Loading summary
Transcript56 lines
- [00:00]
A
You've got this whole cohort of eager young upstart Y Combinator startups who are absolutely ready to snap at our heels and disrupt us.
- [00:11]
B
I remember where I was sitting the first time I shipped a Polish pr. It feels good.
- [00:15]
A
The start of an S curve is a great time to be a designer because there is so much to be figured out. We don't know how any of this is going to be played out. Everyone on the team has taken a goal to ship code to production in Q2 in this current quarter.
- [00:30]
B
You have to start imagining, and I think that's like the level of agency that you're talking about here. It's like, what can this be? How can I be early on this?
- [00:38]
A
It is definitely the willingness to roll up your sleeves and experiment and not talk or think about it, but actually just give it a go and do lots of like disposable little experiments.
- [00:48]
B
Welcome to Dive Club. My name is Rid and this is where designers never stop learning. This week's episode is with Emmett Connelly, who's the VP of design at Intercom, which you probably know as the customer success platform. But what you might not know is that over last couple of years they've incubated a new product called Fin, which has now overtaken the core business. So this episode is an inside look at the journey of designing a new AI product and all of the ways that Emmett is pulling the design Org at Intercom into the future and encouraging them to own more and more of the front end. So let's start off by hearing the origin story of Fin.
- [01:30]
A
Intercom is a 10 year old, thousand person company, more than 10 years old. And then like for a lot of companies, you know, I guess the short and pithy version is chatgpt happened. Or put more accurately, I suppose this moment arrived in the tech industry when it became slowly and then suddenly and quickly apparent to people at different speeds that this was going to signify a real big change. When ChatGPT arrived, like it was kind of like a dawning realization over the course of a weekend. A lot of people talking to each other on Slack, going, holy shit, have seen this, what do you think it means? And really it did not take long for us to understand that this was going to be a big deal for us. I guess in part it's because we've been trying to make the bot conversational bot thing happen for several years, you know, and we had been working with machine learning, creating bots, and we'd actually had a partnership with OpenAI, which I guess helped us even before ChatGPT. That meant that we could very quickly, like, take the API that they made available. And I think within three to four weeks of ChatGPT coming out, we released a bunch of AI features that were like, when you're in the Intercom inbox, you can rephrase or make longer or shorter or more formal or whatever. Your response, very basic stuff, but it helped us kind of get our beaks wet and realize what was possible. We then partnered with OpenAI and on the day that they shipped GPT4, that was the day that the first version of Fin launched. Fin being Intercom's customer service agent. Right. So I guess since then, until now, it has been a story of both kind of us internally and the world at large as well, slowly realizing more and more and more, yes, this is actually important. This is actually the new thing. And I guess for us at Intercom has slowly turned into this thing where we've realized that Fin is bigger than Intercom. And so there's been this kind of remarkable story of the last couple of years of Fin being incubated within Intercom and I guess what we might see in other places, but I haven't seen a whole bunch of yet, which is a true AI, native AI first kind of new startup and new product emerging out of a, you know, legacy SaaS business. And I think that's something we're pretty proud of, actually. At Intercom, a lot of companies have been, I think, quietly trying to figure out what this new wave means for them. We have been trying and, you know, lots of false starts and mistakes and so on, but I think we're finally at the point now where we can confidently say that we have products that we really think is compelling and that we really think is like a prime contender in a really hot space. And I mean, that's the last thing I'll say is, like, there's a certain degree of fortune and circumstance to this, and I guess Fortune favors the people that are there waiting for it to arrive. And we've gotten a lot of value for more than 10 years of working on that problem within Intercom. But then you're gifted, like mana from heaven almost, this new technology that gives you so many new ways of tackling the problem and just the opportunity to be at the cusp and the cutting edge as well, of figuring out what this kind of new era of software development is. So it's been a, you know, exciting, intense, fun, challenging period of figuring out a lot of things at the product level. The company level, the strategy level, the design level, or all sorts of things. Yeah, it's been a blast.
- [04:48]
B
Real quick message and then we can jump back into it. So I've been doing a ton of demos for Inflight. I mean, Sometimes I'll have four 30 minute calls in a row and honestly, I have no idea how I would do it without Granola. It's like Apple Notes, but it transcribes information each meeting for me and then enhances my notes afterwards. That way it's easy to get little summaries or pull out quotes for the team. I can even share a link in Slack or export it directly to Notion. But what's really crazy is I can then set a context window to ask AI questions about those demo conversations. Like I can tell it to create a bulleted list of all of the feature ideas that people have mentioned to me over the last 10 or so interviews. It's crazy. And I firmly believe every single designer should be using Granola for all product conversations. They're offering a big time discount for Dive Club listeners. If you head to Dive Club Granola, you and people on your team can get three months for free. Remember Mengtu? He had one of the most popular episodes on building with AI. Well, he recently tweeted that Lovable is insane and called it a designer's dream. And he's right. I mean, I've tried every tool out there and Lovable is by far my favorite. There's simply no easier way to bring your designs to life in code. Just describe your idea in your own words and watch it transform into a fully functional app. And then iteration is easy because they have a visual edit feature so you can make any granular changes and get the finer details just right. If you haven't used Lovable yet, I promise it's going to blow your mind. Just head to Dive Club Lovable to start building today. That's L O V A B L E. Now on to the episode. Honestly, I kind of just have to tip my cap to what you all are doing because this type of paradigm shift is the exact opportunity that other startups needed to disrupt the incumbent that is Intercom. That's kind of had the, the primary grip on this space for so long and yet you all had the courage to proactively disrupt yourselves. It's cool to see.
- [06:46]
A
We've spent a lot of time over the years at Intercom, you know, writing and speaking and stuff like that about product and strategy and how to run a company and so on. And I guess in that sense of Maybe fancied ourselves as students of strategy here. And then suddenly you're right, the moment comes along, you're like, oh shit, we are the ones about to be disrupted here. Oh yeah, disruption theory to test and dust off your Clayton Christensen and so on. It's put us in a very interesting strategic position because, you know, I'll take the compliment, but we aren't the biggest incumbent in the space either. There are much larger companies, especially, you know, that serve larger markets than us. And at the same time, yes, because customer service is like such a no brainer for AI to completely change the game. You've got this whole cohort of eager young upstart Y Combinator startups who are absolutely ready to snap at our heels and disrupt us. And so yeah, it was that disruptive, redisrupted mentality that I think we had to adopt.
- [07:52]
B
All right, so let's go a little bit deeper then because something that you said to me earlier was you talked about how you spent basically two years steering this tanker towards an AI first company. What were some of the other design challenges that you had to figure out during that journey?
- [08:06]
A
I guess the primary one was making Fin work, right? And figuring out what the dynamics of actually making that system work were. When we shipped the first version of Fin, we kind of out of the box. The numbers are approximate, but roughly right here out of the box, we were getting an average of a 24% resolution rate. That is every time Fin was saying, I think I know the answer to this and trying to do it, it was getting it right that amount of times, right. That was a great start. For some teams. That is like a game changing kind of number already. That is basically what you get when you point your docs to Fin and let it slurp them up and learn from them. There's so much more you can do to train and onboard and educate Fin on how your team works, what they know, give it the capabilities to take actions in other places, right? So there's all of this kind of unhobbling that you have to do in order to give it the ability to drive that resolution rate up and up. Figuring out what those things are. And the answer is it's a combination of lots of different things. But figuring out what those things are has been a key part of the challenge. And really over the last two years, I think our average resolution rate has gone from 24% to 60% today. That's again the average kind of out of the box. So train Fin and on average are getting that some Customers are getting way higher. What we kind of thought was maybe theoretical a couple of years ago has been proved to be a reality. So the core design challenges all along will be how do we help as many people as possible get up to that level? What you realize you have to actually build is a whole application for training Fin. You think about Fin and you think, oh, it's a simple little agent that has some back and forth chats with you and a relatively simple surface area. But in the background is everything that requires you to train Fin with everything that's known about your company, everything. Anyone who's great on your team would know about everything about your product to analyze what Fin doesn't know, then train it, then to test those changes and see if it actually works, then to deploy. And you kind of have this iterative circle. So we had to kind of design this, analyze, train, test deployment workflow in the background that actually goes into training Fin. Making that useful and accessible to as many people as we can has been kind of the background design challenge and one of the big Fin design challenges.
- [10:30]
B
Any specific learnings from that training process and running through those iterations that you think other teams could benefit from?
- [10:36]
A
One of the things that we have that we are still trying to figure out is how much control or access is it appropriate to give the user, right. If you think about Fin potentially having like system prompts where you can write a prompt that Fin always understands, we call it guidance in Fin and tasks that Fin can carry out, one of the things that we've realized in building this action taking capability and trying to make it available to customers is it's amazing, it's super powerful, but sometimes you can shoot yourself in the foot with it. So with great power comes great responsibility and you can actually tank your own resolution rate if you do the wrong thing with an open system prompt. So it's kind of a dial you can turn up, you know what I mean? You can just give Fin access to a load of static information. That's easy. It'll do a great job of understanding it and inferring it and cross referencing from loads of different places to weave together an excellent answer. But it's still stuck at this level of informational answers. Next, it needs to be able to know, hey, Rid, you have a specific question about your account and Finn needs to be able to go and look up your account, the status of it, the level, the tier of membership that you're on, how many projects you've created, whatever, and come back and tell you specific things to you, oh, you've exceeded your limit of accounts created, so you need to upgrade. Do you want me to help you do that? And that last part, then the action taking part is want me to do that for you is where FIN can actually go and complete actions with every stage of capability unlock there. You need to give FIN access to more data and more capabilities. And therein lies the challenge that gets more and more complicated and more and more fraught or dangerous. And so for some of the most, let's say close to the metal stuff that feels again closest like system prompt level control. We're still working directly with our customers to figure out what's safe to give them. We want to give them that access, but we haven't yet been able to make it fully self serve. An interesting thing I'm interested to figure out in the next six months or whatever it is, is, is that an innate property of AI systems where ultimately you give people more and more control up to the point where they're controlling the model almost directly themselves and what they want it to do versus you're putting enough guardrails in it where people can't do like make terrible mistakes or so on. Where does that stop? Will we be able to make this super advanced, almost programmatic control of AI of FIN fully self serve or will you have self serve up to a point under some kind of professional services thing that you engage with Intercom's team directly and the team behind FIN to like implement for you? And obviously you're talking like a bigger volume or something like that, but I think we're seeing some of that with how our customers are adopting agents already and not just us. You know, it's a fun little spectrum.
- [13:35]
B
That I'm now only seeing for the first time in my mind where most people like myself included, you know, I'm living at the system prompt level. And then I'm thinking about, okay, how can I tweak this in order to create a user experience that makes sense. And yet you almost have to think about what's the right interface to allow a customer to tweak their own system prompt because obviously you're not going to expose the entire thing. But if you go too far into, well, we'll create a UI for everything, then maybe that's too rigid and you're not giving enough control. And so finding the right place on that line is quite the design challenge that I'm only now really appreciating.
- [14:10]
A
Yeah, and something I think that separates or creates the opportunity for when is a thin wrapper not enough. Right. So this whole like, debate over, like something is just a thin wrapper around ChatGPT or something like that. Right. Being this knock. And I think what we're finding is that, you know, true AI first companies, and I would count it intercom modestly among one of those, but other examples might be, for example, Lovable or Granola is another great example, I think, where it isn't actually like taking a prompt and passing it off, but it's wrapping not just a killer user experience, but a lot of model improvements and safety around passing back and forth, you know, and so that's a good kind of non disruptible area to be in, you know, I mean, ChatGPT recently added some kind of meeting notes type thing and the question becomes, oh, did granola just get Sherlock? Did they get like, you know, killed overnight, squash stood on, whatever? And I think it's getting harder and harder because it's like, no. Granola is a pretty substantial product actually, even if it seems very simple on the surface. And so at least that's my hope. But it would be a pity if the end point of all of this is there's just one piece of software now. Like, you know, you interact with your AI and that's it. I hope that's not where we end up because it wouldn't be as fun, I think as the alternative, it felt.
- [15:30]
B
Like that for a little bit. I think even 18 months ago, maybe when I would play out where this is all headed, a lot of the versions of the world that I could imagine ended in, oh my gosh, are we just going to have these all in one tools that do everything? And I think since that point I am coming around to this idea of like, okay, there's a little bit more defensibility in the app layer than I originally expected. And I think that Granola is the perfect example of a tool where, yeah, I'm going to keep using that. Not only because I like the interface and I like the improvements that they've added, but also there is this level of personalization and when you've dumped context into a tool, the switching cost then becomes extreme to port that over to another tool that doesn't know as much about you. And from that standpoint, I want to imagine, you know, if I'm using Intercom and it's two years later and in theory, who OpenAI or somebody comes out with like the perfect solution. Yeah, I wouldn't be so quick to jump because I would be leaving all that existing context.
- [16:30]
A
I'm not so sure. For what it's worth, by the way, I mean, like, maybe there will be one piece. I go back and forth all the time. Which is kind of the fun of riding the wave of the current moment as well, is like looking at what's going on and trying to do the impossible thing of predicting where it's going to go. But I mean, my latest kind of jam is using cloud code inside of Cursor on my computer, right? It has made me kind of go, oh, maybe this is the final piece of software. I mean, not that, but it very much feels like it because it is. You know, I started using that setup for code generation. Hey, maybe this is a slightly better version of lovable or whatever. But what it made me realize is you can do anything, anything on your computer through that ui. You can tell it to write a script on your computer and run the script that does anything. You can tell it to pull a bunch of stuff through, all done through claude code in cur. I mean claude code in your terminal if you want, I suppose. But it's kind of like having a uber nerdy like Linux sysadmin type person next to you and you just say to them, and this is how I publish stuff to my blog, which I used to do through like run the GitHub commands and now I just go into cloud code and you can say, just push the latest post to my blog. And it like prepares the GitHub thing and it says, okay, that's done, that's live now. So it's a different way of thinking about using your computer. Not everyone's going to do that, right? That's the uber nerdy like end of the spectrum. But again, I think there's going to be like one of those spectrums. And one of the interesting things I think is the uber nerdy end of the spectrum is pretty overserved right now because like for the longest time programmers have been like nerdy people that love having like their text editor set up, you know, Emacs versus Vim. And I got my set up just the way I want it. And they want the complexity because the complexity entails power comes at power, right? But what's interesting is the programming community has dramatically changed overnight, almost right? You and me are programmers now and every designer in the world is a programmer. And so I do see the opportunity for more user friendly interfaces to that type of power, you know what I mean? Like access to the control everything on your computer without having to go into something that looks like a very 1980s hacker user interface. So I think there's tons of opportunity. Is that a thin wrapper? Maybe, but maybe it's adding a whole bunch of value at the UI layer as well. So I'm interested to see where that goes to.
- [18:59]
B
Let's talk about that more future state a little bit. So I want to return to the tanker analogy that I like as a picture that you shared. So, okay, you're steering the tanker. You're moving towards this AI first world. What are some of the ways that you've had to intentionally evolve the way that the practice of design is performed within Intercom? Over the last couple years, at a.
- [19:22]
A
Practical level, we reorganized design to be centralized. When it came to the point where we realized we were going to have to go all in on fin and actually do fin and truly invest in it from an R and D point of view, we didn't have the time or patience, frankly to go and do a big reorg and redesign our reporting lines to match how we wanted to do the work. So we centralized design, and not just design, like all of R and D really. And we started organizing instead around work streams. Work streams were areas of effort that we could spin up or down for an indefinite amount of time. A workstream probably be as short as three months or some of them have been going on or, sorry, three weeks, some of them have been going on for over a year, I would imagine now, you know, but it means you can. There's one DRI for a work stream, not a triage. Very centralized responsibility rather than diffuse responsibility. And the team structure is very cross functional. So not just R and D people, but sometimes also sales and marketing people pulled into those teams also that has created a certain amount of chaos is too strong a word. But like, you know, instability or uncertainty or ambiguity about how we're working. But it has given us and it has been the only way that we have managed to do what we've done, which is like your analogy of steering the tanker of being able to take parts of the org and really properly 100% put someone's focus or a large group of people's focus on an important thing and be able to make a decision on a Monday and already be working and shipping even on a Friday against that goal. Something I say quite often is the start of an S curve of innovation is a great time to be at designer. An S curve is, you know, that we're at the start of an S curve now where the beginning of the technology is slow and then feels like it's advancing quickly. And when you're in the middle, you don't know when it's going to flatten out. That's where we are now. How much further is this going to go? Almost vertically before it flattens out. Think of like the first iPhone you got. The first few were incredible and awesome and every new release was. And then like what's the difference between the last few? No one really knows. Right. That's where we are with AI right now. The start of an S curve is a great time to be a designer because there is so much to be figured out. We don't know how any of this is going to be played out. There is so much low hanging fruit, so many big open questions and we are busy, busy at Intercom because of all these questions I've been talking to you about that we've had to figure out about fin and how agents should work and so on. The other thing about the start of an S curve is it's not a great time to be stuck or beholden to the old org structure and the tendency to get pulled and dragged back towards shipping that. And that's why at the start of an S curve when there's a lot of the possibility, space is wide open. We have many, many futures that could play out from here now and it's very hard to pick. And so that's why the centralized design team with a lot of optionality is a better structure. Some of the ways the role of the designer has changed at Intercom. Well, we've created a specialist AI designer role. We only have three of them compared to about I think 20 product designers we have at Intercom just to get. But these are designers who are like deep in there. In the AI team we have an AI team of I think about 40ml scientists and researchers and they work directly with them on often on the models. You know, on the prompting there's a lot of looking at the output of the models and eyeballing us and developing an intuition. There's a lot of quantitative assessment and a lot of deep and technical understanding of how the models work. And there's a bit of design as well as in UI design, our product designers are also leaning hard into this AI world. We haven't like mandated any new requirements or skills. I think it's still very early. But my advice to everyone on the team has definitely been to lean in hard and just experiment. And that's what we've been doing really. So everyone on the team has taken a goal to ship code to production in Q2 in this current quarter that we're in, we're in the middle of trying to do that. So we're running workshops that helps every product designer get set up with a local dev environment of intercom and fin, and then helping them using AI assisted tools like cursor to find something very simple and fix a bug or make a copy change or tweak some CSS or fix some padding or something like that. Something small. Right. But I think that will help us understand, like, where's the expectation bar, you know? And it turns out, I mean, by the way, I wouldn't ask people if I did it myself. And I was like, okay, well, at the very least, it's actually eminently doable. Way easier than I thought it would be. I'll tell you something, I wrote about this right on LinkedIn and it got, like, a surprisingly large reaction. I have been surprised at the, wow, how do you get people to do this? Or how do you do this in the first place? Maybe a certain amount of this is the culture at Intercom and the way it is open to such experimentation, or certainly open to being very AI forward. Maybe I get it for free and I'm like, doing this on easy mode. But the answer is you can just give it a shot and find a friendly engineer, maybe get a technical designer on your team who wants to champion this. That's what works really well for us. And then just go and give it a shot. And I think once you try things, you will find they're a lot more accessible and easy. But it is definitely the willingness to roll up your sleeves and experiment and not talk or think about it, but just give it a go and do lots of, like, disposable little experiments.
- [25:03]
B
I saw the screenshot of everyone enthusiastically talking about their first pr and it's cool, right? Like, I remember where I was sitting the first time I shipped a Polish pr. It feels good. And so to be able to give that moment and even dedicate the resources where you have, like the workshops to help people get environments running. And that's the blocker, right? Like, the code piece is not hard. It's just, can you get an environment running, man?
- [25:25]
A
I will tell you, a lot of people have said to me or people have even emailed me about that thing. That's why I'm surprised at that. And they're like, it feels so good to ship to production. It actually, like, it's such a dopamine hit. And I think it's a little bit the Plato's cave analogy where, like, you're looking at the shadows and flickers of reality on the wall. That's like a little bit what it's been like for designers who are like designing in Figma and then someone else takes their work and makes it real and then it goes to the users. And it's very hard sometimes in design to get access to users. We all talk about it, but it's a hard thing to do. And very few product designers go out there and like spend, spend the day with customers and feel that thing, feel the reward or feel the pain of someone using their product. And I think it's the same even just shipping something to production. You're like, I did a thing. Actually, it was different before and now everyone sees it my way and there's some tangible feedback loop. I think that has been kind of missing and a little bit neutered about the design process. And I think really having skin in the game is a super important thing for designers. And the more you can get close to what's shipping to users or close to the customers, I guess. I mean, we always say it as a truism, but interestingly, shipping code has had that byproduct and then suddenly you've got this pride feedback loop that kicks in. We can fix polish features. And so that's where I hope it goes from here, is that everyone ships to production, we get to learn. And then people go, okay, I'm starting to learn what the level of complexity. Oh, it's not just adding a where one was missing, it's changing some padding. It's not just changing padding. And now we do have designers at the adding features level of what they're building and shipping. So what's going on there? What's going on is the designers are nibbling away, right, from the kind of P3 quality bug fixing, quality of life thing. But they're nibbling away. And some of the more pioneering designers are even starting to develop their own front end features. At the other end of the spectrum, those same designers or other designers are also like Vibe code prototyping and lov and so on, right? And they're building these commonly, I think throwaway, like 70% their prototypes, like better than a wireframe, but doesn't quite look like our product and isn't finished, but it's, you know, it's a good fast way to get going. Think about those two things. Think about design like performing this pincer movement almost on what they're doing. They're Vibe coding throwaway prototypes. But more and more, I Think you can imagine like that getting hooked up to your actual design system, those prototypes becoming more and more realistic and maybe even being like, like the first draft build that you might hand off to engineering to actually go, you know, it's got a few react components or something, but probably over time taking more and more of that as well and building more and more finished prototypes. And again from the other end to doing your P3s, then your P2s and actually building features. I see a possibility of a future where designers, where those trends meet in the middle and designers are essentially writing from scratch or iterating in production full aspects of the front end code and possibly even ultimately the entire front end experience that would dramatically redraw the role of what it is to be a designer. Right? And it would redraw the relationship between designers and engineers. Like we have this very jagged frontier of handoff right now and everyone tries to get away from handoff in design, you know, in a sense, while at the same time doing it in a very like detailed way. But it's a hard part and it's very lossy part of design. Often frankly design are doing their best to like draw everything out to hand off to an engineer who's like less visually oriented, let's say than a designer might be, you know what I mean? Less interested in all of the details. And there's just a bit of a frustrating loop to get the details right, you know, I think the front end, back end distinction would be a way better interface point between designers who have the capability to code a lot of that front end and engineers who are going to manage the much more complex engineering challenge of the backend. I'm not going to be totally naive. There's some front end security and all this kind of stuff, man, the agents are going to get good enough to code front ends. If you think about it, one of the very finest training corpuses in the world is view source of HTML, CSS and JavaScript. It's the ultimate open source data resource. So if these bots can do anything, anything in the world at all, it's right great front end code including that's like advanced and secure and performant and accessible and all these kind of things. So I mean, I'm not vying for this, I'm not leading an incursion into engineering's territory, but I will make one more point. Okay, which is, is, which is this. If you go Today to the OpenAI careers page and you look up one of their roles, which is forward embedded engineer, have you Ever. Have you heard of this phrase forward embedded engineer before?
- [30:30]
B
I have not, no.
- [30:31]
A
Okay, so forward Intercom has this concepts now as well. I'll try and explain what it is. Here's the thing we and many other people have learned when you're building these AI agents, I was talking before about like there's a frontier of like complexity where you're, you're giving people a very sharp, sharp blade, right. And they might nick themselves at these blades, this blade if they're not careful. If you give them too much control over what the agent can do, they might inadvertently make a mistake and like tank the resolution rate or something like that. So when we're dealing with new customers we have realized that like a little bit of hand holding, especially for some customers who are coming saying like we have a large volume of queries that we want FIN to answer. Some of them are very complicated. They're dealing with advanced systems like issuing refunds or something like that. So we get in there and in a kind of a professional services style way, we help them set up fin in a way that's exactly right for them. Right. So you have this dynamic a lot now with companies like Intercom, other companies we're competing with do this as well. And companies like OpenAI where you have this role of, for a deployed engineer that goes into the business and embeds in the business with them and finds out what it is that they need and helps them get it set up. And it's a way of managing the, at least the phase we're in right now with these AI agents that like high level of complexity of integrating with them. Here's what's interesting. Go to their website, look up the job listing and there's something there in the job description that says something along the lines of as a forward deployed engineer at OpenAI, you will embed with customers, understand their problems, synthesize them into solutions, research how to do it and propose solutions that solves their problems. And I'm looking at that going holy shit, that's a designer. Like you're describing the role of a designer, right? So these are engineers going to do exactly the role of a designer and using their coding skills to solve these design oriented problems. So I think we designers should a make no apologies for using our design skills and adopting coding skills to also solve. But look, there's a fluidity to the roles that we should embrace and accept and welcome fluidity to the roles.
- [32:40]
B
And also like a land grab crab is almost the picture in my head right now. You know like adapt or die kind of.
- [32:46]
A
Do you think engineers wake up every morning going, I can't wait to close some three P3 bugs right now. I think they don't want to do like, I think that's less interesting than a lot of the stuff that they.
- [33:00]
B
I agree.
- [33:01]
A
Get out of bed in the morning to do now. Like designers don't either. But it will be a lot easier for designers. And I'm only, by the way, the P3 bugs framing is a very. I'm trying to like almost minimize what this actually means means, you know, but like that's where we are today. It's only a land grab if it's not a mutually beneficial setup. Right. The titles perhaps, you know, is a healthier place to start from. I know that's trendy as well. Like there's a lot of that. You know what I mean? We don't do UX design anymore. And I'm. I think that's awesome. But I also think it might and I. There's no shade intended in what I'm about to say at all here. I think it's maybe thinking small a little bit because like I think that that's a great change to make. But it's simply one little ratchet.
- [33:45]
B
Yeah.
- [33:46]
A
That are going to happen of much bigger, more important ones than like titles and things like that. So let's say a couple of weeks ago Lenny's podcast came out with this survey and it was like designers, 24% of designers are like, I can't remember, like miserable or not optimistic about the prospects of AI in their role or whatever. It was like 1% above the others. And yet everyone was like, oh my.
- [34:12]
B
God, designers are so designer and so on.
- [34:15]
A
And I'm like, it's. I mean that is one way of looking at things. But there is a much more agency minded, if you'll pardon the pun, where like the, the individual can have their own agency way of taking this. I think there's a whole, there's big opportunity to have a whole bunch of fun and ride the wave of the changing roles. And I think that's a way better mindset than the like hunker down and, and see what kind of a storm is going to blow over us here. A much healthier mentality.
- [34:45]
B
Agreed. And I see the latter mentality quite a bit. And you can tell that there's a very large and very real chunk of the market for design right now that is feeling a little bit uneasy that yeah, AI and code based responsibilities introduce all these new opportunities. But at the cost of what really matters in design. And there's this pressure. The UX piece is, is yeah, it's flippant and easy to make that change but for a lot of people that's the richness of design and it feels like it's kind of getting squashed down. I'm curious if you have anything to say to those types of listeners because I'm sure there are people listening to this right now that feel some of those emotions. Right.
- [35:26]
A
You know, I wouldn't poo poo it or I wouldn't dismiss it or anything like that. Like that's all valid and I guess like different people came to design with different motivations. You know, some people are again guess like have this notion of themselves as being extremely craft oriented and that's their self identity and maybe they see like a, a wider definition of design as being like a threat to that or whatever. As someone who's been in the tech industry for a while now, I personally I feel like I can get to be more excited than scared or intimidated by the change that are coming because I've seen changes like this happening and I do see that a lot of people in tech have not. We had a very stable, unusually stable decade in tech there and in the tech industry. And so this is very different for a lot of people who have experienced that. In those early days of the web there were, you know, these print designers and younger web native designers I guess you would call, call them, which I was one the print designers, I mean had an amazing experience. I learned a huge amount from them and seeing their work. But you could see, you could look at their work and you could tell that was a print designer versus like a more native web designer. And you can look at work from that period. Some of it by the way is what like the charm of the from that period is from these like super rich detailed like image mapped images that had to be sliced up into a million different like overlapping table views and so on. But over time the web found like its more native expression of what it wanted to be which was a more like rigid I guess kind of sense of layout. And I just feel like we will start to see a similar mindset shift either among designers in this era of adapting and adopting an AI first mindset. I think no matter whether you're a woodworker or a web designer or whatever, like you got to know the grain of your material, you've got to understand like what it wants to do and what it doesn't want to do and things like that.
- [37:20]
B
Well, there's opportunity even in a similar way, tying it back to what you were talking about with Fin and Intercom and proactively disrupting yourself. That same type of alpha exists at the individual level right now with the assumption that like this is not going away. You know, there have been bubbles and the people that have leapt on Crypto, for instance, fell on their face. Yeah, I don't think this is that. And there is an opportunity here where when you can be quick to adopt these new technologies, be the ones that are playing with the new material, man, you can put yourself on a different type of career trajectory. Like right now there is that opportunity because, yeah, maybe tying it back to your print versus web native parallel. This is that level of a transition and it's happening right now.
- [38:05]
A
Yeah. Or bigger. Potentially.
- [38:07]
B
Yeah, maybe way bigger.
- [38:08]
A
And I mean, the thing is it still looks dumb or broken in parts today or it still looks. Well, it's good for the first 80% and then it falls down or whatever. But again, if I look at a lot of the output of lovable. Right, right. Often aesthetically, but even like culturally almost, it reminds me of like a GeoCities or the MySpace aesthetic. Right. Everything is a little bit like there's a Starfield background and it's lasers and kind of fun and kind of disposable and it lives on this like URL called electric toothbrush, whatever, Lovable. Just like you used to have like geocities Avenue or whatever, you know. But you wouldn't look at those dumb old Starfield websites like with animated under construction GIFs and go, oh, the web is silly not going anywhere. You have to be able to look at that and go, where did the web go from there? You know, or likewise, I will say I. I worked at Google in the, you know, early stages of Google through their kind of adoption of, of design as a thing. And at the time that I joined, I mean, Google was a search box and that was what it was notorious for. And you could look at Google and go, wow, that's a very plain form, as plain a web form as you can get. But from that, that very simple starting point flourished a whole suite of apps and products and AJAX and mapping and online cloud storage and all this kind of stuff. So from very like toy like, or Humble or Simple Beginnings or, you know, I guess what I'm. The analogy I make there is I look at the Google single prompt input UI from 2004 and the ChatGPT UI from 2024 and I'm like, both of those are just, just the nascent beginnings of something massive. And it's all going to go somewhere different from here as well.
- [39:55]
B
The way that I kind of like to think about it is that moment in time with the videos and some of the early generative AI stuff where everything had like six fingers and you had Will Smith eating the spaghetti. You know, every new use case of AI has to cross through that spaghetti moment. And then at that point in time there are two types of people. There are the people pointing and saying, look, he has six fingers. Or there's people being like, oh my gosh, this is kind of goofy, but can you imagine where this is going to go soon? And those cycles are condensing and they're getting shorter and shorter and shorter and man, there's no time to point and laugh at the sixth finger. You have to start imagining. And I think that's like that level of agency that you're talking about here. It's like, what can this be? How can I be early on this? And man, it pays dividends.
- [40:44]
A
I do think that getting more technical actually is again the start of an S curve is an advantage because you can just like dig in and experiment yourself and then realize like ways in which this actually sucks for everyone. And you're a designer and you can make it better, but you have to get in there and understand those pain points a little bit first of all as well yourself to see those opportunities.
- [41:04]
B
I'm a big believer in the power of video to explain my thinking as a designer. So when it's time to get feedback, I'll drop a loom link and slack and another link to a FIGMA prototype and feedback will be scattered everywhere. And I mean it's a mess. So I'm building the product that I've always wanted to exist and it's called Inflight. You can kind of think of it like an Async crit. It's an easy way to share a video walkthrough along with an interactive prototype or whatever you're designing. And then AI interviews the people on your team to get you the feedback that you need and organizes everything for you in a beautiful insights page. So right now I'm only giving access to dive club listeners. So if you want to be one of the first to use Inflight, head to dive club/inflight to claim your spot. Let's zoom out for a second and maybe we get to the plateau of the S curve, things start to settle, become a little bit more predictable. I kind of want to have you Talk about the role of design. And it's a little bit hypothetical in nature, but in that world what are the durable skills? Like what are the most successful designers doing to really make an impact at a company like Intercom in your mind.
- [42:17]
A
Just jump in, be willing to take like to learn, be a self led learner. Being a self led learner is probably the most important like characteristic and being willing to toy and play with the role. And the other side of it I think is how like you know, the identity crisis perhaps in design from the last few years it's been kind of interesting to see. So the Apple brought out the liquid glass stuff, right? And it's such like an interesting kind of milestone because in a way it is like Apple just did what Apple does and like what designers everywhere have been trying to do, you know, a certain cohort at least of designers to like take this level of attention to detail and craft and dial it up to such a ridiculous degree that like it's a crazy amount of resources and effort and willingness to push. And in that regard it's kind of a spectacular achievement. So it feels like that dialed in craft version of design where it's all about aesthetics and detail and showing the value of your work through an almost inordinate stupid amount of detail to include in the thing. Like the precision and craft being demonstrated that way. And like Jony, I've talks about this in a beautiful way, do you know what I mean? Like seeing that someone else in the world cared enough to do that thing. I'm not denigrating it, but it does seem like the wrong version of design right now or the very craft oriented end of the design spectrum compared to, to the problem solving oriented end of design. And I think that design has had this very like, maybe not split because it's not different factions, but a very broad sense of what is design like? Is it a very UI craft oriented thing or is it a very like deep systems, problem solving oriented thing? The answer of course is that it's both. But the craft folks, I think sometimes to their, their detriment fail to take into account what does the world care about, what does the business care about, what really makes a difference, what's going to really make design valued in the company and not just be a little bit of a like, well, my design buddies think that the work I do is cool, right? So there is in reality a spectrum there of how people think about their work. Meanwhile on the other end there's probably a dearth of appreciation or sophistication in the craft craft and how the importance of really putting that level of detail and care into the problem solving work that you ship matters as part of the overall thing. I would hope that we would get to a place where these things can be married. And this is not like, contrary to how I see some people think about it, oh, we have to learn code that pulls me further away from the craft. I'm like, no, it doesn't. It gives you a tool to better engage with the craft. Your craft. Craft is not drawing pictures of the real thing. Your craft can become the real thing. Your craft doesn't have to be red lines describing the distances. It can actually like. You can play with the responsive layout and the speed and movement and things like that. And you will always be able to produce something better yourself if you're building it more directly than building an abstract representation of it. So like, I think the absolute weapon designers of the future are the ones that can marry both and see that like going in a more technical direction is actually in a sense also bringing you closer to the craft side of things. But at the same time, this is not going to be craft for craft's sake. It's going to be craft for, you know, a business oriented, problem solving, customer value point of view. I think that's what we've been trying to do in design for a long time. There's always been these little subgroups that value one or the other. And there probably should be. There will be room for sub specializations in future. But another thing that is, I think a truism at the start of the NS curve is there's a lot more space for generalists than specialists. We will get back to a place where specialists have their place. But I think for now, while we're figuring out what the hell is going on here and what the new rules are and what the new handoff points are and what you're supposed to do. And so like, it will always favor people who are willing to be more generalist.
- [46:26]
B
I have another kind of zoomed out hypothetical for you. Maybe we could talk about a project lifecycle for a second. Because in a world where designers kind of take control of the front end, how do you think that changes the product development process and the way that we collaborate engineers and I'm particularly interested even in the sequencing within that process.
- [46:47]
A
The process looks different. Remember I was describing to train fin, you go through this analyze, train, test, deploy kind of loop, right? And a few weeks ago we launched the analyze part of that, the kind of final keystone in that in that process. So this is kind of essentially reporting that helps you understand in an AI enriched way what are all your customers talking about, what are they struggling with and what are the big opportunities for you to go and train FIN and make it better? Right. Okay, so I'm just giving you the intro to like AI infused reporting. Think about it that way, right? How do you approach a product like that? In the old days we would have come up with some jobs to be done, you would have started some wireframes, you would have mocked things up, made sure and probably gotten to almost a very appealing looking high fidelity stage. You might have done some early user studies to validate those things and then eventually you'll get to a pro stage in the design where you feel like you can start to build and it won't be a last cause and so on. And there's like, you know, people talk about the double diamond and all sorts of processes. Right. Leading up to this, for our analyzed product, we took a very different approach because it was an AI first product, we couldn't have done what I just described there, the kind of linear approach, because we didn't know what shape of data was actually going to work, you know what I mean? We didn't know across all of the different customers, let's say. So an aspect of analyze is topics. We automatically analyze what everything your customers are talking about. You break it down into topics and subtopics. Automatically you get this amazing map of what all of your customers are contacting you about. Never been possible before. Super insightful and then useful to go and it's telling you what to go and train FIN on to make FIN better. Right. We couldn't have designed that up front, so we had to almost do the build up front. Get to a point of confidence, I'll explain what that means and then do a more detailed design phase. So our AI designers worked with our ML scientists and some customers that we partnered with and just got it basically working for them. And this was like a lot of massaging the data. There's no user interface. There's like producing reports for them in spreadsheets maybe and sharing with them and they're going, oh, that looks good or it doesn't or whatever. And understanding what do we need to do in the back end with the model such that FIN can produce for a wide range of different customers a good, good kind of topic subtopic breakdown. Then you're like, what does that look like? Do we have tons of topics or just a few or whatever. And then you can go and start to design your ui. You show that to your customers, by the way and like essentially very quickly, code, vibe, code your ui, you know, with engineers. We did it, but it could be a throwaway prototype the way I was describing it as well. Anyway, here's the interesting thing. Once we pass a certain threshold of we figured it out, this is the rough shape of the solution that now then we come and we start to do a lot of like high fidelity design work. Then we started to add a lot of like detailed UI styling. And I'm going to be honest with you, it was kind of nerve inducing because we were a few, couple of weeks out from shipping and we were still like working on the front end quite heavily, you know what I mean? At least from an aesthetic point of view. We actually even brought in your friend Von's Mans who worked on project, so Fon's worked on this with us. It was a lot of fun. And so it just kind of suggests a different way of approaching the work where upfront you're doing a lot of what I would call like material exploration, figuring out the data, figuring out the shape of the data. And sometimes it's like, is it even possible? No, we can't come up with good suggestions for what you're supposed to do on that amount of data. Okay, we need a different approach then, you know, and then you're figuring out roughly what the shape of the solution is in a very rough, prototyped, co created with customers type way. And then you're like productizing that. Now maybe that's not a million miles away from how a lot of companies do design, especially if you have great access to a small number of customers. That's the way a lot of small startups do design. But for bigger companies it's harder to do that I think because there's a predictability that gets yanked out of the whole process there. We didn't know what the product was going to look like. We knew how it was going to work and we actually had higher confidence that it was going to work. Great, great. But lower confidence and like what's it going to look like? And so there's just a kind of a different mindset that you have to approach the whole project with even as a stakeholder internally in the company. You know, I mean it's not a.
- [51:11]
B
Million miles away, but it's still backward, you know, like it is flipped, it is fundamentally flipped. Which then makes me wonder how does the role of research evolve as the product development process is kind of flipping.
- [51:25]
A
There has been so much kind of going on at the high level strategy level that a lot of our research has been oriented towards foundational, you know, what's the state of the industry, what are our customers thinking? A lot of that foundational research work to just inform our strategy. So we're not doing a huge amount of let's say user study research. And by the way, I think again we're very early depends on where you are with your own product evolution. But like we're not quite at the, like tighten the screws everywhere, optimize everything about Fin, you know what I mean? Going in and like hyper optimize everything. I still think there's some fundamental like shit we've got to figure out about how the product works and figure out what are the big things that we got mostly right but not quite and iterate them. So I don't think we're really in the fine tuning stage yet. I do think user research is going to be a really interesting thing to see to what degree do interesting products or do reliable agent driven products come in to manage that because it's always been such a manual labor aspect of the design process process. It'll be interesting to see if like something good can happen there that makes that a lot simpler like automated testing. Engineers have this idea of lint testing right where they have like hey, is there any regression in like the code changes I just made? And it'll warn you it would be cool to have something like that automatically built into figma or whatever tool it is that you're using that is trained on the experience of hundreds of thousands of users on the Internet and how they respond to common patterns and so on, on and giving you some warnings about that. So that might be interesting too.
- [53:02]
B
Okay. We've talked about a lot of different changes and a lot of different futures of varying degrees of certainty, but it's quite clear a lot of things are changing. So how has all of this impacted what you are looking for when considering adding new people to the design team?
- [53:20]
A
We have made the decision ourselves to decide to try and invest in an AI centric future that we believe is kind of coming. Might get the details wrong but like broadly speaking I think we'll be correct and so we'll be directionally moving fast in the right place. So you know, internally in the team today we're like well where is everyone? And trying to figure out how to get going and what internal supports are needed. I think for anyone new joining the team, we would certainly want them to be philosophically aligned with that and ideally like have some demonstration of some evidence that they're already doing that as well. Well, there's a certain amount of, as a, as a let's say design leader of facilitating and laying the path and greasing the wheels or whatever for people to go down a path that you. But ultimately only to the point where it becomes a self fulfilling thing where people are like, oh, actually this is, I want to, I want to drive myself down this road now. You know, that is probably the primary thing. You know, the kind of healthy disregard for boundaries or roles or dogmatism or whatever would be another thing to carry through. And then I guess I could say characteristics. I don't know if you can turn these characteristics on or you're just born with them or you acquire them through experience or whatever. But like curiosity, drive, a little bit of determination, a little bit of that thing of agency that we were talking about. Low fear index, I guess, you know, for like what could possibly go wrong. I know people listen to these shows looking for the insider baseball thing. I think there's interesting opportunities in, in design systems right now. So, you know, design systems have had their role and I've been maybe fighting the good fight valiantly for a long time. I think design systems have a really, really big opportunity to emerge as an important part of the tool chain in this new world. I'm talking about where your design system helps you very quickly build. You know, you're building in lovable or whatever it is with your native design system and it's very quickly actually the design system becomes much more a about enablement and building really quickly. I talked earlier about how we're moving to a centralized design team and that has increased the importance of design ops for us. With the centralized design team, there's just a lot more like trains to keep running. We've never had a design ops person because how we've liked to run the teams. There's a lot of responsibility at the individual level and team level and so on. But I think the changing shape of the team means you just need some other ways of running the team. Our design managers are becoming much closer to the work now and becoming involved in the work. When you have a centralized team and a lot of people in maybe disparate tracks of work now it becomes really important to have the people at the layer above them knowing what's going on with the product and trying to tie it all together and trying to orchestrate it all together so it's not like a disconnected bunch of 12 different work streams doing their other thing. But actually the important things know about each other and they hook up and become something like simple and considered and so on. We've always had a principle around being simple by default and flexible under the hood. Intercom is a super powerful and deep tool. How do you keep that accessible to a broad population of users but still have that power that's there if you need to access it? And our catchphrase of simple by default, accessible under the hood has been a very simple way of thinking. Oh, okay. So you can kind of progressively disclose those things and make it accessible. Today I found myself saying agentic by design crud under the hood. I think good design principles and good approaches to design should be relatively evergreen. This isn't a fashion moment that we're following. So you have to figure out what are the good design fundamentals and how do you express those same design fundamentals in this new world? So agentic by design crud under the hood is a cute little extension. Here's an observation I have about our own work in the last two years working on Fin. We built an AI agent, fin. And you can talk to Fin and interact with Fin and it can solve all your customer service queries. But if you're a business and you're setting up Fin, you still have to log into Intercom and the back end of Fin and train and test and deploy. And there's a bunch of SaaS screens to do that, like CRUD screens, create, read, update, delete, you know, web forms. The same web forms we've had for like 20 years. An interesting different approach we could have taken with, with that completely was to make the training of Fin agentic, right? Was to make it so that you chat to a teammate version of Fin and you get it to shadow your calls or you, you know, upload loads of stuff to it, or you to it extensively. And it's more like your teammate and so on. And maybe that's the direction that it will go. Like just my observation there is we've built an AI agent, but the back end of the agent is not agentic and maybe it shouldn't be, or maybe it will be. I genuinely don't know. I'm not making some prediction here. But what I am saying is more and more you can break that down at a feature level and you can kind of say, is there an agentic way of doing this or is this better is served with an old crud kind of. And crud sounds like old fashioned and cruddy and bad and so on. And so I'm not trying to make CRUD seem bad, but it might be the case that we're moving towards a future where a lot of our UIs are agentic and you end up popping the hood and going down and going okay, there's actually a bunch of like dials and knobs and sliders and stuff like that that I can tweak with if I want to get to it.
- [58:49]
B
What you're talking about is literally the thing that I'm going to to end this recording and work on designing later today is like how much of a setup flow is crud with predictable UI versus how much of it is like, well, what if you just started talking and had a back and forth and then it automatically translate all of your preferences into these settings behind the scenes and you don't have any ui? And that tension is definitely going to be a theme of the next year or two.
- [59:16]
A
And I see this with every AI product I'm using. You know, you start to use AI products and then you start to wonder about it. So the example that I actually just wrote about today, I wrote a blog post about it today, was I've been using Cursor a lot to like try and build actual proper applications, you know, and there's lots of stuff that you learn to do, right? You create a markdown file that has like a to do list for the things that you want it to do. And you're saying like always go to the to do list and update the markdown file to show you. And there's a PRD file and there's all this kind of stuff. So you set up your coding workspace like that and then there's a terminal and you're like starting up servers and so on. And I'm like, man, a lot of this could be friendlier. Like there could be some buttons to kill and start the servers and you know, there could be friendlier things that would make this easier to work with. There could be an actual little checklist of to do in the sidebar of Cursor and I could like see them getting checked off and so on. Now there's something I like about the whole like super flexibility of the, of the cursor experience. But like I was saying, you can kind of earlier, like you can kind of do almost anything on your computer with that, I've realized. And it's often better to wrap these multi purpose, deep, low affordance UIs in something more accessible and more easy. And it's not just so that you don't have to type NPM server start every time or that that's too geeky and you can't handle it sometimes even if you know how to do it, it's usually it's easier to click the green traffic light button button or something like that, you know what I mean? To start your server. So for all of these things, I think we're going to figure out how much of that should be agentic versus how much of it should be more traditional ui. And there are still things, even with the most hardcore AI tools that I'm using that I'm like, I want a little bit more traditional UI to come out here and I wouldn't mind some different ways to do this apart from a single text input, like almost a command line interface.
- [61:12]
B
You know, I think it makes me appreciate a lot of the things that you are implementing at Intercom because it's not just coming from this economic place or even solely a pressure to adapt or die, but it's also like you as the longtime VP are getting into the weeds of cursor and exploring and tinkering. And I think it says a lot about who you are as a leader. So I really appreciate that mentality and also, also you coming on and just giving the full behind the scenes of all the different things that you are wrestling with and thinking through and the changes that you're putting in place.
- [61:46]
A
Yeah, I mean the last thing I'll say is I'm having a lot of fun with the tech at the moment.
- [61:50]
B
I can tell.
- [61:51]
A
I think if everyone managed to tap into that, I think design would have a real golden age potentially. Because I do think we're on the cusp of an awesome new period for the design industry. But it's require a bunch of change management and stuff like that. And yeah, as I said, I appreciate you and what you're doing here as well. Giving a platform to folks like me to talk about it because might be a bit intimidating, I get it. But like we'll definitely get there together and things staying as they were was not, was not going to be a great long term situation for designers anyway. There's so many things you got to look at our industry or our discipline as a designer as well. There are so many things about how we've been working, working that could be better that are only the way they've been because that's the way they've always been or whatever. So yeah, I'm, I'm excited to fix a bunch of that stuff and onwards.
- [62:43]
B
Amazing. Well, appreciate you coming on today.
- [62:45]
A
Thanks Vision.
- [62:46]
B
Before I let you go, I want to take just one minute to run you through my favorite products because I'm constantly asked what's in my stack. Framer is how I build websites. Genway is how I do research. Research Granola is how I take notes during crit. Jitter is how I animate my designs. Lovable is how I build my ideas in code. Mobin is how I find design inspiration. Paper is how I design like a creative. And Raycast is my shortcut every step of the way. Now I've hand selected these companies so that I can do these episodes full time. So by far the number one way to support the show is to check them out. Out. You can find the full list at Dive Club Partners.