Transcript
I am here joined by two awesome people. I'm here with Ricky Robinette, who is the Senior Director of Developer Marketing at Google Cloud. He is a Brooklyn native, and he's been building things for 15 years for developers. We also spent seven years at Twilio together. And I would say, you know, he joined right after me, probably because of me. I'm just guessing, that's why. Yeah. And he is our perspective. He sees developers across countries, across companies. He has a very high-level view of what's happening in the developer world. And we're also joined by Jono, Jonathan Alford, who is the founder of Roboto Studio. And he builds every day with Sanity and Next.js. He's also a Sanity Pioneer. Ask us about that program afterwards. And, you know, Jono is building every day. He's making stuff work by Friday for his clients. And so his perspective is much more on-the-ground builder. He kind of mentioned what the session is about, which is, you know, when the cost of code is zero to generate, judgment is everything. GitClear did an analysis of 623 million lines of code, and they found that duplication was up 81%. You know, all this stuff was kind of like, it was just a lot of stuff. It was a lot of code, a lot of generation. And we all have the sense that engineering prowess used to be about who has a giant repo and is contributing on GitHub every day, and that's just no longer the bar. That's not the baseline anymore of what good engineering is. It's much more about quality. So I brought these two together to really talk about what the bar is right now and what quality means to them.
Stack Overflow just did a survey this year, and they found that 84% of developers are now using AI in their coding. That's going up. But they also found that 29% of those developers don't trust the output, which is going down. And usually when the quality, when the trust is going down, usually the usage goes down as well. But that's not happening here. And I'm wondering, Ricky, why do you think this is happening? And do you see this in the developer community?
Yeah, I actually see this personally. I'll share a quick story. I don't know if Andrew is still around. I loved how much Andrew talked about using AI to build dashboards and analyze data. When I started using agentic tools, one of the first things I built was a YouTube dashboard for my team so we could look at all the videos we were shipping for developers on YouTube, and we could measure how they perform. And one of the things that I'm going to say I built, my agent built, was a histogram for each video. So you click into a video and it's like, here's how the views went. And I shared it with my team. They're like, oh my gosh, this is amazing. And one person on my team was like, this really blows my mind, because I've tried to build this so much, and I've never found a YouTube API that can give you historical data on the video. Like, this is incredible. I'm like, yeah, it just found it. And then like a week later I was thinking about that comment and I went into the code and there was just a comment in the code that said, there's no API for this, I'll just make up the data. Right? Yeah. And I was like, oh, crap. And the best part of the story is I built the same dashboard at Google like a month ago, and it did the exact same thing. Like two years later, I went and had the histogram and I was like, no, that data is made up, and it fixed it. And I think as developers, at least as I use this more and I talk to developers who use this more, you just have more stories like that where you're like, I'm going so fast, faster than I ever thought. But if I don't know what I'm running towards, it's going to just run in the wrong direction in moments.
Yeah. Jono, you know, you build every day with this, right? Like, do you trust it? Well, I guess even more important is this, like, an old pattern that we're just renaming? Like, we used to just copy comments off Stack Overflow, and we just trusted that. But, you know, is this something new or is this something that's always been there?
It's interesting. So we've kind of had a little bit of a split in the team. So we obviously have some people in the team that are very old school, and they really want to focus on doing everything very secure, you know, running tests over things. And then we have people that are completely on the opposite end of that spectrum where you're basically building, like, you know, speed needs to be speed. And we actually had a fun incident as well, where essentially I've been building the main website that we have for quite a long period of time. And essentially we had to get ISO compliance and we started refreshing all the computers. And I was building for probably two months straight, using Claude Code, obviously. And I realized I hadn't installed an IDE. And then basically I looked at what we had built and I went, wow. Okay. So this is wildly different to obviously what we do with client websites, but some of it was complete slop, but some of it was actually pretty decent, what it had actually output. So it's kind of a mixed bag. And I want to say that you should always have those sensible people on the team to make sure that they push the slop code back from that position. But yeah, I think it's a great empowerment tool for clients and things like that, because you're getting more people that have never touched code before actually building something, which is awesome.
Yeah. So a little bit trust but verify. And it is important to have the people on the team who can actually do QA. Yeah. So I want to talk about what clears the bar for you right now. Right. Like, Jono, you know, you can build anything. You are in the business of meeting with the client and pitching them on what the vision is going to look like for them. But do you show up to that first meeting now with, you know, just five prototypes and let them pick, or are you still investing the time into setting the vision with them?
It's interesting. So I would say that one thing that we're doing a lot more these days on the initial conversation is we use a tool called Paper Design, which is very similar to Figma if you haven't used it before. But what's interesting about it is it kind of blurs the in-between point between development and design. And so we're seeing a lot more of the actual building of the user experience that the client wants to see before we even actually start touching the development. So you're seeing static HTML pages. Here's how the filter will work. Here's how the page structure works. And then start translating that when you actually go through all of the discovery stages and start building, if that made sense. Yeah. So vision is still important. It's kind of the whole thing. More than ever. I think everybody's so tired of... well, at least for me anyway, I'm so tired of sloppy UI patterns. So I think anything that really makes it a more distinguished, more interesting UI is, you know, doing really, really well right now.
Yeah. I want to talk about that. Ricky, you have always been a fan of the MVP. I remember when, at Twilio, I pitched you on rewriting our entire event stack so that we could have events, national or international events, and we'd have this whole registration thing. We were like, what's the MVP look like of this? Like, what's the Excel spreadsheet version of this that we can run? But is that still the case for you? Do you still believe... like I know your daughter is ten. She codes, you know. Are you telling her, start small, honey, don't do anything giant? Or is it just YOLO? Just build the whole thing.
Yeah. Yeah. Well, so my daughter started coding, you know, two or three years ago. She made a video of her vibe coding when she was eight. And, you know, it's been so fun to watch it through her eyes. But the thing that surprised me. So in 2011, I don't even know if you know this, I did this personal challenge where I was going to launch a new app every month and it got press and it was like a big deal. It'd be like somebody launching a new thing every month. And now it's like, if you're not launching a new thing every hour, people are like, oh, are you really vibe coding enough? And so I actually am a step before MVP in conversation with people now, of like the why. Why are you building this? How are you thinking about it? The scale can be anything, right? The tools can let you build so much. But if you don't know the direction you're going, it can get out of control. And one of my favorite stories: I've always been passionate about taking people that have never built before and giving them the tools to be able to build. And maybe about eight months ago, somebody came to me for the first time like, I want to build a thing that does this. I'm like, great. And they're like, okay, I talked to Gemini and it gave me the five agents I need to build. So now I need to build five. And I'm like, what are you talking about? And so helping people zoom out and think of those first principles of software development, for those of us that have been building a long time, I think has been more critical than ever. But it's like before you even get into MVP, making sure you're thinking that way.
I love that. I was talking with a few people last night at the Pioneers event, and something that came up was, you know, a lot of folks here are interfacing with non-technical teams, right? And they're introducing AI workflows. They're introducing content ops that are all new. And, you know, I know that everyone in this room is pretty sold on AI and agents as part of our build process. We have a whole Build with AI thing happening right now. And this is true in every major tech hub. But, you know, there's a lot of people who are not using AI. I think the latest Gallup poll said 24% of US employees are not using it weekly, or only 24% are using it weekly. Like, that's a stark difference from, I think, where all of us are at. And so I'm wondering, you know, Rick, you spent a career getting developers thinking about new things and convincing them to try new things. What's different this time? Is this normal? Is this just fatigue? What's the gap here and what needs to happen?
Oh my gosh. How controversial do you want to get, Jared? That's Hacker News material. Yeah. This is just going on Hacker News. It's getting clipped. Look, my wife is an artist, right? And I spend a lot of time in the arts. I think that it's easy to, first of all, lose sight of the antagonism people feel towards AI across the world. And I think, as an industry, we haven't really helped by saying, guess what? You're not going to have a job in five years and there's no plan. And, you know, all of this stuff. And so I think first and foremost, I like to change that narrative and be like, what do you care about? How can I help you get there faster? Right. The second thing is I'm very anti-slop, which is probably less controversial, but I think the counterbalance to slop is having core philosophies of how you think about AI. And I think a lot of folks, even myself, didn't take the time to think about that. Like, you know, I went to my team when I joined Google and I said, one of the core philosophies of this team, I want it to be: writing is thinking, right? And people offload the writing to AI, and then you're not thinking, and then you end up in the conversation and you can't actually debate or weigh the pros and cons. And so again, I think of those 75% of people who haven't used it, it's this blend of, hey, they probably hate it for some reason, right? Like some group of them may feel that it's ungrounded in the why. And then third, they may not feel empowered and feel left behind. And I saw a post on X that was like, the only way to stay on top of AI is to be unemployed, right? And I feel that way. I'm like, oh my gosh, I wish I didn't have a job so I could try all this stuff. But imagine if my job is to try this stuff, and you don't have that. So I think there's a lot of power in people who are further in the journey taking a step back and sitting with people and being like, how can we help you solve your problems with this?
Yeah, I think we've all seen the instance where building the thing and showing people how their lives can be improved goes a lot of the way to just showing them that this is a tool that can build things for them and help them. You know, Jono, when you tell a content team that there's going to be a new workflow, what are the objections you hear and how do you handle that? I think a lot of people would love to hear how you navigate that water.
Yeah. Well, if you want to have a look at something where you have a workflow, or not even a workflow, if you want to see what can happen with AI today: the next time you have a problem, a grievance with a product or whatever, set an AI up to use computer use and go through the entire contact form without having to touch it, because that in its own right is a great example of where, you know, if you need to return something, there's a million paper cuts to do it, right? I mean, it's very intentional. But in the same way, when you want to actually publish something, when you want to put a piece of content out there, naturally we're not as fluent as just literally talking about the content. There's a whole bunch of steps. So a workflow is basically breaking down those barriers to get the content out there. And I think, essentially, trying to create a small MVP where it just removes one paper cut is huge. So that's generally what we do as a first engagement. Go get rid of that paper cut. What's the biggest problem? Show that, and then nothing else outside of it.
Yeah, that's great. You mentioned UI, and one of the things that I don't know if you all feel, but UIs are just all turning into the same thing. It's like a vanilla Tailwind output from every agent that you tell to build a UI. What's the workaround? Are you guys introducing ugly styles into your builds? What are you doing to differentiate what your teams are building and what you're building?
Yeah. Well, I'm sorry, I have a story about everything. I work at Google. I asked Gemini to make me a slide the other day, and somehow the slide looked like that same UI. I'm sure under the hood we were having it write code. I'm like, how does my slide now look like Tailwind, right? This doesn't make any sense. So I am very anti everything looking the same. I just spend most of my time writing markdown instructions. It's so weird. I spend more time with markdown than code, and try to get very opinionated and make sure I'm giving my tools a very strong, opinionated line to work off of. And then, you know, I don't know if I'd say I "uglify" stuff or add ugly, but I want my tools to know what my personality is. I want them to know what it would look like to build something in my personality and brand. And I would just say, if everyone in this room did one thing, it would be: spend two hours with your markdown files and brush them up, and you'd be amazed how much better you get from the standard results on this stuff.
Yeah, I think that's kind of interesting. So I am going to throw a little bit of shade at you, but with Material Design, when it first came out, I thought this was kind of an interesting one, because the first time it came out, everybody got tired of it within a short period of time. And so every Android app looked exactly the same, and the same for Flutter. But one of the things that I actually think is interesting, that unfortunately Google does catch a little shade for, is the divergence from standardized, you know, user-experience-friendly UI patterns. And I think if I were hiring right now for this specifically, I would look out there for artists and people that were creatives and not yes-people. I would pick those weird and wonderful people that can do amazing stuff that really goes against UX patterns. And I think that is going to be how people brand their design in the future, because we can do it for free, basically, now.
Yeah. Plus one to that. Yeah. I was talking with a creative director here, and she was like, is AI coming after my job? And I firmly believe that design is the differentiator now. Like having someone... I mean, our design team. Big, big round of applause for our design team. This year looks amazing. From our swag shop to our boards. And I think having people who are opinionated, who are like, this is where we are going as a brand, this is where we are going as a company, and making sure that gets down to the agent, I think that is kind of actually... Have you done any kind of design system build, working that into skills and making that part of your process?
Oh, 100%. So we literally ripped off Vercel's design.md file when they put it out there. And it's really nice, because I actually think that's one of the biggest things you can do. So if you're building monorepos, I think having a design.md file that everybody in the entire company can point at, can point an AI agent at, and then basically create something from that markdown, is really important. And I think, by proxy, creating the press pages that you have with it, you know, like the press release packs, and it has the icons and the logo and everything. We put quite a lot of effort in to make sure it just doesn't look like shadcn. Yeah, if that makes sense. So I think the more you curate smaller prompts that articulate a specific kind of, like, brutalist, or more off the beaten track of design, the better, essentially. So yeah, we do that a lot with design.md files.
I love that. I have a two-part question, and the first one is you'll have to all entertain me a little bit, because I'm very curious to know the answer to this. I build a lot of software for one. Does anyone else here just build software that's literally only used for them?
Yeah, yeah, lots of people. So I built a meal planner app on Sanity. Everything's overkill about this meal planner app, but I needed to know the macros for myself and my family. I needed to know that. It needed to create a shopping list based on where I shop, because some things are not at Trader Joe's. I don't know if you've realized this. And I also wanted to do portion size, right? Like my portion is going to be bigger than my wife's and my family's, because we're all on kind of different goals right now. And I wanted to build this. It doesn't exist anywhere. You can import New York Times recipes. I love this thing. I'm literally hacking on it every week and making sure it's, you know, maintained. But I'm curious, do you all have an example of software for one that you've built?
Yeah, it's hard to pick. I have something similar to you, Jared, but I'll pick something different. So my daughter is super into Spider-Man, as am I, so it's kind of nice. And so, for Christmas this year, she woke up, she opened all her gifts, and then suddenly there was a note from Miles Morales saying, hey, you know, the multiverse has broken into your Earth, and I've installed an app on your dad's phone, and I need your help saving the multiverse. And she opened my phone, and there was a Spidey Sense app. And basically, it was like Foursquare, if you remember Foursquare. And she went around the city and she'd take pictures and Nano Banana would put the Spider-Man characters into the pictures, and she'd meet new Spider-Men, and it ended with this battle with Mysterio at the Brooklyn Bridge. And, you know, it was amazing, because I built it in a day, and I was thinking, I never would have made this for her, because before AI it would have been, you know, hundreds of hours to get this all together. And I could just hack this together. And we had this incredible family moment, and she was over the moon. And then she built her own version, which is classic. My daughter, she's like, now I want to learn how to make that. And so, yeah, anything I can do that is audience of three, really, for my wife, daughter and I, I'm always stoked to build.
That's so nice. That's so wholesome. Yeah. Yeah. Well, mine's completely the opposite. So we actually had this issue, which I'm sure a lot of people are seeing now, where if you actually set up a recruitment page or you set up some sort of pipeline, you're going to end up with a tremendous number of people. I know this isn't the most ethically great thing to do, but we created a careers Tinder. So basically the idea is we have this page where we can basically pipeline and then flip through the first version. Because one of the biggest things that I always used to hate, and a lot of people told me this from recruitment, is ghosting. And so essentially we made a very, very streamlined version, which was like, okay, unfortunately you're not going to make the first interview, but I'll make sure I get back to you. I'll make sure you get all this information. And on the other side of that is that if things go forwards, we can book a time, we can streamline it through something like Cal. So the whole process of building personalized software, which I know has fallen out of favor now, like people don't talk about it as much. But I would say right now, with the models being as good as they are, personalized software could not be better, and it's obviously not a fad anymore.
Yeah. I mean, there used to be a little voice in my head that was like, don't spend time on this idea. There's no market fit. It's not a good idea. And, you know, that voice has diminished quite a bit. But on the flip side, how can you build something that's going to last ten years? Right. I think all of us are hoping to build things that last. You know, Sanity's coming up on that. And I think that this software only gets better and better. I wonder, you know, how do you maintain something? You have to build things that a client manages well after the engagement, you know? How do you think about that bar? And how do you make sure that it's going to be maintainable, specifically if you didn't write all of the code? How are you going to go back in two years, three years and actually maintain the thing?
It's interesting. We've been having a lot of conversations about this with some of the folks at Vercel as well. So QAing is, I guess, the next big frontier. Because right now, you know, you have basically unleashed Pandora's box and, you know, you can't put it back in. So people are naturally going to go build something and build a whole ton of stuff very, very quickly. But the big problem is that tools like, you know, CodeRabbit and Greptile, they're great, but they obviously have the same problems that we do, which is, you know, token usage. You have to commoditize the thing. But I do see something like this effort basically being set up as a custom QA system. We actually have this internally at the moment, and it works really, really well, because obviously CodeRabbit can't throw something like a mega-expensive model at something if, you know, you're doing 100 PRs an hour or whatever. But if it's your own personal one, obviously, providing you have a way of saying, this is a smaller change, so don't use an expensive model; this one's a bigger change, make sure you use the big models. I think that I can see the future, but I know that there's not that many people that have cracked it yet, which is interesting.
Ricky, do you feel the same way? Are you telling your team, you know, make sure this software lasts for ten years?
I mean, I don't want to get too existential at a tech conference, but I'm like, I don't know what ten years from now is going to be like. And I've lost a lot of my career to trying to plan too far in advance. And I've very much become, like, now is the only thing that exists. And I don't mean that in a reckless way. I mean that in a stoicism-type mentality. And so I would actually encourage more people to be in the now, be in the moment, experience a moment, and let go of the future. The future is going to work out. And if it doesn't, oh my gosh, we'll have bigger problems, right? But I think that a lot of people being present in the moment, building the things that matter to them, is actually the answer to solving for ten years from now.
So you're telling the folks on your team, just build the thing, we'll worry about maintenance tomorrow.
Yeah, I like... That's what we've always done. We just pretend like we don't. Right. I think everyone, like, you know, what company have you been at that doesn't reach three years after you launched and say, oh, the legacy codebase is a mess and we've got to redo it? That is just the rock that Sisyphus is going to push up the hill of software engineering for the rest of time, probably.
I'm just glad you didn't drop a banana reference in that answer. Let's get it. Okay. Well, I want to ask you all. First off, is there any podcasts people should listen to? Any content that you're all listening to to stay on top of things?
Matt Pocock is a good one. I think it's great. I think Syntax is fantastic. And don't blow up my spot when I say this, but there's a great code for Syntax somewhere on the Sanity website. But yeah, basically, I think it's worth looking at those two in particular, especially if you're getting into more of the in-depth AI engineering stuff, for sure.
Yeah, I feel like everybody's going to think I'm just crazy at the end of this, but I love AI. I use it every day. I'm like, I'll have a deep dive in this. When I'm listening to podcasts, I'm trying to find people who feel differently. So there was just a great podcast on, I think they called it, the AI apocalypse, right? But most of my media consumption is trying to consume media from people who disagree with my perspective on AI, because I get enough in my personal life and on X of people who are as hyped as I am about it. So yeah, I wouldn't encourage everyone to go down the doomer route, but it's helpful for me to know the full picture of how society is viewing all of this stuff.
Yeah. Plus, it gives you, you know, knowledge into the objections that you might get when you're telling developers about it.
Okay. Last question. If you're hiring a developer, if you're bringing someone on the team, what's your advice to them on the one thing they need to get better at right now?
Oh my gosh. I used to say ship, right? Like ten years ago, I'd be like, the only thing that matters is shipping. Can you ship or can you not ship? Obviously not the challenge now. I tell my team the only thing that matters at this moment is taste. You touched on this, right? I want to know that you know what good is. I want you to have opinions. I want you to have an informed why of why you made a decision. And so, you know, my favorite thing when I interview people is like, hey, okay, this is the thing you did. Why did you make this decision? Why does it look like this? Or show them something we've done: what do you think could have been better? And if you don't have that taste at all, if you don't have those opinions, I think you're going to get lost in the mass of people who do and show that they do.
Well, yeah, that's spot on. I think one thing that I would highly recommend is trying to understand where the two things meet, where you have business goals and development skills, and understanding the future of AI. I definitely think it's worth just looking at X all the time to find these things. As much as I don't like it, I think it's worth knowing everything there is to know about things. Even if it's, you know, this week it's loops, next week it's graphs, God knows what the week after is. Those kinds of things are worth at least investigating and trying to bring to the team and say, you guys are doing it wrong, let me show you something cool to try.
Well, rad. If you all have questions, please find Jono and Ricky on the floor afterwards. I'm sure they'll be happy to talk to you about all of their opinions. Thank you very much.
