Why AI Could Make Apps Disappear

Welcome. I think it's time to question everything, and I'm probably going to piss off some programmers with this one, because I think software is already dying. Not computers, not technology, not even necessarily the companies that make software. I mean, something much more specific software as something humans have to operate. Applications, menus, buttons, workflows, training, subscriptions, and eventually maybe even the code itself.

Because here's the question I can't get out of my head. If artificial intelligence eventually understands what I want. Why should I have to understand the software? And if artificial intelligence eventually writes, tests, repairs, and optimizes the code, why should the code even be written? For me to understand? That's where I think this gets uncomfortable, and maybe a little bit weird.

In the last memo, I talked about series. And there's one line from that story I want to carry into this one. The catalog wasn't the business. The catalog was just the interface. People didn't actually want a catalog. They wanted stuff. The catalog was simply the best technology available at the time for connecting human desire to an outcome. Then the internet showed up and the interface changed.

The need didn't. So now look at your computer. Look at your phone. Look at all those little squares on the screen. Applications. What if the application is just the catalog? For basically my entire career? Software has worked in the same way. I have a problem. Somebody has built a tool that can help me solve that problem. And now I have to learn that tool.

Want to manipulate photograph? Learn Photoshop? Want to analyze a bunch of numbers? Learn Excel? Want to make a presentation? Learn PowerPoint. Want to edit a video? Learn editing software. And these are incredible tools. I'm not shitting on Photoshop or Excel. Excel may be one of the most useful pieces of software ever created. That's actually what makes this very interesting, because the bridge doesn't have to suck to be a bridge.

Sometimes people spend 20 years becoming absolute experts at using it, and then the mechanism changes. Take Photoshop. What do I actually want to wake up thinking, man. I really hope to use the selection tool today. Oh no. I want something changed. Remove the background. Move that person. Fix the lighting. Make this larger. Change that color. Put this product on a white background.

Make this look like it was shot at sunset. That's the outcome. Photoshop is the mechanism I currently use to get there, and for a long time that mechanism required skill. Real skill. Yet understand layers, masks, selections, colors, tools, shortcut shortcuts. And if you were really good at it, you could accomplish things that other people couldn't. And that skill had value because the capability was scarce.

Then I shows up and increasingly I can say remove the background and it does it. Not perfectly, not every time, but again. I'm not interested only in where this technology is today. I'm interested in the direction because if the direction is better, faster, cheaper, easier, then at some point the question stops being how good is I at Photoshop and becomes why did I even open Photoshop?

Excel's another great example. Excel is unbelievable. People have built entire companies inside spreadsheets, budgets, forecasting, inventory analysis, databases that definitely should not be databases. And for decades, knowing Excel has been a legitimate professional skill. Formulas, pivot tables, macros, Vlookup, xlookup whatever black magic you excel people are doing over there. But again, what does the person actually want? They don't want a spreadsheet.

They want the answer. Which products are losing money? Where did expenses jump? Forecasts the next quarter? Show me weird transactions. Which stores are underperforming? Turn this mess into something I can understand today. You may still use Excel to get there tomorrow. Maybe you hand intelligence to data and ask the question. If that happens, Excel can exist underneath everything, but your relationship with it changes.

It becomes infrastructure, invisible, something the machine uses instead of something you spend six hours learning and operate. And that's a distinction I think people are missing when I say software is dying. I'm not necessarily saying software disappears. I think software may become more like electricity. There's an unbelievable amount of complexity behind the light switch in your house power plants, transmission lines, substations, transformers, markets, control systems, wiring engineering.

But when you walk into the room, you don't give a shit. Let's switch light. That's the experience. The complexity is still there. It just became invisible. And I think software is headed in that direction. You don't say open this program. Go into this menu, export this file, import it over there, apply this formula and then send to me.

You say figure out why sales fell last month. You don't say open Photoshop, create a mess, clean up the edge and replace the background. You say make this look professional. You don't say build the spreadsheet with these seven formulas. You say, tell me whether I can afford this. The human moves farther and farther away from the mechanism and closer and closer to the outcome.

That alone threatens the traditional application. But I don't think that's the controversial part. The controversial part is what happens to the people who build the applications. There's a phrase everybody loves right now vibe coding. And I've done it. I built some pretty large applications using AI, although at this point I don't even know I'd call what I'm doing vibe coding.

I connect AI directly to the code base, so I basically become the orchestrator. I tell it build this, that broke go figure out why I fix it and I'll test it. Look at the security cleaning this up. Change the interface. This isn't what programing look like for the last 30 years. I'm increasingly not writing the code. I'm directing the thing that writes the code.

And the improvement has been ridiculously fast. A few years ago. I can build your website. Great. And then I produce some shitty little website that look like it was built by a toaster, right? But today it absolutely doesn't suck in the same way. And this is improving incredibly quickly. And this is where experienced programmers come in and they look at a generated code and say, this code is shit.

And you know what? Sometimes they're absolutely right. It might be inefficient. It might have security problems. Maybe it isn't committed correctly. Maybe the architecture is terrible. Maybe it did something in a way no sane developer would ever do it. Maybe it looks like somebody dumped spaghetti in the GitHub. Fine. I get it. Because if you went to school for 4 or 6 years and then spent ten, 15, 20 years writing software, you learned what code is supposed to look like.

These are conventions, patterns, architecture, documentation, security practices, efficiency, maintainability. And there's a very good reason for all of that. Humans need to understand the code that a developer leaves. Another developer comes in. They have to be able to look at the thing and figure out what the hell is going on. That's programing. That's reality. Today. But this show is about bridges.

So I have to ask the uncomfortable question why does the code need to be human readable? Seriously, why programing languages themselves are interfaces. Think about what we did. Computers operate at a level humans absolutely do not want to deal with. So we kept building bridges, machine code assembly, higher level programing languages, frameworks, libraries, integrated development environments, low code, no code and now natural language.

Look at the direction over and over and over. Humans have been moving farther away from speaking the machines language, and the machine has been moving closer to understanding ours. So why do we automatically assume that today's programing language is the destination? Maybe Python is just a bridge. Maybe JavaScript is just a bridge. Maybe C plus. Plus is a bridge.

Maybe the entire idea that a human being should open a code editor, read thousands of lines, understand them, change them, and maintain them is just the bridge. Now imagine where this goes. One AI creates the software, another AI reviews it, another attacks it. Looking for voting vulnerabilities, another optimizes another test performance, and another runs the application and looks for failures.

Something breaks. AI finds the problem, repairs it, tests the repair, deploys it, and eventually maybe the software underneath all of that doesn't look anything like something a human programmer would choose to write. Because why would it? Maybe a machine develops a representation that is faster, smaller, more efficient, and completely incomprehensible to human eyes, and another machine understands it perfectly at that point.

Why would I care? Most people have absolutely no clue what the engine in their car looks like. They don't care where the spark plugs are. They don't care how the fuel injection software works. They don't care what sequence the transmission control uses. They get in. Start the thing, put it in, drive and go. The outcome is take me somewhere.

And increasingly, they don't even need to understand the driving part. The mechanism keeps disappearing. So why would software be any different? Why would it be immune? The customer doesn't give a shit if your code is beautiful. They want the product to be fast, reliable, secure, and they want it to do what they need. If machines eventually become better at achieving those outcomes, then beautiful human readable code may be something incredibly strange, a limitation we preserve because programmers are accustomed to reading it.

Think about that. We could actually reach a point where forcing AI to write software the way humans prefer makes the software worse. I don't know if we get there, but I think it's a completely reasonable bridge to question. And if you're a programmer, this is the part where I would pay attention. I'm not saying that because I want programmers to disappear.

I've spent more than 30 years in technology. These are my people. You know that. That's exactly why I'm looking at it. Because the most dangerous bridge might be the one you've spent your entire career becoming really good at crossing. Think about that phrase we've heard for the last few years. AI isn't going to take your job. Someone who knows how to use AI.

Well, I've said it myself. And for right now, I think there's a lot of truth to it. But what if that is also a bridge? First, the programmer writes the software, then the programmer uses AI to write more software. Then the programmer becomes the orchestrator of AI. And after that, I don't know. But you better be asking a question, because your value may move farther and farther away from I don't know how to write code in towards.

I know what needs to be built. I understand the problem. I understand the architecture. I can tell when the answer is wrong. I understand security, I understand the business, I understand the consequences. I know what questions to ask because if the cost of how collapses, then knowing what and why becomes more important. And there's another side to vibe coding.

And I think almost nobody wants to talk about it. Everybody right now is saying you can build an app. Cool. So can everybody else. And this may be one of the strangest traps in the entire AI transition. I could make it possible for almost anybody to become a software developer at exactly the same moment. Software development stops being scarce.

That's important because scarcity creates value. If if it once took 20 developers two years and $3 million to build some specialized piece of software, that that software at that point had enormous economic value, partly because it was difficult to create. Think like Facebook. But what happens when one person and I can build something similar in a weekend? Great software just became abundant, but the abundance has a consequence.

The thing itself becomes cheaper. So before you spend two years building the greatest vibe card application anybody's ever seen, I think you need to ask a more important question. What the fuck are you building? Technically, conceptually, what problem does it solve? And more importantly, will that problem still exist by the time you finish it? Right, because you could build an amazing bridge to a destination that nobody needs anymore.

That's the danger. Imagine building an entire application whose purposes? Organize this information. Move it over here. Transform it like this. Send it to another application. Create this report. Notify the person. That could be a fantastic product today. But what happens when I simply say do this and intelligence figures out every intermediate step? Your product didn't get beaten by a better application.

The need for the application disappeared. That is a much more dangerous competitor. And now let's take it one step further. If software creation becomes abundant. If images become abundant, if video becomes abundant, if music becomes abundant, if intelligence itself becomes increasingly abundant, then where does the economic value go? Because it doesn't vanish. It moves. And I think this is where the next decade starts getting really interesting.

Digital capability may become incredibly cheap, but there are things I cannot simply prompt into existence electricity, compute, silicon factories, networks, robots, raw materials, land, physical infrastructure, machines that can actually touch the real world. Software can become nearly infinite. The physical world cannot. And that means the center of gravity may be moving for the last few decades, some of the most valuable companies on Earth were built around software because software was incredibly scalable.

Build it once, sell it millions of times. Beautiful business. But if software itself starts becoming cheap, what then becomes scarce? The infrastructure underneath the intelligence, the chips that run it, the energy that powers it, the networks that connect to it, the machines that execute what it decides, the water that cools everything down, and the companies capable of joining all those pieces together.

Remember that, because we're going to come back to it a lot in this series. This is why I don't think the revolution is ultimately about chatbots. I don't think it's ultimately about who has the coolest AI button inside their app. That's the obvious part. The much bigger question is what happens when intelligence becomes a commodity? What happens when software becomes abundant?

What happens when digital creation costs almost nothing? Because then the really valuable pieces, maybe the things intelligence still needs, but they cannot create by typing a prompt. Energy. Compute. Connectivity. Manufacturing. Physical execution. Scarcity. And if you follow that trail far enough, you start finding some very interesting companies, some very interesting people sitting at the other end of it.

Let's get there.

But first, connectivity, because every AI, every device, every robot, every autonomous machine, every piece of invisible software has a problem. It needs to connect. And right now, we're still carrying around devices we call cell phones. Think about that name. Cell phone. How much of what you do on your phone is actually a phone call? You watch video, send messages, use AI, navigate work, listen to music, upload photos, make video calls, browse the internet and almost everything you actually value is data.

So if the phone really isn't a phone anymore, I think we need to ask another uncomfortable question. Why does it still need a traditional cell network? And what happens when the sky becomes the network?

I don't know where this bridge ends, but I think it's worth finding out. Goodbye.