just want to thank you, Kristen.