WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=L3RuP_q8Bwc NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/L3RuP_q8Bwc.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:00.960 --> 00:00:01.720 Hi. s2 00:00:01.720 --> 00:00:02.880 My name is Kat. s3 00:00:02.880 --> 00:00:10.560 I'm a senior design and developer advocate at Progress Software, and I have spent most of my career focused on the front end. s4 00:00:10.560 --> 00:00:19.040 After taking a somewhat meandering path through design and into development, I eventually found my niche in the space where those two worlds overlap. s5 00:00:19.040 --> 00:00:28.400 My design background has given me a lot of insight into the UX side of software engineering, and the user research that I get to do as part of my work as developer advocate s6 00:00:28.400 --> 00:00:31.000 is one of my favorite parts of the world. s7 00:00:31.000 --> 00:00:39.000 I have spent a lot of my career teaching developers about design, about UX, and about how to build more user-friendly software. s8 00:00:39.000 --> 00:00:45.800 Something that's always been valuable, but is more crucial than ever now that AI has entered the picture. s9 00:00:45.800 --> 00:00:49.960 I don't need to tell you that AI is a pretty unique new technology. s10 00:00:49.960 --> 00:00:59.760 It is rapidly developing and has so much potential to improve systems and software in ways that we couldn't have even dreamed of just a few years ago. s11 00:00:59.760 --> 00:01:04.440 However, along with all of that potential is also quite a bit of risk. s12 00:01:04.440 --> 00:01:14.880 AI's non-deterministic nature means that the output can wildly range in quality, format, style, and accuracy, even when we're trying our best to prevent that. s13 00:01:14.880 --> 00:01:19.960 On top of that, it requires entirely new types of interaction patterns to use. s14 00:01:19.960 --> 00:01:27.240 In fact, we've seen the emergence of new vocabulary, techniques, and sometimes entire roles meant to address that need. s15 00:01:27.240 --> 00:01:39.680 For those of us that are living and breathing this technology every day, concepts like prompt engineers, hallucinations, retrieval augmented generation, and so much more have all become regular parts of our lives. s16 00:01:39.680 --> 00:01:45.360 But for the vast majority of our users, that simply isn't the case yet. s17 00:01:45.360 --> 00:01:53.520 In any kind of technology, there's a gap between the developer, who has by nature kind of become the subject matter expert, and the user. s18 00:01:53.520 --> 00:02:00.520 The technical literacy of the average software engineer is always going to be higher than the average user of said software. s19 00:02:00.520 --> 00:02:02.280 That's where UX comes in. s20 00:02:02.280 --> 00:02:10.080 It's how we bridge the gap between what we intended when we built an application and what the user actually experiences. s21 00:02:10.080 --> 00:02:18.840 After all, it doesn't actually matter what our software can do if our users hate using it so much that they will avoid it at all costs. s22 00:02:18.840 --> 00:02:23.280 And right now, AI has a kind of serious UX problem. s23 00:02:23.280 --> 00:02:34.160 Our users are seeing AI features get added to basically everything right now, from their work software to their social media apps, right down to their computer and phone operating systems. s24 00:02:34.160 --> 00:02:43.520 But, as a wise technologist once said, "Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should." s25 00:02:43.520 --> 00:02:51.240 We're really pushing our users to adopt these new tools and features, but we're not always making it easy to do so. s26 00:02:51.240 --> 00:02:59.959 It's common to see situations where a user is presented with a poorly designed AI feature that they really only have the barest idea how to use. s27 00:02:59.959 --> 00:03:09.519 Then, when they try it and it doesn't do what they wanted or expected it to do, they feel annoyed and off-put and kind of negative about AI generally. s28 00:03:09.519 --> 00:03:17.000 The more times they try an AI feature and get sub-par results, the less likely they are to engage with it again in the future. s29 00:03:17.000 --> 00:03:28.800 Which means that, as the creators of these features and software, we really only have so many chances to get this right before our users just disengage and stop trying at all. s30 00:03:28.800 --> 00:03:37.840 In my personal opinion, this knowledge gap between developer and user when it comes to AI is one of the widest that I've seen with any technology. s31 00:03:37.840 --> 00:03:46.800 That makes it extremely hard for us to build AI-powered features that our users are actually going to get the full benefit from. s32 00:03:47.040 --> 00:03:56.480 When Macintosh introduced their graphical user interface, it was, like AI is now, something that was very different and unfamiliar to their users. s33 00:03:56.480 --> 00:04:05.000 But a large part of their success and reputation was built on their ability to create a user experience that met users where they were. s34 00:04:05.000 --> 00:04:08.840 They borrowed vocabulary and user flows from real life. s35 00:04:08.840 --> 00:04:17.880 They included lots of icons and visual representations, and they really made sure to walk users through how to get the most out of their new personal computers. s36 00:04:17.880 --> 00:04:26.160 And over time, as the average user became more comfortable and more literate in the space, the user experience of the Macintosh evolved with them. s37 00:04:26.160 --> 00:04:30.320 We start to see more technical terms and fewer abstractions. s38 00:04:30.320 --> 00:04:41.600 Look at the difference here between the control panels in Macintosh system 1 and system 6. If you had presented a system 1 user with the words rate of insertion point blinking or RAM cache, s39 00:04:41.600 --> 00:04:44.520 it would have been pretty much meaningless to them. s40 00:04:44.520 --> 00:04:52.720 Similarly, we see that those low and high volume icons are gone in system 6, along with the little turtle and rabbit speed icons, right? s41 00:04:52.720 --> 00:04:57.240 That kind of more literal depiction wasn't needed anymore. s42 00:04:57.240 --> 00:05:04.120 Right now, we are probably somewhere around system 3 in this metaphor of introducing AI to users. s43 00:05:04.120 --> 00:05:13.280 We may not need the turtle and rabbit icons anymore, but we also can't yet assume that they will sit down with our AI software as experts. s44 00:05:13.280 --> 00:05:19.880 The good news here is that, unlike the Macintosh UI, we're not totally starting from scratch here. s45 00:05:19.880 --> 00:05:26.160 Our users have existing mental models about software usage that's going to carry over to what we're building. s46 00:05:26.160 --> 00:05:30.320 They're just not always going to map over exactly one to one. s47 00:05:30.320 --> 00:05:32.160 But that's where we come in. s48 00:05:32.160 --> 00:05:42.120 As developers, it's our role to leverage these patterns that our users are already familiar with, start to combine them in new ways, and then introduce them to our users gradually s49 00:05:42.120 --> 00:05:45.160 in a way that's not going to feel overwhelming. s50 00:05:45.160 --> 00:05:52.520 A good user experience will not only provide a user with the tools, but also guide them how to use them. s51 00:05:52.520 --> 00:05:56.040 For instance, let's look at an AI chat interface. s52 00:05:56.040 --> 00:06:04.440 From a purely UI perspective, chatting with an LLM is almost, but not quite, like chatting with another user. s53 00:06:04.440 --> 00:06:15.520 So, we'll be able to borrow some familiar patterns here, like having the user's chats show up on one side and the AI chats on the other, messages that appear above an input field with a send button, s54 00:06:15.520 --> 00:06:18.400 a scrollable message history, and more. s55 00:06:18.400 --> 00:06:26.400 That gives our users a really good jumping off point and a lot of visual clues about how to start interacting with an AI chat. s56 00:06:26.400 --> 00:06:35.200 However, the experience isn't quite close enough that we can just lift an existing interface from direct messages and repurpose it wholesale. s57 00:06:35.200 --> 00:06:46.680 A chat interface that was originally built for two people is not going to have the accommodations that we need for things like adding reference sources, pausing or stopping an in-progress reply from the LLM, s58 00:06:46.680 --> 00:06:50.160 leveraging agentic tools and integrations, or more. s59 00:06:50.160 --> 00:06:56.440 So, those are going to be the places where we have to adapt existing UX and UI patterns into something new. s60 00:06:56.440 --> 00:07:03.040 We have to be the bridge and help our users transition into familiarity with these new AI experiences. s61 00:07:03.040 --> 00:07:04.840 We can see the difference here, right? s62 00:07:04.840 --> 00:07:09.760 Compared this Copilot research agent new chat with the new Teams chat. s63 00:07:09.760 --> 00:07:11.800 Both of them have familiar aspects, right? s64 00:07:11.800 --> 00:07:16.480 We see text boxes, we see the plus to attach files, and so on. s65 00:07:16.480 --> 00:07:20.320 But, the Teams chat assumes a much higher level of familiarity. s66 00:07:20.320 --> 00:07:31.600 It's not spelling out any of the iconography it uses, and we don't see some of the things intended to lower the barrier of entry, such as the examples, large text buttons, or voice interaction options s67 00:07:31.600 --> 00:07:34.600 that show up on the agent chat. s68 00:07:34.880 --> 00:07:42.320 Now, before we really dive into this, I do think we have to take a moment to address kind of the elephant in the room, right? s69 00:07:42.320 --> 00:07:46.040 Why not just have AI create these solutions for us? s70 00:07:46.040 --> 00:07:54.200 Why do we have to be the ones who design and build new AI patterns when we have technology that can generate interfaces for us now? s71 00:07:54.200 --> 00:07:59.960 The problem with that is that AI can only really remix things that already exist. s72 00:07:59.960 --> 00:08:10.120 It's fantastic for looking at those existing common patterns and replicating them, but the patterns for AI interfaces don't fully exist yet. s73 00:08:10.120 --> 00:08:18.520 Or at the very least, they are still in the process of being rapidly developed, and they are changing all the time as that technology advances. s74 00:08:18.520 --> 00:08:26.360 There's just not a long history of standardized and familiar AI-related interfaces that we can reference to solve this problem. s75 00:08:26.360 --> 00:08:32.360 So, at least for now, we're kind of on our own here, and we can't yet defer it to AI. s76 00:08:32.360 --> 00:08:38.800 Maybe that will change down the road, but for the time being, this is still a very human problem. s77 00:08:38.800 --> 00:08:48.120 Additionally, as many folks have begun to notice, the design that an AI creates is going to be a lot like every other design out there. s78 00:08:48.120 --> 00:08:54.000 AI can reference and remix, but you're just not going to get brand new concepts from it. s79 00:08:54.000 --> 00:08:58.520 An AI-generated UI tends to look pretty darn average. s80 00:08:58.520 --> 00:09:04.360 And while that can be a really great starting point, it's not usually a great ending point. s81 00:09:04.360 --> 00:09:07.880 If you just need a quick landing page, it's probably going to do the trick. s82 00:09:07.880 --> 00:09:21.760 But since we're dealing with whole new interaction modes that need more attention paid to the UX rather than less, it's just not going to be enough to get us where we need to go right now at the quality level that our users need and deserve. s83 00:09:21.760 --> 00:09:30.800 So, if we're going to build new patterns and new interfaces to help our users get the most value out of this technology, where do we start? s84 00:09:30.800 --> 00:09:34.680 Well, like most things in design, it's best to start with the user. s85 00:09:34.680 --> 00:09:37.760 Specifically, the user's problems. s86 00:09:37.760 --> 00:09:49.600 That means that when we're thinking about building our AI features, we need to be thinking about what challenges our users have with AI and the user flows that will mitigate those issues as much as possible. s87 00:09:49.600 --> 00:09:58.680 From the user research I've been doing and the users I've had the chance to speak to, I've started to group the main challenges here into five categories. s88 00:09:58.680 --> 00:10:05.520 Trust, clarity, control, transparency, and meaningful benefit. s89 00:10:05.520 --> 00:10:18.120 In this talk, we're going to discuss each of these in depth, understanding the problems our users have when they're using AI, and looking at some examples of new patterns and techniques we can employ to address them. s90 00:10:18.120 --> 00:10:32.240 By making sure that we are addressing each of these, we can start to create AI experiences that support our users through the introduction of this new technology and ensure that they are in a place to use it to its fullest potential. s91 00:10:32.240 --> 00:10:34.702 So, let's get going. s92 00:10:34.702 --> 00:10:34.720 [snorts] s93 00:10:34.720 --> 00:10:36.720 Trust. s94 00:10:36.720 --> 00:10:46.560 The number one, by far, biggest hurdle that we need to overcome in order for our users to engage with the AI features we build is trust. s95 00:10:46.560 --> 00:10:50.480 Right now, to the average user, AI is a black box. s96 00:10:50.480 --> 00:10:55.280 Many simply do not understand from a technical perspective how it works. s97 00:10:55.280 --> 00:11:02.200 And when you don't understand how something works, it makes it very, very hard to trust the output. s98 00:11:02.200 --> 00:11:09.720 On top of this, just about everyone who has interacted with an LLM before has, at some point, seen them hallucinate. s99 00:11:09.720 --> 00:11:18.600 Even as the models keep getting better, we're still not at the point where we can claim any tool will be 100% guaranteed hallucination-free. s100 00:11:18.600 --> 00:11:24.600 There's always a chance, even if it's a small one, that the AI is going to provide an incorrect response. s101 00:11:24.600 --> 00:11:30.040 So, how many times can a user see output that's wrong and still trust it? s102 00:11:30.040 --> 00:11:37.760 It can be easy to think that the answer to this problem is to try and position the feature that we're launching as some kind of exception to this rule. s103 00:11:37.760 --> 00:11:42.120 We say things like, "Other AI tools may not be trustworthy, but ours is different. s104 00:11:42.120 --> 00:11:45.200 Ours is higher quality, it's safer, it's more reliable." s105 00:11:45.200 --> 00:11:47.160 So on and so forth. s106 00:11:47.160 --> 00:11:50.440 But not only is this questionably true, right? s107 00:11:50.440 --> 00:11:54.520 I mean, after all, and not many of us are really training our own models here. s108 00:11:54.520 --> 00:11:58.240 So a lot of this is going to be simply outside our control. s109 00:11:58.240 --> 00:12:02.840 But it's also a very difficult thing to try and sell your users on. s110 00:12:02.840 --> 00:12:09.040 In situations where we cannot promise the truth, we have to go above and beyond to earn it. s111 00:12:09.040 --> 00:12:20.040 Our honesty about current capabilities and limitations of our AI tools will go a lot further with our users than the denial of any potential problems. s112 00:12:20.520 --> 00:12:32.880 So rather than trying to convince users that our solution is inherently trustworthy, we can instead build in patterns that empower them to see when incorrect information is returned and give the tools to correct or mitigate it. s113 00:12:32.880 --> 00:12:35.280 You know the saying, "Trust, but verify"? s114 00:12:35.280 --> 00:12:37.080 That's kind of the goal here. s115 00:12:37.080 --> 00:12:48.080 By citing the references in an AI-generated response and linking users back directly to the source material, we give users the information they need to validate AI output themselves. s116 00:12:48.080 --> 00:12:58.560 The more a user is able to click through and see where an answer came from, the more they'll be able to trust the content, even if they don't choose to check every source every time. s117 00:12:58.560 --> 00:13:04.640 In addition to allowing our users to vet the answers, citations also have a second important purpose. s118 00:13:04.640 --> 00:13:10.600 They allow users to repurpose the material in their own work while maintaining a trail of accuracy. s119 00:13:10.600 --> 00:13:14.480 This is really important for their own credibility and reputation. s120 00:13:14.480 --> 00:13:23.600 After all, how often would you share content from an unverifiable source if you knew that any errors would ultimately be attached back to your own name? s121 00:13:23.600 --> 00:13:30.000 Generating output for the user might be the last step in our process, but it's really just the beginning for them. s122 00:13:30.000 --> 00:13:35.920 Users want to take that content and turn it into a report, an email, a campaign, a presentation. s123 00:13:35.920 --> 00:13:41.720 And if the output can't be cited and trusted, then it's just not going to be used. s124 00:13:41.720 --> 00:13:50.200 There are a handful of different ways we can implement this, and which one you pick is going to depend on what you're building and the context in which it's being used. s125 00:13:50.200 --> 00:13:56.880 But a few popular approaches include things like tooltips, inline links, and side panel reference windows. s126 00:13:56.880 --> 00:14:03.240 Tooltips allow you to provide a small snippet of a relevant quote, which can be really great for increasing credibility. s127 00:14:03.240 --> 00:14:12.760 Links, of course, are ideal if the content is coming from an external source, like another webpage, as opposed to an internal source, like a document in a shared drive. s128 00:14:12.760 --> 00:14:22.240 And if the feature you're building is meant to support more research-oriented work, then you might consider adding a side panel, where the user can explore the source material in more detail, s129 00:14:22.240 --> 00:14:29.560 which kind of positions the AI assistant as more of a librarian than inherently a subject matter expert. s130 00:14:29.800 --> 00:14:38.120 The other place where trust factors heavily into AI work is agentic workflows, where the AI is thinking and executing work on its own. s131 00:14:38.120 --> 00:14:47.640 This, understandably, can be kind of concerning for users, depending on the stakes of the project and how easy or hard it would be to roll back any incorrect actions, s132 00:14:47.640 --> 00:14:52.040 which is when keeping that human in the loop really becomes crucial. s133 00:14:52.040 --> 00:15:02.080 One of the best ways to help users trust these systems enough to use them is to show an action plan to the user and allow them to approve it before the agent begins work. s134 00:15:02.080 --> 00:15:05.480 This has become a pretty common flow in a lot of foundation models, right? s135 00:15:05.480 --> 00:15:16.080 If you're using agentic features in Claude or ChatGPT, you will have seen it create a list of the steps it will take to accomplish a given task and then ask for your confirmation before beginning. s136 00:15:16.080 --> 00:15:27.360 Of course, there are also settings you can add to toggle this off or always allow it, all of which is important if you're going to allow users to repeat a flow over and over and don't want them to have to baby sit it. s137 00:15:27.360 --> 00:15:33.240 Spoiler alert, we will talk a little bit more about permissions in the transparency section later. s138 00:15:33.240 --> 00:15:48.080 But if you're creating some kind of an agentic tool and no plan is ever shown to the user and an action happens without them understanding why or how, it can be very, very hard for them to trust both the result of that action s139 00:15:48.080 --> 00:15:51.400 and the agentic tool itself. s140 00:15:52.720 --> 00:16:02.680 There is also, undeniably and I would maybe even argue correctly, quite a bit of skepticism around whether or not content is AI-generated right now. s141 00:16:02.680 --> 00:16:11.440 You've probably seen exchanges online where someone shares a photo or video and another person is immediately commenting to tell them that's not actually real. s142 00:16:11.440 --> 00:16:21.960 When we spend so much of our time and energy second-guessing and investigating content that's shared with us, it doesn't exactly foster an environment of trust. s143 00:16:21.960 --> 00:16:26.600 For now, AI-generated content is highly polarizing with users. s144 00:16:26.600 --> 00:16:34.720 While some have heavily leaned into using it in their daily lives, others will react to it negatively and dismiss it as slop, right? s145 00:16:34.720 --> 00:16:42.800 The ability for AI to generate text, images, and video that is nearly indistinguishable from human-created work is very new. s146 00:16:42.800 --> 00:16:47.200 And for many users, that still feels uncanny and unsettling. s147 00:16:47.200 --> 00:16:49.640 I don't say this to start any kind of debate, right? s148 00:16:49.640 --> 00:16:57.960 I mean, seems harsh to say it, but our own personal opinions on AI-generated content don't really matter too much in this context. s149 00:16:57.960 --> 00:17:08.400 What's more important is the awareness that our users will react to AI-generated content in a wide variety of ways and not all of those ways are going to be positive. s150 00:17:08.400 --> 00:17:18.760 So, if we want to include AI-generated content in our applications and we're not exactly sure how our users will feel about it, what can we do to retain their trust? s151 00:17:18.760 --> 00:17:25.240 One of the easiest things is to just mark any AI-generated content as AI-generated. s152 00:17:25.240 --> 00:17:38.600 This could be as simple as a disclaimer in the App Store description, a watermark uh on a photo, whether that's digital or visible, or maybe even just an asterisk at the end of a sentence that denotes something as AI-created. s153 00:17:38.600 --> 00:17:50.760 When we do this, we remove that potential user perception of them feeling like we tried to sneak AI-generated content into our apps or somehow like pull one over on them. s154 00:17:50.760 --> 00:18:01.400 If we're using AI and we think it's the right fit for our work, or we feel strongly that our users should be able to share AI-generated content, then there's really no shame at all in designating s155 00:18:01.400 --> 00:18:03.240 where it's been used. s156 00:18:03.240 --> 00:18:16.000 Depending on where that AI is used, it might also be helpful to clue our users into which aspects of output they might need to manually verify, rather than having to review everything as though it was AI-generated, s157 00:18:16.000 --> 00:18:19.040 if that's not truly necessary. s158 00:18:20.080 --> 00:18:23.280 Hand in hand with trust comes clarity. s159 00:18:23.280 --> 00:18:32.960 Clarity is one of the most important aspects of creating a good user experience kind of generally, and the stakes only get higher when we start incorporating AI. s160 00:18:32.960 --> 00:18:42.760 Because, as we said, so much of AI still feels like an unknown to our users, we need to be as descriptive and straightforward about what's happening at any given time. s161 00:18:42.760 --> 00:18:52.280 The more insights we can give them about what our software is doing and what they can expect to happen next, the more comfortable they will feel engaging with AI. s162 00:18:52.280 --> 00:18:58.200 I've noticed that there can be a kind of impulse to frame AI to our users as magic, right? s163 00:18:58.200 --> 00:19:05.320 We don't even have to look further than the prevalence of the sparkle icon to designate AI as an example of that. s164 00:19:05.320 --> 00:19:11.120 However, while that might feel kind of mysterious and cool, it's really not accurate, right? s165 00:19:11.120 --> 00:19:16.560 AI is just another technology, and our users deserve clarity over dramatics. s166 00:19:16.560 --> 00:19:26.440 They want to know what's going on, how a given output was created, and there's really no reason for us not to pull back that curtain whenever we can. s167 00:19:26.800 --> 00:19:32.720 One of the most obvious ways for us to do this is with streaming text or similar live generation of content. s168 00:19:32.720 --> 00:19:37.800 Not only is this a really great way to disguise the latency issue of AI generation, right? s169 00:19:37.800 --> 00:19:45.280 Nobody likes to watch a loading spinner, but it's also a fantastic way to show the user what's happening in real time. s170 00:19:45.280 --> 00:19:53.120 If we make the user wait until an entire response is generated, there's a non-zero chance that it won't actually be what they wanted. s171 00:19:53.120 --> 00:20:00.560 However, if we start showing a partial response as it's being generated, we give users the chance to start assessing it immediately. s172 00:20:00.560 --> 00:20:09.280 That means they can be coming up with follow-up questions or adjustments or maybe just stopping the process entirely if it's not returning what they expected. s173 00:20:09.280 --> 00:20:19.680 This also keeps the user as an active participant in the process, preventing that like XKCD my code's compiling experience of pressing a button and walking away. s174 00:20:19.680 --> 00:20:26.560 You've probably experienced yourself how much harder it is to complete a task if you are constantly disengaging from the workflow. s175 00:20:26.560 --> 00:20:35.320 Even if you know a process will only take a minute, if there's not something to keep you there, you're going to start checking your emails or replying to Slack messages s176 00:20:35.320 --> 00:20:45.560 or scrolling on your phone and just like that, a 1-minute wait has become a 5-to-10-minute side quest, making it much harder to pick up where you left off. s177 00:20:45.560 --> 00:20:52.280 The more we can prevent this for our users, the more productive they can be when they're using our tools. s178 00:20:52.520 --> 00:21:03.680 Another thing that can help remove the wall between the AI and the user is having it think out loud as much as possible as it processes a request, especially in chat-based interactions. s179 00:21:03.680 --> 00:21:14.160 Similarly to live streaming, this improves the user's confidence level in the output cuz they can kind of follow the breadcrumbs and have a deeper understanding of how a given conclusion was reached. s180 00:21:14.160 --> 00:21:18.280 After all, we do this all the time when we talk to other people, right? s181 00:21:18.280 --> 00:21:26.800 If we're presenting an idea, we know it's going to be helpful for us to kind of talk through the rationale behind how we got there and why we think it's the right choice. s182 00:21:26.800 --> 00:21:32.560 Explaining ourselves is such a natural part of human interaction that we've really come to inspect it. s183 00:21:32.560 --> 00:21:37.040 And when we don't see it mirrored in the systems that we use, it can feel kind of frustrating. s184 00:21:37.040 --> 00:21:41.400 Like something is being unintentionally kept from us. s185 00:21:41.400 --> 00:21:48.800 Additionally, seeing that chain of thought process allows for better, more accurate revisions when the output wasn't what they expected. s186 00:21:48.800 --> 00:21:56.720 If the user can understand all the steps that happened between point A and F, then they can really pinpoint where things went off the rails. s187 00:21:56.720 --> 00:22:03.480 And that allows them to give more specific instructions or clarify ambiguous parts of their request. s188 00:22:04.560 --> 00:22:11.120 Some existing patterns are a little bit easier to adjust from their previous non-AI usage into this context. s189 00:22:11.120 --> 00:22:13.720 And highlighting new content is one of those. s190 00:22:13.720 --> 00:22:26.560 It's not groundbreaking, but when it's combined with everything else, it can be immensely helpful in helping users keep track of AI actions, especially in situations where an agent is acting autonomously. s191 00:22:26.560 --> 00:22:35.760 The basic idea is that when new content appears or something gets changed or updated, it gets specially marked for the user in a way that draws their attention to it. s192 00:22:35.760 --> 00:22:40.920 Sometimes that might mean literally moving the focus up or down the page to see what's changed, right? s193 00:22:40.920 --> 00:22:47.200 Like when new messages appear in an ongoing chat and the message history scrolls down to display it. s194 00:22:47.200 --> 00:22:53.360 Other times though, it could just mean drawing the user's eye back to a space they might not have been actively watching. s195 00:22:53.360 --> 00:22:56.720 Even if that doesn't include literal movement on the page. s196 00:22:56.720 --> 00:23:05.560 Things like changing the color of revised text or putting a box or highlight around new material or even marking lines of code that are different. s197 00:23:06.280 --> 00:23:15.360 AI tools are often framed as being a kind of personal assistant for the user, an intelligent junior co-worker that they can delegate tasks to. s198 00:23:15.360 --> 00:23:24.400 However, as anyone who's ever delegated a task can already tell you, sometimes that means that things don't get done in the way that you would have done them. s199 00:23:24.400 --> 00:23:31.320 When users are working with our AI tools, they need to know that they are still ultimately the ones in the driver's seat. s200 00:23:31.320 --> 00:23:41.320 After all, there's very little value in being able to assign the AI tasks if you can't then step in as you like to make adjustments, corrections, and fixes. s201 00:23:41.840 --> 00:23:52.120 In that vein, it is also important for the user to be able to stop or override an AI action at any point, whether that's during content generation, partway through an agentic workflow, s202 00:23:52.120 --> 00:23:56.000 when it's running code or scripts, or something else entirely. s203 00:23:56.000 --> 00:24:04.120 If the user is unable to abort the process, then they're not actually the one in control, and frankly, that's not really acceptable. s204 00:24:04.120 --> 00:24:13.960 One of the most reassuring things we can offer nervous users is a clear and prominent emergency brake that they can slam to bring everything to a halt for any reason. s205 00:24:13.960 --> 00:24:21.920 That means it can't be tucked away in a menu somewhere or involving some specific command they have to remember, which they won't in a high-stress moment. s206 00:24:21.920 --> 00:24:29.200 We want to give them the equivalent of a big red button they can slam to make everything stop whenever they need. s207 00:24:29.560 --> 00:24:34.680 Of course, once a user has stopped a process, we can probably guess what they want to do next, right? s208 00:24:34.680 --> 00:24:37.960 They want to undo the incorrect things that changed. s209 00:24:37.960 --> 00:24:46.440 Even way before AI, Nielsen's heuristics list safe exploration as one of the core principles required for a good user experience. s210 00:24:46.440 --> 00:24:56.840 That means that users should be able to navigate back and forth, click and unclick things, and generally just kind of mess around in a piece of software without getting stuck or unintentionally s211 00:24:56.840 --> 00:24:59.480 making big permanent changes. s212 00:24:59.480 --> 00:25:07.000 Sometimes a user will try something and then change their mind, decide they don't like it, or it didn't do what they thought it would. s213 00:25:07.000 --> 00:25:13.720 In those cases, we want to make it as easy as possible for them to roll things back to an earlier state. s214 00:25:13.720 --> 00:25:24.400 This kind of safety net was always valuable, but now that we're dealing with non-deterministic output, having some kind of a version history is going to be pretty much a non-negotiable. s215 00:25:24.400 --> 00:25:29.560 Of course, the level of version history that's required will differ based on the task. s216 00:25:29.560 --> 00:25:37.720 The Git style of version control that we are used to as developers is probably more than you will need in a user-facing application. s217 00:25:37.720 --> 00:25:42.160 If this is a truly simple conversational interface, you may not need it at all. s218 00:25:42.160 --> 00:25:47.520 The user could just rephrase their question and try again if they don't get the answer they wanted. s219 00:25:47.520 --> 00:25:58.480 However, if your tool is doing something more complex, like creating and editing a document, simple undo and redo actions that allow users to move back and forth within the last s220 00:25:58.480 --> 00:26:01.920 10 or so steps would be really helpful. s221 00:26:01.920 --> 00:26:09.480 And if you're wanting them to do truly advanced work with your tools, then it's worth building in the mechanisms for checkpoints or save states. s222 00:26:09.480 --> 00:26:21.040 Consider what kind of tasks your user is going to be working on with your AI features, and think about how significantly each step forward is likely to change the existing content. s223 00:26:21.040 --> 00:26:26.600 Ideally, this will also allow users the specificity to make targeted targeted adjustments. s224 00:26:26.600 --> 00:26:33.920 You don't necessarily want them to have to wipe out and replace everything that happened in a given step if they weren't happy with it. s225 00:26:33.920 --> 00:26:38.520 It's common for AI-generated results to have a mix of quality in the output. s226 00:26:38.520 --> 00:26:42.200 Users may want to keep some aspects while reverting others. s227 00:26:42.200 --> 00:26:49.880 And the more control we can give them over that creative process, the more enjoyable it will be for them to collaborate with our AI tools. s228 00:26:49.880 --> 00:26:54.600 This is especially true now as the cost of tokens is starting to tick up. s229 00:26:54.600 --> 00:27:04.320 Having granular control over what gets reworked allows for more specific and production productive iteration without them just having to say, "Try again." s230 00:27:04.320 --> 00:27:08.160 and start over from scratch each time. s231 00:27:08.720 --> 00:27:14.600 AI tools are strongest when they seamlessly integrate into the rest of a user's workflow. s232 00:27:14.600 --> 00:27:22.360 Referencing internal documents, sending emails, making calendar appointments, drafting and sharing content, so on and so forth. s233 00:27:22.360 --> 00:27:27.320 However, most users won't integrate what they can't see and don't understand. s234 00:27:27.320 --> 00:27:40.280 If the AI applications we build are going to become part of their daily lives, then users need transparency into exactly what they do, what they have access to, and what it will cost them in both time and money. s235 00:27:40.280 --> 00:27:47.320 The more transparent we can make these features, the less hesitation our users will have in adopting them. s236 00:27:47.600 --> 00:27:58.480 The concept of asking for user permissions is certainly not AI specific, but the stakes can feel a lot higher when you're asking users for permission to allow AI to act on their behalf. s237 00:27:58.480 --> 00:28:10.040 That means we need to make it easy for users to see which applications we've allowed our tool access to and in what ways, keeping in mind that permissions are often more than just a binary yes or no. s238 00:28:10.040 --> 00:28:13.800 So, can your agent just reference data from this database? s239 00:28:13.800 --> 00:28:16.200 Or is it also allowed to delete tables? s240 00:28:16.200 --> 00:28:20.920 Can only read your user's emails, or can it send from their addresses well? s241 00:28:20.920 --> 00:28:23.760 Can it run scripts, search their local files? s242 00:28:23.760 --> 00:28:28.280 Each one of these actions requires direct user sign-off. s243 00:28:28.280 --> 00:28:31.360 Permissions are also not a one-and-done situation. s244 00:28:31.360 --> 00:28:39.640 A user might feel comfortable allowing an action to happen once under their direct supervision, but they might not want to allow it permanently. s245 00:28:39.640 --> 00:28:43.320 They could give access to a specific folder, but not every folder. s246 00:28:43.320 --> 00:28:47.640 They might approve an action, but then want to be notified every time it's taken. s247 00:28:47.640 --> 00:28:56.360 Consider which of these gray zones might exist in your permission structure and try to accommodate as many different options as you can. s248 00:28:56.600 --> 00:29:00.360 Another common sticking point with AI is data collection. s249 00:29:00.360 --> 00:29:10.080 Often, it can be helpful to save information from past interactions, but storing this kind of data requires real attention to user permissions and data management. s250 00:29:10.080 --> 00:29:18.480 If you want to create a system that can remember things, then you also need to make sure that it can forget and that the user can not only see, s251 00:29:18.480 --> 00:29:22.520 but has the final say in what exactly gets remembered. s252 00:29:22.520 --> 00:29:32.160 An ideal permissions flow will not only ask for a user's approval in the moment, but also create a space where they can see the history of what they've given access to and when, s253 00:29:32.160 --> 00:29:37.560 as well as allowing them to revoke that access or permission at any time. s254 00:29:38.200 --> 00:29:49.960 Now, this one's pretty simple, so we won't spend too much time on it, but another crucial aspect of transparency is making sure users know exactly what they are committing to when they approve an action request. s255 00:29:49.960 --> 00:30:02.040 In addition to approving those actual steps and plan, as we already discussed in the trust section, they also need to know how much time something is going to take and what it's going to cost them in money or tokens or credits, s256 00:30:02.040 --> 00:30:04.480 however you are counting this. s257 00:30:04.480 --> 00:30:18.400 Even if we can't specify an exact amount, if we can provide a rough estimate, that's generally going to give users enough information to work with and to potentially revise their request if it's going to exceed what they're comfortable with. s258 00:30:19.160 --> 00:30:31.360 Similarly to the marking that denotes which content was AI-generated, it's also a really good example uh of an chance to have a visual signal for when an AI tool is acting independently. s259 00:30:31.400 --> 00:30:44.000 If, for example, you're going to allow an agent to take control of the user's browser, then there needs to be some kind of banner or sidebar or outline, an indication that the user is no longer driving that interaction. s260 00:30:44.000 --> 00:30:53.320 Not only is this a good thing to do just for transparency, so the user understands the current state of the system, but it's also helpful to make sure they don't unintentionally s261 00:30:53.320 --> 00:30:57.160 interrupt or confuse an ongoing process. s262 00:30:57.160 --> 00:31:05.720 It's an especially important consideration for processes that you know are going to take an extended period of time where the user might have stepped away and then come back s263 00:31:05.720 --> 00:31:10.840 without necessarily keeping track of everything going on while they were gone. s264 00:31:11.920 --> 00:31:22.280 Finally, all of the interesting technology and cool functionality in the world is meaningless if what we build isn't solving the problems that our users need solved. s265 00:31:22.280 --> 00:31:34.240 To do that, we need to create experiences that make it as simple as possible for our users to get the output they need from our AI features and start to leverage it in their work or daily lives. s266 00:31:34.680 --> 00:31:41.640 One of the biggest mistakes we can make when developing AI features is assuming that our users know how to use them. s267 00:31:41.640 --> 00:31:44.120 As mentioned at the beginning of this talk, right? s268 00:31:44.120 --> 00:31:54.960 We know how to do things like provide context or specify formatting requirements or break large tasks into smaller ones and then iterate, but our users often don't. s269 00:31:54.960 --> 00:32:06.160 So, when we place a blank text box in front of a user and just tell them to ask AI, we're actually kind of asking them to do a lot of work in figuring out how to really use it. s270 00:32:06.160 --> 00:32:15.800 They need to understand what kinds of problems the tool can solve, how specific they need to be, what information the AI tool needs to succeed, if they need to add sources or references, s271 00:32:15.800 --> 00:32:19.440 how to recognize when a response needs refinement. s272 00:32:19.440 --> 00:32:32.960 Rather than expecting this level of AI literacy from our users on day one, we can help them by providing examples, templates, suggested prompts, and guided workflows that demonstrate what success looks like. s273 00:32:33.840 --> 00:32:39.040 Now, what will your users be doing with the content that your AI feature generates? s274 00:32:39.040 --> 00:32:43.960 If they're going to run a search or analyze a spreadsheet, what happens next? s275 00:32:43.960 --> 00:32:45.920 What do they do with the results? s276 00:32:45.920 --> 00:32:49.000 If they create an image, who are they showing it to? s277 00:32:49.000 --> 00:32:50.440 How are they sharing it? s278 00:32:50.440 --> 00:32:55.480 Is it getting printed or posted online or sent to a friend? s279 00:32:55.480 --> 00:33:04.040 By introducing next step action buttons, we can make it as easy as possible for them to leverage the output of our tools in their own work. s280 00:33:04.040 --> 00:33:10.880 If they're not able to action on the content that was created, it's never going to be more than a novelty to them. s281 00:33:10.880 --> 00:33:13.640 It's not going to be of realistic use. s282 00:33:13.640 --> 00:33:20.080 But we can help guide them to suggested next actions that will help them make the most of what they created. s283 00:33:20.080 --> 00:33:26.000 If we want to take this one step further, we can start to build in direct integrations with their other most used tools. s284 00:33:26.000 --> 00:33:32.440 Like, maybe they want to generate a new document with the content in their word processing software from a draft. s285 00:33:32.440 --> 00:33:35.000 Maybe they want to push new code to a linked repo. s286 00:33:35.000 --> 00:33:39.920 Maybe they want to open a new ticket in a tracking system with their findings from a test. s287 00:33:39.920 --> 00:33:49.480 The simpler we make it for them to get the information from our tool into the rest of their workflow, the more useful it's going to be to them. s288 00:33:50.160 --> 00:33:59.080 Now, AI can generate content, write code, analyze data, automate tasks, empower our users to do all kinds of amazing work. s289 00:33:59.080 --> 00:34:05.120 But only if they're willing to try it, and only if we make it easy enough for them to do. s290 00:34:05.120 --> 00:34:17.360 When everything our AI tools are doing is hidden behind a curtain, it makes users feel like things are happening without their input, which is hard, especially when many of them are already kind of feeling some level of AI skepticism s291 00:34:17.360 --> 00:34:19.320 or hesitation. s292 00:34:19.320 --> 00:34:32.560 If we unintentionally create AI experiences that take away our users' power, understanding, and autonomy, then they'll never be interested in using what we build, because they'll never feel truly comfortable engaging with it. s293 00:34:32.560 --> 00:34:47.440 By focusing on patterns that reinforce those pillars of trust, clarity, control, transparency, and meaningful benefit, we can set guardrails around our AI features that will help make our users feel safe and confident. s294 00:34:47.440 --> 00:34:50.399 Cuz the technology is here, right? s295 00:34:50.399 --> 00:34:56.760 The models are already really good, and they just keep getting better and better, faster and more efficient. s296 00:34:56.760 --> 00:35:02.840 That means that the differentiator for the AI-powered software we build isn't performance anymore. s297 00:35:02.840 --> 00:35:07.440 It's the quality of the experiences that we can build around them. s298 00:35:07.440 --> 00:35:12.040 The line between design and development is blurring a little bit more every day. s299 00:35:12.040 --> 00:35:23.600 How users interact with the system, how much information they see, how much control they have, and how we earn their trust are now questions that developers have to consider when building AI-powered software, s300 00:35:23.600 --> 00:35:27.160 whether your title includes designer or not. s301 00:35:27.160 --> 00:35:36.800 The technology may be incredible, but for it to be truly successful, users still need to be at the center of everything we build. s302 00:35:37.200 --> 00:35:40.080 Thank you guys very much for listening to this talk. s303 00:35:40.080 --> 00:35:46.760 I really hope this is helpful to you as you start integrating AI features into your own applications and software. s304 00:35:46.760 --> 00:35:51.440 You can find these slides, as well as a full transcript of this talk, at the following links. s305 00:35:51.440 --> 00:35:55.760 And, of course, feel free to reach out to me online if you have any questions at all. s306 00:35:55.760 --> 00:35:59.160 I'm always happy to chat.