The Evolution of App Development

This is William. William's a portrait painter. He lives in the 1800s. At this time, the epitome of portraiture was creating the truest possible likeness. So, to achieve these skills, William has spent over a decade learning anatomy, light, composition, how to flatter a face. He really is at the height of his craft.

This is William's neighbor, Thomas. He's a sign painter and he works at his father's sign painting business, also called Thomas, where he's been working since he was 10. No labor laws back then. While he has no understanding of human anatomy, he is creative. He's pretty good with color, and he has a great steady hand and a great eye for detail.

And, finally, a bit further down the street in a fancy house, we have Clara. Clara is a young lady from a wealthy family, and she's just started her very first lessons in painting. She's looking to take this on as a hobby. They are all about to live through something groundbreaking. The first camera is invented and it's gaining popularity. It's big and clunky and tedious at first, but it is quickly becoming a smaller and more efficient and a mainstream way to take photos. And what is something a camera can do better than any other method in the world? It is to capture a likeness, an exact likeness.

So, the thing that all these artists for hundreds of years have been striving for, you can just do it. No skill or experience required. It's difficult to appreciate now just how much of a stir this invention caused at the time. But here's a quote that's often attributed to a famous French painter, who apparently, upon seeing some of the earliest photographs, said, from today, painting is dead. Does it sound familiar? What does this mean for our cast of characters? Clara may choose to skip the whole tedium of learning to paint altogether and instead use a camera. Without knowledge of light and composition, her photos might not be particularly good, especially at the start, but she'll get a basic version started very quickly.

Good enough to capture the likeness of her friends, pets, and maybe even some scenery. Thomas, who already has an eye for detail and experience in art in a way, is actually perfectly positioned to benefit the most from the camera. He could branch into photography and quickly eclipse what Clara is able to do and produce photos faster and with better likeness than poor William ever could. So, both Thomas and Clara are to benefit from this new invention. And this leaves William, whose world has been turned upside down. He has spent years honing this craft, and that can be replicated by anyone, with a camera and a click of a button.

But can it? This is based on a very real history of events surrounded by the invention of photography and its impact on painters. As we know, history has this tendency of repeating itself, and I can probably believe you can see that this is a thinly-veiled analogy of how building mobile apps has changed in the past year. William, Thomas, and Clara's experience with adjusting to the advent of photography is not dissimilar to all of our experience as engineers, adapting to the changes in how we build.

Hi, my name is Cady Crammond. I'm a developer success engineer at Expo, and this talk is about the evolution of how we build apps. Expo, as I hope most of you know, is a framework for building mobile apps with React Native. We've been around for almost as long as React Native itself, and we've also always been a favourite for React and web developers because you can build native iOS and Android apps in one codebase, writing mostly JavaScript. We've also been positioned to be a favourite of AI tools due to how quickly you can get an app up and running, and speed is of an essence when it comes to the impatience of prompt engineering. And with Expo Go, and the amount of training data there is for open source for LLMs.

In the past year, we've actually been doing what every dev tool company has been doing. We're trying to adapt to these changing landscapes, to continue to make the experience of building apps delightful to users. And, historically, these two have been the archetypes that we focus on. We have the veterans. So, veterans are the folks that have been building React Native apps since before auto-linking existed. They know what they're doing. They just need the tools that work efficiently to get things done. And then we have the experts, but we're saying experts in a related field. So, not yet a React Native engineer, so anyone who is, let's say, a React engineer, web developer, infrastructure engineer would be in this category. But now, we have a new kind of builder, the newbie, building mobile apps for the very first time, and sometimes completely disinterested in becoming a full developer. The way we've learned has changed forever.

So, say that I asked you right now to build me a one-off command line tool for budgeting. How would you go about it? Until recently, you might start by Googling about CLI tools. You might read some documentation on CLI libraries. You might find a YouTube video and teach yourself about how to build CLI tools. But today, I bet most of you would ask your agent. You might start by researching tooling first, but most likely, you still use AI to build this tool.

Question is, especially if you focus on web or mobile and don't really do CLI work, would you even read the code? The crazy thing about newbies going into mobile development is that the barrier to entry to having your own little mobile app used to be huge. You needed either a huge amount of time or a huge amount of money, or both. These days, though, you can prompt your dreams into existence.

There's a problem with this approach, though. Well, there are several, but there are two psychological ones. You've probably heard of the Dunning-Gruger effect. It's basically when you learn a smidgen, just a little bit, about the domain. You tend to feel like an expert because this is the whole sum of what you know. Well, an actual expert tends to be very hesitant in making confident assertions about a domain because they are aware of the gaps in their own knowledge. And there's another thing. It's called the Gell-Mann amnesia. Here's something that we can probably all relate to. Have you ever used AI to help you investigate something in a domain that you're already an expert in?

Maybe it's a really gnarly bug in React, or it's just something about a country that you live in. And you suddenly see it say something with absolute confidence that is blatantly wrong. So you go, hold on, that's wrong. So it does some more noodling and comes back with, oh, you're absolutely right. I was wrong. Here's the real thing. The amnesia part of this is that even though we have all lived through this, and we know that this happened, and we know that this tool is fallible, we still tend to trust it even when it's confident in areas that we are not experienced in. This here is called the Rumsfeld matrix, and it's basically a way to categorize knowledge, mapping it against the level of awareness and level of understanding.

So you have things that you know you know, and no, you don't know. And this is the easy part. But then comes the interesting part. There are things that you don't know you know, and things that you don't know you don't know. And this maps perfectly to the problems that we have using AI to build our new tools. So in the case of this newbie who might be code-coding a new app, they know what they want to build. They know the users, they know the screens, they have a great idea of what they want the app to be. They also have some known unknowns. So, for example, they know that they don't understand how payments work, so they'll just use Stripe. But this is the scary part. We have things that we don't know that we know. So this is where the AI might be wrong, but you're going to trust the output anyway. And also, this is probably the scariest part, considering all the attacks that we've had with open source, the unknown unknowns. So from a user's perspective, you press a button, and it works. We don't know what's going on behind the hood. You don't know that there are secret secrets in client-side code that shouldn't be there that could be exploited. And you don't even know that that's a bad thing. At Expo, we talk about this thing called the pit of success. You might have seen some talks about it. And even, I mean, it's a good pit. You want to go in it. It's basically making it so being successful in reaching your goals, your app development goals, is the path of least resistance. So that when you, like falling into the pit is the easiest thing in the possible to do. So you have to go out through your way not to do it. And for newbies, AI makes this both easier and harder. It already guides you to a success, but it doesn't give you the pitfalls. One thing we need to recognize is that this whole, this is a whole new category for developers. It makes it easier to get started, too easy even, because you can start with build an app that, and be so far removed from the code that you have no hope for bridging the gap into development. So this is extremely early days. So one thing that we are definitely recognizing is that the path for AI coding into our development is different from the path for developer developers coming into our development.

So as of recently, we have actually broken our tutorial on expert.dev into two, where you can do our getting started tutorial without focusing on the code at all. So we'll see how that goes. But the more important indicator of success is something we can all do, which is to check our own thinking. So to fight against the two fallacies we just mentioned, for the unknown unknowns, we can use AI to check that AI works. So for example, you could use a different AI agent to say you're a security researcher, review this app, see if it can be exploited. And the other thing that we should all do, which is to trust but verify. So it is very helpful, it's very common for AI to hallucinate based on past experiences, but an easy way to get around it is to ask it to show you're working with references.

Okay, I think all of us here are experts in some fields. In this case, I'm referring to being an expert in development, but not mobile development. You might already be a mobile development expert, which is great, but you might be a web development expert. So the experts in adjacent fields has long been a sweet spot for experts, especially for folks coming from the web. And we have a huge amount of guides for it. And this is something I myself have focused on a lot because I come from the web way back when I only produced enough to teach a couple of React Native courses. And something I've been thinking about recently is what would the next version of this course look like?

So if I was going to design a new React Native course, what would it look like? I'll tell you, it wouldn't look like the course is there. I was talking about this the other day and I ended up writing down like a rough schedule, and this is based on my experience of what can be taught in a course, what is interesting, how I use AI and what can be automated. And in the past, most of the course has actually done me live coding the whole experience and then talking through the fallacies. But because day-to-day live coding character by character isn't what we do anymore, we can spend more time on the higher level aspects of it, on the architecture, on the thinking, on the actual interesting parts of design, finding solutions, getting things to the app store. Comparing this to how the existing courses run, the percentages make it quite interesting. So, in the existing courses, 80% of the time is spent just coding things that we're not focusing on anymore as much, because you can offload it. Instead, we are focusing more, we can balance the course and we can balance the experience to spread it out to also focus on design and getting things to the app store. In particular, I would like to draw your eye to this line. That was a bit of a revelation to me. Previously, when you do a course, the course is the whole sum of the knowledge that you are getting. You're doing the course, and then this is what you've learned. Whereas now, that is less relevant, because if I did a write-up of a course, you could just read it with your AI agent, and that's not that valuable. So, really, the valuable thing is for me to give you the places where you could find solutions for yourself, solutions that I don't even know that you need. So, for the expert, what is going to help them help themselves? It's literally giving them access to the tools that you need.

So, for Expo, you really, truly need only two, in my opinion. And the great thing is that you can set them up and as we improve them over time, you get all the updates. The first thing is, of course, skills. And I personally love skills because of how simple they are. It's all about branding. I'm convinced that the person who came up with skills is the same person who came up with the air fryer. Air fryer is just a tiny oven. Skills are literally just documentation. They're markdown files of documentation on how to do a specific thing, written so that an AI is slightly easier to consume it, but you could read it, too. So, at Expo, we have skills on various topics that users often get wrong, such as upgrading Expo, adding app clips and best practices of mobile and web development, using API routes, etc.

And the second thing is the Expo MCP. So, what an MCP is, it's like skills, but it is tailored. It is built by a specific company to interact with their tools. So, the MCP is useful if you're using EIS, but also it has general information about telling you how to learn to learn. So, using the Expo MCP, you can search documentation, read documentation, add libraries, but then also view things on EIS. So, I'll give you an example of what the Expo MCP could do. So, this is a build that I made yesterday and, oh no, it's failing. So, what I can do is I can literally just, in plain text, paste in the URL and say, why did the build error. And because I have the Expo MCP installed, it does it so behind the scenes. You don't even know that it's doing it. So, that's the pit of success for you. But you can see it's calling the correct plugin and it knows how to see my build. It knows how to analyze the build and it actually correctly identified that I had pinned React Native to an unsupported version.

And this brings us to the veterans of the industry. Where does this leave you? Well, what happened to painting? Is it dead? Well, painting didn't end, but it was freed from the job of just creating a likeness, and this is where impressionist comes from. The boring mechanical parts of the work got automated and then the human part became the differentiator. So, the camera didn't replace painters. It replaced the part of painting that was never the point. It's the same way with AI. AI is not going to replace development.

It's going to replace the tedious part of developers. So, your job is not in danger if you code. Sorry. So, your job is only in danger if you code without thinking. The same way that the camera would have ended the career of mediocre portrait painters whose only value was creating a likeness. Your job has irrevocably changed. You're an architect now. You're a manager, an overseer, a source of wisdom, and the one who remembers to question when others follow without thought. It's not a loss. It's a gain. And you can now do less with more, freed from the tedium and repetitive code.

Don't forget to build your next app with Expo. Thank you.

So, the first question we've got is, do you have any advice for how individuals can overcome the Dunning-Kruger effect? How they could overcome it? I think it's self-awareness, isn't it? I mean, psychology is really interesting because in all areas of psychology, being aware of your fallacies does not make you immune to them, which is interesting, right? But I think thinking about that constantly really helps. And actually, one way of understanding that you don't understand something is to try to explain it. Because you might feel like you have a surface level understanding on a specific subject. But if you're talking about, let's say, oh, I press, you know, the analogy of I press this button and it works. You know, I press this button, like, um, I, the purchase happens. Try to explain what, like, the steps that the button actually goes through to make the purchase, to make sure that you understand what happened. And all the time you'll be like, oh, actually, there is a bit that I can't explain to someone else. Therefore, I don't really understand it. That's really nice.

From the perspective of veterans, we've got the new wave of junior engineers and the people who are just coming into software engineering. They have all these tools. They have AI now that can help them learn. Yeah. They're not going through the same thing that people had to go through 10 years ago in order to understand how a technology works. How do you think the veterans can be better teachers as a result of this? Because there is a divide in, in how they've learned and what they have to learn. Do you have any advice for that? How they could be better teachers. I think the first thing is that as a veteran, you almost have this temptation to be a bit miffed, I must say. It's just like, oh, man, I spent all these years, you know, painstakingly balancing binary trees and then, you know, AI is going to do it at this double hat. But I think it's important to recognize that the knowledge that you have is going to set you apart as the expert. So actually, so for me, in terms of photography, the photography was actually invented by painters, so by artists. And the most successful photographers were people who were prior previously artists. So I think I wasn't very clear in my analogy with that. But basically, the fact that you, you know, the painters know what they're doing, the painters know compositions, they know anatomy, they know what the lighting, they know what looks good. And then that set them apart to being able to be the best photographers, exactly the same with developers. So because, you know, you already have all this knowledge about architecture, and how things fit together, and importantly, what's the what the security concerns are, like you are in the best position to like, be the most successful in this area. So how it is translated to teaching newbies, I think it is to, you know, the box with the unknown unknowns, things that people like try to make that super clear. So the only way you can shift an unknown unknown into a known unknown is to talk about it. So think about the things that people who haven't been in the weeds of the code would never have thought of. And I just be like, by the way, make sure you think of security. That's already like shifting it up to the known unknown.

And then that's the best you can do. Yeah, I love that. It comes back to your idea of introducing solutions earlier in the learning process. That's great.

How can we train experts today? How do you train? How can we train experts? I guess, how can experts learn? I think experts need to relearn how to learn in some ways. Because the methods of learning the methods of doing most things that we've had in the past are less efficient now. So it's in a way that we have to learn from the newbies how to learn, right. And then we can we can be good again.

Do you think newbies should still be convinced to code by themselves to have a better understanding of what they're doing? And if so, how do you motivate them to do that? That's a really tricky question. I think that's something that we need to figure out by asking people who are new to code because me with my experience of already knowing how to code, it's always going to be different. Yeah, no, sure. And if I think back to when I was learning to code nearly a decade ago, like it was almost a bit of a rite of passage to take down production to do something that was a big mistake. And maybe we're not giving newbies the opportunities to make mistakes, because we've almost got this like, higher quality bar and expecting more of them from the beginning because of AI and because of all of these developments.

And another question is, do you think newbies should still be convinced to code by themselves to have a better understanding of what they're doing? And if so, how do you motivate them to do that? That's a really tricky question. I think that's something that we need to figure out by asking people who are new to code because me with my experience of already knowing how to code, it's always going to be different. Yeah, no, sure. And if I think back to when I was learning to code nearly a decade ago, like it was almost a bit of a rite of passage to take down production to do something that was a big mistake. And maybe we're not giving newbies the opportunities to make mistakes, because we've almost got this like, higher quality bar and expecting more of them from the beginning because of AI and because of all of these developments. Yeah, exactly.

Do you have any advice for players who are new to the development field in the recent age of AI? Remember, the matrix, the matrix of knowledge based on understanding. So this is not a new thing for me. I've been thinking about this for like years and years and years. The fact that there are things that you don't know that you don't know. And then in development, try to make that box for yourself as small as possible. So question everything. And you can use AI to do this as well. So, you know, as you're learning, rather than just taking everything from face value, just question everything. Why? How? If you were a senior engineer, how would you review this? If you were a senior engineer? Does this make sense? Could you break this? Is this a security risk? Yeah, I love that. Because you can use that to your advantage and almost ask the questions that you might be afraid to ask a senior engineer, and like not have the fear of embarrassment for asking why, how to centre a div. Silly, those like early, early questions. Yeah, no, I had, like recently, I had an opportunity to mentor this, this person, who was like a career changer, she was like changing into AI. And this was the person that I worked with very closely, who was like completely new. And I just learned to code. And actually, it was really powerful, because I just gave her like general guys, like general keywords, I was like use TypeScript instead of JavaScript. And she was like, say less, I'll figure it out. You know, and then she would just go away, figure things out. And then she what she would do is like she would, would have these calls every week. And then she would write down the things that didn't make sense to her. And then we just talk through those. And these were like, really, like things that AI would just have vague answers to, or like she had so many gaps in her knowledge that I needed to like go deeper to explain something. But AI is just an incredibly powerful tool for newbies, so you should use it. Yeah, fantastic. Thank you so much.

That's all we've got time for. There are a couple more questions for you. But I suggest that people go and find you in the extra booth. Yes, and go and ask those questions. Thank you so much. Thank you.