(AI) Bubble Trouble

Summary

Paul Ford and Rich Ziade explore the duality of the AI moment: the tools are genuinely better than ever, yet the industry may be overheating into bubble territory. The episode begins with a listener question from Christopher asking what Aboard means when they say they’re “not vibe coding.” Rich explains that vibe coding tools rush to produce code immediately, while Aboard takes a blueprint-first approach — gathering requirements through AI-guided conversation before assembling vetted code components, more like ordering off a menu than fast food.

The conversation pivots when Paul shares a transformative personal experience: taking a synthesizer-building class at the School of Visual Arts in NYC. Using Claude to help him write C++ code for a Teensy microcontroller, Paul went from soldering buttons onto a PCB board to building a drum machine, creating a Lua-based operating system for his hardware, and conceptualizing a complete groove box — all in about a half hour of AI-assisted coding. He describes doing five things he’s never done before without having known about them beforehand, illustrating how AI dramatically reduces friction for experienced technologists to explore adjacent domains.

Rich then draws the crucial distinction: people with existing technical knowledge are getting rocket boosters from AI, but the promise that anyone can just say “make me a thing” and get it remains far away. The bubble, they argue, is in the expectation that the same box people use to order pad thai will help them refinance their mortgage. The real transformation is in reducing friction for skilled people to do new things — but that is not the trillion-dollar consumer revolution investors are betting on.

Highlights

”I did five things I’ve never done before in a half hour”

yt-dlp --download-sections "*11:55-12:50" "https://www.youtube.com/watch?v=g0b1vZ-qG5I" --force-keyframes-at-cuts --merge-output-format mp4 -o "g0b1vZ-qG5I-11m55s.mp4"

“I did like five things I’ve never done before. I did them without thinking about them, having known about them. Soldering was one that was physical. But the brain stuff — I’m going to build an operating system for this one particular piece of hardware, I’m going to couple it tightly to the hardware that I just made, and build a way to quickly load multiple modules.” — Paul Ford, 11:55

”Experienced technologists get rocket boosters”

yt-dlp --download-sections "*13:50-14:45" "https://www.youtube.com/watch?v=g0b1vZ-qG5I" --force-keyframes-at-cuts --merge-output-format mp4 -o "g0b1vZ-qG5I-13m50s.mp4"

“This is a weird moment because the old people have an absolutely exponential advantage. If you basically understand what technology is and how it works, your access to like 50 zillion other things in the world — the cost has gone down and is approaching zero.” — Rich Ziade, 13:50

”The pad thai box is the bubble”

yt-dlp --download-sections "*17:05-17:55" "https://www.youtube.com/watch?v=g0b1vZ-qG5I" --force-keyframes-at-cuts --merge-output-format mp4 -o "g0b1vZ-qG5I-17m05s.mp4"

“Everyone is thinking the same box I use to order pad thai is going to be the box that helps me refinance my mortgage. That is the bubble. That’s the promise. And that is the bubble.” — Paul Ford and Rich Ziade, 17:05

”All this convenience leads to depression”

yt-dlp --download-sections "*21:30-22:20" "https://www.youtube.com/watch?v=g0b1vZ-qG5I" --force-keyframes-at-cuts --merge-output-format mp4 -o "g0b1vZ-qG5I-21m30s.mp4"

“All this convenience I think just leads to depression. You on episode nine with pad thai noodles dripping on your shirt because everything just sort of comes to you. There’s such a gift on the other side of struggle. If everything is too convenient — how do you feel good about what’s on the other side of something you achieved?” — Rich Ziade, 21:30

”We’re not vibe coding — we’re blueprinting”

yt-dlp --download-sections "*3:10-4:00" "https://www.youtube.com/watch?v=g0b1vZ-qG5I" --force-keyframes-at-cuts --merge-output-format mp4 -o "g0b1vZ-qG5I-3m10s.mp4"

“What we do is we don’t rush to building the thing. We draft a blueprint. The conversation you have with Aboard for the first 5 minutes are pretty much filling in the details that go into a blueprint. And that is very different than typical vibe coding tools, which love to go to build as quickly as possible.” — Rich Ziade, 3:10

Key Points

• Vibe coding vs Aboard (1:05) - Vibe coding rushes to code artifacts; Aboard gathers requirements first through guided conversation, then assembles from vetted code components
• Spec-driven development (3:55) - Vibe coding products are now adopting “spec-driven development” — writing specifications before code — which Aboard has always done
• Home Depot analogy (1:29) - Vibe coding is like buying wood and nails to make a shed; Aboard is for building a wing to the house, which requires a blueprint
• Mr. Bubble and the AI bubble (5:00) - The hosts agree we’re in a bubble where the market can’t only go up; if AI replaces all jobs, no one has money to pay for the products
• Synth class at SVA (7:00) - Paul takes a 6-session, 3-hours-each synthesizer building class at the School of Visual Arts in NYC
• Teensy microcontroller (8:50) - The hardware platform uses a Teensy running in the Arduino ecosystem, programmed in C++
• Vibe coding a drum machine (10:00) - Paul used Claude to vibe code a C++ drum machine for the Teensy, doing two types of programming he’d never done before
• Lua operating system layer (10:30) - Paul built a Lua-based operating system to load different modules (drum machine, chord generator) on the same hardware
• Five new things in a half hour (11:55) - Soldering, embedded programming, C++, operating system design, and hardware-software integration
• Hardware manufacturing access (13:00) - Paul estimates he’s only 2-3 steps away from being able to design, manufacture, and ship hardware from China
• Skilled people benefit most (13:50) - Experienced technologists get exponential advantage; “the cost of accessing 50 zillion other things has gone down to approaching zero”
• The real bubble (17:05) - The bubble is the expectation that the same chatbot for ordering pad thai will handle mortgage refinancing
• The non-bubble (17:30) - Things that used to have enormous friction in the economy but still take work now have much less friction — that’s the real transformation
• Education message (18:30) - Group-based learning, project-based learning, and applied learning matter more than “intro to AI” classes
• Convenience vs. depression (21:30) - Too much convenience leads to depression; the gift is on the other side of struggle

Mentions

Companies

• Aboard (1:05) - The hosts’ company; uses AI for guided requirements gathering before building software
• Replit (2:20) - Vibe coding tool that will debug code for hours and charge you for it
• School of Visual Arts (SVA) (7:00) - NYC school where Paul takes his synth-building class

Products & Technologies

• Claude (10:00) - Paul used Claude to rewrite and vibe code C++ for the microcontroller
• Teensy (8:50) - Microcontroller running in the Arduino ecosystem, hundreds of megahertz but limited memory
• Arduino (8:50) - The ecosystem the Teensy microcontroller runs in
• MIDI (9:30) - 30-40 year old digital music standard, still heavily in use today
• Lua (10:30) - Tiny scripting language that fits in 8K of memory, used for the microcontroller operating system
• C++ (9:50) - Programming language used for the synth hardware; lower level than Paul’s usual scripting languages
• Visual Studio Code (9:55) - IDE used for programming the microcontroller

Surprising Quotes

“If the fantasies all come true about how valuable this is, no one will have any jobs or money with which to pay for the products.” — Paul Ford, 5:40

“I’m no longer scared of creating and shipping hardware.” — Paul Ford, 22:30

“It’s born as a consumer tool. ChatGPT 3.5. Anybody can use this and it will help you with your homework. And I don’t necessarily think it’s that. I think it’s this sort of weird infrastructural dynamic change machine.” — Paul Ford, 17:00

“It just extrudes strudel all day long.” — Paul Ford, 22:00

“My brother once tried to bring a tube television — he leaned it on the edge of the bathtub. My mom walked in. He was forced to write a long essay on electrical safety.” — Paul Ford, 5:20

Transcript

0:00 Hi, I’m Paul Ford. And I’m Rich Ziade. This is the Aboard Podcast. The podcast about how AI is changing the world of software. And Rich, we got a listener question from Christopher. Okay. Hello Christopher. Thank you for listening. Let’s play a theme song and we’ll answer it.

0:38 Christopher had a good question. He wrote us and he said, “Listen, you guys keep saying you’re not vibe coding.” Yes. You’re not vibe coding. You’re not vibe coding. Great. Okay. I got it. I heard you. But maybe you could explain what the hell you mean by that. Sure. Thanks for the question, Christopher.

1:05 Vibe coding lets you type in a box and then it starts making software immediately. Focus on the second word: coding. It writes code that then runs as software. It’s about getting you to that artifact as quickly as possible. And it works for certain things. Just as I can go to Home Depot, buy some wood and nails and make a shed, I can make little things with vibe coding.

1:50 Software is quite complex and it becomes exponentially more complex if it’s a bigger thing. We build much more complex software with Aboard. It’s designed for building a wing to the house. You’re not just going to go to Home Depot, grab a bunch of stuff, and say, “Honey, we’re going to have an extra bedroom before you know it.” It doesn’t work.

2:20 Vibe coding is when you type in the box, you get some code, it edits the code, and you go around and around. A tool like Replit will debug your code for hours — and charge you for it. So, it’s very much about cutting engineering cycles out of the loop so you can get to your thing.

2:31 Our use of AI is different. We create a little conversation with you. We gather your requirements and use AI to extrapolate about what your business is and how it might operate. Let’s go back to my analogy. Do you need a blueprint for a shed? Ideally, it’s nice to have one, but you don’t need one. You do need a blueprint for a bedroom with a bathroom and pipes and electricity.

3:10 What we do is we don’t rush to building the thing. We draft a blueprint. The conversation you have with Aboard for the first 5 minutes are pretty much filling in the details that go into a blueprint. That is very different than typical vibe coding tools, which love to go to build as quickly as possible and then fix as best they can.

3:50 It’s a little bit more like you’re ordering off the menu and then they go cook it, as opposed to the fast food of vibe coding. And what you’re seeing amongst vibe coding products is something called spec-driven development — let’s write something first before we run the code. We’re much more diligent about that. The spec isn’t a casual step for us.

5:00 Remember how a couple weeks ago we were talking about how we’re in an AI bubble? Do you remember Mr. Bubble? He was great. You got in a tub and it was all bubbles. You don’t really take a lot of tubs as a grown-up. My kids take a lot of baths. Kids love baths. If you could just get the iPad in there, it would rule. They do.

5:30 Speaking of electrical safety, we concluded last time that we’re pretty much in a bubble. The market just can’t only go up. At some point, if the fantasies all come true about how valuable this is, no one will have any jobs or money with which to pay for the products. So I don’t think we’ve unlocked the trajectory.

6:30 I’m going to go two steps back. My midlife crisis hobby is learning piano and fooling around with synthesizers. Normal dad stuff. I take my piano lessons on Tuesday mornings and then at night I go beep boop around and practice my scales and my Bach pieces. Recently I saw on Craigslist somebody was like, “Hey, if you ever want to learn how to build a synthesizer, we’re teaching a class at the School of Visual Arts.”

7:00 I started going to the synth class. It’s six classes of three hours each. Pretty low commitment. My kids are a little older. I got a PCB board with — I had to solder buttons on it. I always had a block about soldering, so it was cool to actually get taught. I assembled my board. It works pretty well. I may go by for the laser training so I can make a case.

8:50 It has a microcontroller called a Teensy. It runs in the Arduino ecosystem. I have put a lot of little buttons and sliders and these are by nature organic — you hit them with your fingers and they make little pulses. They all go into the Teensy, which is hundreds of megahertz but it is a little tiny computer. Not a lot of memory. It takes those from analog button world into computer world — zeros and ones — and turns that into signals it sends over MIDI to a laptop or synthesizer.

10:00 I had Claude rewrite the code for me. I started to tidy stuff up. That’s normal vibe coding. Then I was like, what if I wanted to do a drum machine? So I vibe coded a C++ drum machine for the Teensy. Two kinds of programming I’ve never done before — embedded microcontroller programming and C++.

10:30 Then I go, maybe I’ll make an operating system layer. I start to realize I want a drum machine mode, 16 buttons, eight sliders, or I could go over here and have a chord generator. There’s a little tiny language called Lua — 8K in memory every time you instantiate it. So it fits nicely on the microcontroller. Let me skip to it. A half hour has gone by.

11:55 I did like five things I’ve never done before. I did them without thinking about them. Soldering was physical, but the brain stuff — building an operating system for this particular piece of hardware, coupling it tightly to the hardware I just made on the PCB board, building a way to quickly load multiple modules and switch between them to create a whole groove box experience. I could not have said any of that a week ago.

12:40 This is a wild level of acceleration. I could argue from first principles with a high degree of confidence that AI will absolutely accelerate humans doing things that they might not have been doing otherwise. If you look at China — all these machines that make things. That world is hard to access if you’re a web programmer. I really only have two or three steps left to learn before I could do a really bad job of making a piece of hardware and getting it shipped back from China.

13:50 This is a weird moment because the old people — skilled people — have an absolutely exponential advantage. You as a curious, thoughtful technologist who understands many high-level concepts just found rocket boosters. You understood the context to use them within. Nor is a sneering senior engineer telling you you could never understand. You can play and screw up and it’s not expensive.

15:00 I actually don’t think building a really good digital device that people would want to use and buy and marketing it — that’s about as hard as it used to be. But getting started is almost free. The promise of just saying “make me a microcontroller and a drum machine and handle the shipping to Asia” is far away.

17:00 It was born as a consumer tool. ChatGPT 3.5. Anybody can use this. I don’t necessarily think it’s that. I think it’s this weird infrastructural dynamic change machine. Everyone is thinking the same box I use to order pad thai is going to be the box that helps me refinance my mortgage. That is the bubble. That’s the promise. And that is the bubble.

17:30 There’s a secondary non-bubble where things that used to have enormous friction in the economy but still take a lot of work just have so much less friction. That slowly will bloom into everything being cheaper, things being domestically manufactured by more robots. But it’s not tomorrow.

18:30 If there’s a class called “intro to AI,” we already lost. Group-based learning, project-based learning, problem-solving, applied learning — so that high-level concepts get put through the ringer and young people are forced to find paths in ambiguous places. My ChatGPT is educating me about how to work within the limits of this tiny computer.

20:00 A lot of incentives exist to create solutions where people think less and get an outcome. If you’re thinking a little less and making transactions happen all day long — if you can skip steps but charge the same amount — that’s smart for business. This is ultimately a challenge for education and culture to think about these tools as empowering rather than convenient.

21:30 All this convenience I think just leads to depression. I think it leads to you on episode nine with pad thai noodles dripping on your shirt because everything just sort of comes to you. There’s such a gift on the other side of struggle. If everything is too convenient — how do you feel good about what’s on the other side of something you achieved?

22:30 It’s incredibly motivating to be able to skip a few steps and figure out what’s actually going on and then backfill that knowledge over time. It tries not to make you frustrated. But I’m no longer scared of creating and shipping hardware.

24:00 I’m looking at high schools for my son right now. We’re leaning a lot more towards project-based, where the tables are set up in circles so they can work together and learn through some applied way. We’ve got to burn our fingers here a little bit. I think it’s necessary and healthy.

25:00 Check us out at aboard.com. The product of a passion project. Probably the most challenging project I’ve ever been involved in. Kind of love it. We got engineers that are just some of the most talented I’ve ever worked with. Take a class too. Send an email to hello@aboard.com. Have a great week everybody.