Interview Intro

‍

Richard Rodger:  [0:00:00] Welcome to the Voxgig Podcast. We talk to people in the developer community about developer relations, public speaking and community events. For more details, visit voxgig.com/podcast. All right, let's get started. 

‍

Developer relations is not just for software. You can also have developer relations for hardware. In this episode I talk to Matthias Wagner, CEO of Flux.ai, which is a desktop tool for designing printed circuit boards. Matthias wants Flux.ai to be the Github.com of printed circuit boards, and he’s already pretty far along. 

‍

Let's talk to Matthias and find out more. [0:00:44]

‍

Main Interview

Matthias Wagner

‍

Richard Rodger:  [0:00:45] Matthias, welcome to the Fireside with Voxgig Podcast, and today we are talking about your company, Flux.ai. And you are a little bit different from our usual guest, because your stuff is to do with hardware. I’m not going to try and explain what you do – it looks pretty cool. I love your website design; it’s really awesome. But I’m going to hand over to you. Matthias, what does Flux.ai do? [0:01:09] 

‍

Matthias Wagner:  [0:01:11] Good question. First of all, thanks for having me. Great to be here with you, Richard. Flux.ai is an online design tool to design printed circuit boards, PCB boards, like the motherboard in your laptop or in your smartphone. We’re talking a modern approach; putting Figma in the browser; it’s collaborative. 

‍

We have AI functionality which makes our electronics design generative, that you can describe or brainstorm with an AI agent around things like your bill of materials. I have this favorite example where I say, ‘”Imagine you wanted to build a smart chocolate brownie oven.” And you can use the AI agent here to iterate what components would be required, what the tradeoffs are. 

‍

You can drill down on your requirements; be like, “This has to be battery powered; I want this to be on the go when I go camping.” Whatever – wherever you want to go with this. And it’s not just helping people who already design PCB boards to design these boards faster and cheaper. We’re also helping people who are designing their first PCB board, to make that- [0:02:14]

‍

Richard Rodger:  [0:02:15] Yeah. And is it – are you – do you aim primarily at the Kickstarter market or is it for professionals or is it both? [0:02:22] 

‍

Matthias Wagner:  [0:02:24] We target engineers; we want to give engineers better tools. And from that sense, we’re speaking to the individual and we’re not worried so much what the application is. But you can imagine here there is – on the professional side, it’s contractors. There’s small and medium businesses; there’s a lot of large Fortune 500 businesses using flux. We also have a lot of students, lot of hobbyists, lots of people coming from other fields. Like imagine software engineers coming in here and want to design a board or mechanical engineers by the building of robot. And they need a controller board or like a sensor board for something. [0:03:01] 

‍

Richard Rodger:  [0:03:04] I like that way you described it, Figma for PCB boards. Did you use that phrase for – with your investors? That sounds like a winning phrase. [0:03:15] 

‍

Matthias Wagner:  [0:03:18] I use it with people where I feel like they know what Figma is, but it turns out not everybody does, so, sometimes I say Google Docs as a replacement. And typically, I say the product is on the intersection of GitHub, Figma and Java, some ironman. 

It’s like GitHub if you think about this global community that has built this public repository of reasonable components. And it’s like Figma if you think about the design tool that’s collaborative and lives on the browser. And it’s like Java ironman; if you want to talk to an AI agent, they will help you build hardware. [0:03:49] 

‍

Richard Rodger:  [0:03:50] And is that important? I remember about 10 years ago exploring the idea of doing a startup in the IoT space, mostly around the cloud end of things and providing a database and that sort of stuff. It had come from a consulting project that I did for alarm systems. But I am not a proper engineer at all; I did mathematics. I’m only a coder. And I did get really into it, because it’s super fun, the soldering iron and burning out LEDs and that sort of stuff. [0:04:30] 

‍

Matthias Wagner:  [0:04:32] Manifesting things in physical space, right? [0:04:33] 

‍

Richard Rodger:  [0:04:33] Yeah, exactly. Do you find that there’s a difference between the engineers that come to you and – if people are coders, they have certain expectations around what their tooling is going to do. Is your stuff inspired by that, or is – does it come more from an engineering philosophy? [0:05:00] 

‍

Matthias Wagner:  [0:05:02] If you think about – it’s definitely inspired by software engineering. A big a-ha moment I had, which led to starting the company, was that software engineering had come so far in my lifetime; it’s been truly democratized as a craft. And that in contrast, hardware design felt like has not moved at all since I first opened a CAD tool, an electronics design tool in the ‘90s. There was clearly something broken, it felt. 

‍

And then there were people that kept sitting round saying, “Hardware is hard.” And I was like, “Well, software isn’t exactly easy. It's just we’ve made it easy.” We’re not punching cards anymore. If I had to build a web app by punching cards, it’d be just as hard as making an IoT PCB board, or worse, frankly. So, started to look into the existing tooling and contrasting that with software development, where in software development, we have fast iteration cycles. 

‍

Why do we have fast iteration cycles or fast feedback loops? It’s because we have modularized almost all of it. If I were to encrypt something, I’m not going to spend the next six months finding an encryption library. I’m going to go to a place like GitHub, where there’s a competition of encryption libraries for different purposes, for different strengths and weaknesses, and just grab one from there. And then contrast that with hardware, if I needed an amplifier, I’d be building it from scratch, right? [0:06:40] 

‍

Richard Rodger:  [0:06:40] Yeah, you would. And the reason I bring up my own experiences from 10 years ago – they didn’t last very long – is, even though I was really enthusiastic about it and I bought all the books, starting from scratch, the very basics of working out your voltages and all that sort of stuff – there didn’t seem to be much of – like you said – reusable. You had to burn out the LEDs yourself. [0:07:08] 

‍

Matthias Wagner:  [0:07:11] Exactly. 

‍

Richard Rodger:  [0:07:13] It sounds like you want to be the – you want to position yourselves like GitHub maybe. [0:07:17] 

‍

Matthias Wagner:  [0:07:19] Yeah, totally. It is investing in these three pillars:  GitHub, Figma and Javas. We want to – we’re taking the transferable ideas from all these three and turning them into one product. [0:07:31] 

‍

Richard Rodger:  [0:07:33] Let’s take a closer look at that then. In GitHub I can share my code and I put it under a specific license, like MIT or BSD or whatever, that people can use. And that’s important, because you can share and use that commercially. Do you guys support the same things? If I’m – if I want to have a little circuit, can I say this is free to use or MIT? How does this work in your world? [0:07:55] 

‍

Matthias Wagner:  [0:07:58] This is exactly how it works. On Flux site, everything is a project, and projects can be infinitely lesser into each other. And you can have that. The amplifier’s talked about as the project and you can design that. You have others – you can others – others can review that; others can contribute improvements. Others can go in forklifts and make their own variant of it that’s more specialized for one over the other application, let’s say, or cost optimization or whatever you’re going for. 

‍

And somebody can take that amplifier and turn that into an audio amplifier. And then the next person, they want to build a room, a vacuum kind of thing, where they need a sound model. And  than they’re going to use the audio amplifier build on top of your amplifier in there. And so, you create these dependency chains, just like you have on GitHub, where one project you pull in depends on five other projects. [0:08:50] 

‍

Richard Rodger:  [0:08:52] Yeah, so the engineers, your relationships with the engineers is central to the company, building that community. So, have you done – have you taken much of an approach to building the community or have you focused on the product first? [0:09:10] 

‍

Matthias Wagner:  [0:09:16] If you contrast what we have been doing to, say, building GitHub. When Tom Preston-Werner, who’s one of our investors, built GitHub, they didn’t have to build Git. They had that piece and they built the hub around that, the community around that. And our case, none of the building blocks existed in a way that we could recombine them, so we had to build everything. [0:09:36] 

‍

Richard Rodger:  [0:09:36] You had to build everything, yes. [0:09:38] 

‍

Matthias Wagner:  [0:09:38] Yes. And so, it was clear from the outset that the vision was grand here, and that it would take a long time. And that we somehow had to figure out a way here, because we needed feedback to build this product, that the product wouldn’t be useful for a long time. We had a build a community of people who believed in the same vision and were willing to deal with how crude it would be for a long time, so that someday, we’ll get to that promised land. 

‍

And that’s always how we communicated this to people. It’s like, “Look, this is a very long road. And we’re here day one, and this is very far away from where we want to go. We offer you free to join us along this long path.” And something we did early on, probably 2-3 months into the company being founded, we sent out a biweekly – we called it – back in the day we called it the change log. 

‍

We keep people updated on where we were at, what we were working on, what we shipped that week. Not with the intent that they would come back and try it out, whatever, although that was great that some people did and provided feedback. But that we would keep people informed that this is moving, that we were committed to make progress over long amount of time, and we were committed. 

‍

Another thing we did then, we launched a public beta really early. The company was founded October 2019 and then in summer 2020 we shipped a public beta, and we kept that running all the way to the public launch. And we had then – we experimented with a lot of alternative channels than email. We tried Discord, a Facebook group, Slack. Eventually we arrowed down on Slack, because we thought – we were spread too thin too early, across too many channels, and then we – the thinking we had is we wanted to go where users already are. 

‍

And at the time, at least, in hardware, there wasn’t many people using Discord or Facebook groups. But everybody had a Slack at work and so, it was easy to switch from their to our Slack. And that worked really well. And then this started really small, with 5-10 people, but if you look at the curve today, it’s flat for years like this. Then you – with your public launch and then you hit that inflection point. And that is – [0:12:03] 

‍

Richard Rodger:  [0:12:03] That’s the SaaS curve; isn’t it? It’s always flat for a long time. [0:12:07] 

‍

Matthias Wagner:  [0:12:07] Yeah, totally. It’s the overnight success a decade in the making – that’s a joke. But we had a strong conviction that this is what we wanted and what other people wanted. And then that helped bridge the gap of keeping at it for years and years, to get critical mass into the functionality of the product, that it was now ready to be launched. And in fact, I would say that even today, we’re just day one here; we’re nowhere close to what we want to deliver. That said, it’s probably late morning on day one, and you can build lots of categories of products with Flux today, and lots of people do; it’s been really exciting. [0:12:48] 

‍

Richard Rodger:  [0:12:49] What you have at the moment is usable, and it’s usable for production stuff. And it’s quite- [0:12:54] 

‍

Matthias Wagner:  [0:12:54] It’s all-

‍

Richard Rodger:  [0:12:55] And it’s quite impressive, if you go play around with it. And I encourage- [0:12:57] 

‍

Matthias Wagner:  [0:12:57] I always tell people. But the dream that we have is that we want to enable a 12-year-old to build their own iPhone. Today you can’t quite build your own iPhone, but you can pretty much build everything else. These high-density, high-skill applications, or some stuff where – let’s say a controller for a manned rocket – that you can’t do today. But we’re working towards that too, to capture this use case too. [0:13:26] 

‍

Richard Rodger:  [0:13:27] Let me know when you’re doing rockets. Maybe we need to talk again. You found – that’s interesting that you mentioned Slack. There’s a lot of discussion in the developer relations community about what tools one should use for online communities. You have your Discords and Slack, various other tools. 

‍

And there’s – you can find quite a few blog posts where people say, “You should use Discord everything else is a mistake,” or “You should use Slack and everything else is a mistake.” But in your case, it sounds like Slack was a good choice, because your users were coming from corporate environments and corporate access to Slack is usually okay. Whereas Discord is seen as a gaming site, so it’s often blocked by the corporates. [0:14:17] 

‍

Matthias Wagner:  [0:14:17] Yeah. And that’s been changing, but what I tell other founders, is go where the users already are. Slack isn’t the perfect tool to build community, not by any means. But we want a free account too, because it would be way too expensive to run a paid account, because the whole pricing- [0:14:34] 

‍

Richard Rodger:  [0:14:34] Yeah, this is-

‍

Matthias Wagner:  [0:14:35] – model isn’t – it’s cumbersome, but this is where users were, and that’s why it worked. And it was – would have been difficult at the time to push them over to Discord. And you can always expand later to other channels. [0:14:46] 

‍

Richard Rodger:  [0:14:47] Exactly. That drives me nuts about Slack, because there’s so many really successful, vibrant developer relations communities on Slack, but everything disappears after three months because nobody can afford to pay. Slack is missing out on a use case, right? [0:15:04] 

‍

Matthias Wagner:  [0:15:06] I totally agree. 

‍

Richard Rodger:  [0:15:07] It makes sense for a company, but not for a community. [0:15:09] 

‍

Matthias Wagner:  [0:15:11] Yeah. 

‍

Richard Rodger:  [0:15:12] And we have-

‍

Matthias Wagner:  [0:15:12] Discord is eating their lunch. The trend largely today is towards Discord because you have much more control; it’s affordable for a large community. [0:15:21] 

‍

Richard Rodger:  [0:15:23] It’s – I’m sure you would, and I know a lot of companies that would pay to have a community Slack. They would pay something, if there was some value there, just not per user. [0:15:35] 

‍

Matthias Wagner:  [0:15:35] Just not $5 per user, yes. That’s- [0:15:37] 

‍

Richard Rodger:  [0:15:37] Exactly. We’ve spoken on this podcast to quite a few guests about how they run their Slack communities, and maybe this applies to Discord as well. What model of community do you have? Is it the one where all of the users are in one place, or do you create sub-communities or do you create individual ones for big customers? What approach do you take? [0:16:00] 

‍

Matthias Wagner:  [0:16:02] It’s all in one big place right now. We have a bunch of different channels, from showcasing projects to requesting help to cool stuff people find on the internet. It’s a combination of product support, indication, talking about cool shit. Meeting others – I need help. I’m going to build a motor controller and I’ve never done this before. Now I need help with antenna design; who’s done this before? And just happen to connect people. 

‍

Our whole team is active there, so it’s also for us a great way to see what people are up to, what they struggle with. Also to go deep on – here’s an idea somebody had to change a feature or add a feature and just kick out on that. And that makes it also then easy for us to – let’s say it’s a dock thing. You can come back with the – I want to try this now. And you have this direct channel. Which otherwise, you have these messaging features in our dock ticketing platform, but it’s not as immediate and not as quick than in chat. [0:17:14] 

‍

Richard Rodger:  [0:17:15] And do you have a rota system for people to manage the community? Do you have a community manager? Or is it just casual and ad hoc, or do you have a system? [0:17:27] 

‍

Matthias Wagner:  [0:17:30] . We iterated for many different models over the years. How it works today is that we have somebody who is in charge of the community, and hey post. But we still post a change log every two weeks, so they post; they’ll just email it out. They also post it on Slack. 

‍

So, the engineer in charge of communication responds and – but then additionally, a lot of our team members hang out there. I have the tab always open to see what’s going on and reply to questions, also some weekends. We have also these days a lot of power users who are active there and help other people out, machine users. 

‍

And then we have these product ambassadors at Flux that we call. These are former hardware engineers that now work at Flux to help us design the product, but also help our users get on boarded and unstuck. And learn from them and listen, and have that flow back into the product in a more systematic way. And they all are hanging out there too. [0:18:32] 

‍

Richard Rodger:  [0:18:36] You guys are doing something really interesting. Get the CEO in the Slack channel, in the community, and that’s you. And you get so many people in developer relations asking, “How can I grow my community? How can I keep it healthy?” Have the CEO participate sometimes, because to the user, for me, that’s amazing. That’s – [0:19:00] 

‍

Matthias Wagner:  [0:19:00] Yeah. I fully agree. 

‍

Richard Rodger:  [0:19:00] This team believes in what they’re doing. [0:19:01]

‍

Matthias Wagner:  [0:19:03] Yeah, I totally agree. I always wish I’d more time for this, and I have that, where I have weeks where I open Slack and it’s – every channel is unread, and I’m like, “Oh my God.” But then- [0:19:15] 

‍

Richard Rodger:  [0:19:15] And you guys are not small. You’re funded; you have 50 people. You have a busy job; you have a proper CEO job, but you still participate, which is amazing. [0:19:24] 

‍

Matthias Wagner:  [0:19:25] I would say so. But it’s also with Slack, that makes it easy. It is a mobile app. I can do this on the couch, or when I’m in the cab or waiting for something somewhere. I do that as other people read the newspaper. And it’s also – I’m a geek and a nerd and I started a company because I am really excited about this, and so maybe it is my way of reading a newspaper. [0:19:51] 

‍

Richard Rodger:  [0:19:52] Yeah. What I like about this is – and we are talking about engineer relations, not developer relations, but they are very similar. But I’ve often spoken to other guests about the idea of the vibe of a company, the developer spirit, which is something hard to create. You don’t create it by having good documentation or lots of examples. It’s often created by the leadership having that spirit already. And it sounds like you’ve captured that for engineers as well. 

‍

If I’m an engineer and I join your Slack channel, there’s Matthias and he’s answering my questions about circuit boards. I understand that the – that Flux.ai cares about what I’m doing. And it’s – some people would say it’s an expensive signal for you as the CEO to send, because you have to spend your time on it. But it sounds like it’s been critical to get your community to where it is, to the success of the company. [0:21:03] 

‍

Matthias Wagner:  [0:21:04] Yeah, absolutely. We – I have done this myself, and the reason it was so cumbersome is why I started the company. And then everybody who is on the team has experienced this pain one way or another, especially everybody at Flux who’s user facing. All our advocates, they’re all used to work nine to five, 40 hours a week, on designing hardware themselves. So, they can see eye to eye with users. 

‍

And we’ve also tried to also have a diverse team, so not just people from Apple here, but have people from Apple, have people that work at SMBs and have people who’ve just been contractors. Because these different user personas, they have different needs and different way to engage, and different problem sets. That is important, but we always say that it works. It’s easier to teach a former hardware engineer how to run a community than vice versa, and that’s probably true for many- [0:22:12] 

‍

Richard Rodger:  [0:22:12] That’s true; that’s true. 

‍

Matthias Wagner:  [0:22:12] -for many startups, is whatever you’re doing, it’s easier to teach domain experts how to run a community than the other way round. [0:22:19] 

‍

Richard Rodger:  [0:22:20] Be slightly more friendly. I have a question for you. You come from a place of working in hardware. You probably have always worked closely with software engineers, but this time you had to lead a team of software engineers to build a software product first. What was that like? Did you find it almost the same or different? I’m really interested in your perspective. [0:22:47] 

‍

Matthias Wagner:  [0:22:47] I’ve done both. I’ve worked at Metta as a product manager and working with software teams. So, I was familiar with both. But maybe there’s a related question in here, which is, making software to build hardware, this is a first for me. And so, that’s a unique challenge here in that. It’s because mistakes in hardware are so expensive; you have to adjust how you build the software to be more fault tolerant. And if you’re going software to – social media software like we did at Metta, the Like button not quite looking up to spec is- [0:23:34] 

‍

Richard Rodger:  [0:23:35] Or missing account, is not going to – people won’t die. [0:23:37] 

‍

Matthias Wagner:  [0:23:38] Exactly. This isn’t going to cost anyone millions of dollars. But when you build a design tool for hardware and somebody designs something in the sense of the manufacturing, it can lose millions of dollars. [0:23:50] 

‍

Richard Rodger:  [0:23:53] But how do you deal with that problem? I’m going to get very nerdy and scientific here. Because from what I’ve read, when you put the hardware together on a board, it’s physics; the elements interact with each other in certain ways. How much of that do you simulate in the software? Is it possible? Or are there – have you come into edge cases where something works and it works perfectly in Flux.ai, but then when you put it into hardware form, it doesn’t, or it behaves differently? Is that much of a challenge or am I going overboard with science? [0:24:33] 

‍

Matthias Wagner:  [0:24:35] No, that’s definitely a challenge. In the ideal world, you would be able to fully virtually test your project, your circuit. [0:24:43] 

‍

Richard Rodger:  [0:24:44] Down to the quantum level, right? [0:24:45] 

‍

Matthias Wagner:  [0:24:46] Yeah, exactly, that would be amazing. And that’s certainly what we’re – what the dream is, where we are today; it’s like, you can definitely simulate whole parts of the project. Especially if you think about logic kind of stuff – that’s fairly easy to simulate and build around. If you think about an EV charger, all the safety mechanisms, the safety mechanisms in that product can’t just rely on the firmware. You have to have actual hard hardware safety mechanisms. 

‍

For example, if the micro-controller fails in the charging, then it has to disconnect. And there has to be an analog circuitry to do that, to ensure that that happens. And these kinds of things, they’re fun in a simulator to test and debark and get them right upfront. And that saves a lot of time, that otherwise you spend in – with atoms, testing that. Especially, around safety features, you can argue that it’s – you still have to test them in the real world, but you can do a lot of more iteration virtually; that’s also a lot safer. 

‍

And we do that, but then there’s other categories of issues you can catch with static rule checking, where it’s like, look, these two things can’t be closer to each oth3er than X. So, you can do that kind of checking. But then also, if you think about the collaborative nature of Flux, there’s a big emphasis on other people. Make it really easy to get feedback from others and have others review what you’ve done here. And that plays a huge role, and the more we simplify that and take the friction out, the more people do that. 

‍

And then another – so, you see, there’s multiple layers to the whole story. There’s also AI agent again; the AI agent too can perform design checks and design reviews, and you can feed that AI agent with your own set of requirements and constraints, what certifications you have to adhere to and so on. And then the AI agent will take these things into account, so it’s a huge leverage here for users. [0:26:53] 

‍

Richard Rodger:  [0:26:54] And on the AI side of things, did you – have you trained your own modules, or are you using normal generative LLMs? Or is it a totally different approach or- [0:27:05] 

‍

Matthias Wagner:  [0:27:06] Both. 

‍

Richard Rodger:  [0:27:07] Both, okay, awesome. [0:27:08] 

‍

Matthias Wagner:  [0:27:09] I was – it’s a hybrid today, but I also feel like – we launched our AI agent in April this year, and the architecture around it has been extremely rapidly evolving, as the whole AI ecosystem has been rapidly evolving. It has, but we’re just six months into GPT4; this is absolute frontier work. [0:27:28] 

‍

Richard Rodger:  [0:27:28] Yeah, it’s crazy. 

‍

Matthias Wagner:  [0:27:31] Yeah, it is. It’s normal now, but it didn’t exist just seven months ago. It’s certainly evolving, but it’s a mix. We use – if you think about the example I made earlier. You want to build a smart chocolate brownie oven. Then – but certainly, having work knowledge of GPT4 has – is beneficial here. Because it knows what a brownie is, what chocolate is and what an oven is and what it means for it to be smart. 

‍

Then we can then combine that with, for example, more custom parts of the stack, like we have a huge library of data sheets for semiconductors, so we can help find the right components and what supporting components they need for such a project. So, it’s a combination of in-house and external models. [0:28:16] 

‍

Richard Rodger:  [0:28:17] Does it hallucinate? 

‍

Matthias Wagner:  [0:28:20] Yes. I call it creativity. 

‍

Richard Rodger:  [0:28:24] Maybe you guys are imposing these physical rule checks, so I guess you catch the hallucinations more than… [0:28:29] 

‍

Matthias Wagner:  [0:28:30] Yeah. We – the art here is – if we think about creativity, we want to keep that in the product. But if you think about brainstorming on the bill of materials for that chocolate brownie oven, you want creativity. But if you think about the spec for a semiconductor, no, you don’t want creativity – you want the actual facts, please. [0:28:50] 

‍

Richard Rodger:  [0:28:51] Yes, you want – yeah, the logic has to work. No more Pentium bugs, right? [0:28:54] 

‍

Matthias Wagner:  [0:28:55] Yeah. We’ve done a lot of work here to – providing actual sources, and citing factual sources for certain categories of tasks. And for other tasks, to retain the creativity that it puts in display and developed really amazing use cases of creativity. Bill of materials, for example, is one of them, but we’ve had the users ask it, for a formula to size a certain sub-circuit. I think it was a signal filter. 

‍

And it came up with a very novel way to lay out the math and solve the problem that was way more elegant than anyone had ever seen in – at that company. Then they had a bunch of PhDs in the field, and they went and figured out if this formula actually works, and they figured out, this is just a novel way to solve the problem that’s more elegant. [0:29:42] 

‍

Richard Rodger:  [0:29:43] Wow, that’s very cool. [0:29:44] 

‍

Matthias Wagner:  [0:29:44] And how these large language models work is probably took from a related field a solution and applied it to this field. Right? [0:29:51] 

‍

Richard Rodger:  [0:29:53] Yeah, amazing. 

‍

Matthias Wagner:  [0:29:53] So that, you want to retain.  

‍

Richard Rodger:  [0:29:54] Yeah. So – and this comes back to the democratization that you were talking about earlier. But here’s another question, which again is maybe a little bit controversial, on the software end of things. I use ChatGPT and AI and all that in my work, but I’m – I‘ve been coding for 26 years. I can – I know how to get it to do what I want, and if it gives me rubbish, I know it’s rubbish. Or maybe I just want the basic outline of how do I use this software library, and then I’ll adapt it to my own thing. 

‍

But I look at the software developers that are starting today, and they won’t have to go through the same pain that I did to learn a lot of stuff. But – now maybe I’m going back to the Stone Age here and I should go live in a tree. Is it a problem that they’ll never have that feedback loop, that they will always have had this help from the AI? 

‍

And will that – will there be a gap in their abilities? Should we – is – maybe I’m an old man shouting at clouds and I shouldn’t worry about this at all. I don’t know what to think about it, Matthias, myself, when it comes to software, and the same thing must apply in the world of hardware, right? [0:31:30] 

‍

Matthias Wagner:  [0:31:31] Yeah. I would generally say that I’m not concerned about the kids or the future. And the reason is that I became a software engineer in a world of Google and you could make the same argument for Google, that just because I can now- [0:31:46] 

‍

Richard Rodger:  [0:31:46] There you go, yeah. 

‍

Matthias Wagner:  [0:31:47] -search for the solution, that I don’t know how to use a library and books anymore, or solve problems on my own. [0:31:53] 

‍

Richard Rodger:  [0:31:55] That’s a very good point; that’s a very good point.

‍

Matthias Wagner:  [0:31:56] Even with Google, there were still plenty of problems to be solved on my own, I would argue. And even with AI, there’s still lots of problems to be solved on your own; don’t worry. You’re not going to get bored. But it just sets a new bar. [0:32:10]  

‍

Richard Rodger:  [0:32:11] Now that’s a really good point; yes, that’s a really good point. But the next question which that generates is- [0:32:16] 

‍

Matthias Wagner:  [0:32:16] It’s like – I have a nice quote here – I think Steve Jobs had that – that a computer is a bicycle of the mind. And AI is an e-bicycle for the mind. [0:32:26] 

‍

Richard Rodger:  [0:32:27] Yeah. You’re right. [0:32:28] 

‍

Matthias Wagner:  [0:32:29] It’s just further, faster, less effort. [0:32:31] 

‍

Richard Rodger:  [0:32:32] Yeah. You’re dead right. I like that point, and it generates a question for me, which is, I wonder is the step change from no Google to Google, as software engineers – because I remember that. I remember a friend sharing the link with me, and suddenly my work doubled in speed. Is that a bigger change? [0:32:54] 

‍

Matthias Wagner:  [0:32:54] But that was a learning curve too. [0:32:55] 

‍

Richard Rodger:  [0:32:56] It was, yes. You had to Google-fu; you had to learn how to use Google. But I wonder is the change from zero to Google and Google to AI – which is the bigger change? It’s interesting; it’s an interesting question. And I would argue the first one is maybe a bigger change. [0:33:16] 

‍

Matthias Wagner:  [0:33:20] It’s hard to say, because it wasn’t that before Google we didn’t have ways to find information on the world on the internet. It’s just that Google provided a 10x improvement to the process. Because the Google-fu you related to, I liked it, especially earlier with Google; you could get better results if you knew how to search for things. [0:33:41] 

‍

Richad Rodger:  [0:33:41] How to phrase it, yes. [0:33:42]

‍

Matthias Wagner:  [0:33:43] How to phrase it. And it has now become more democratized; like Google can use more natural language today. But with AI, it’s the same thing; we clearly see it as people that find out more effective ways to extract what they want from AI today than the average person.  It’s very similar; it’s just on a different level. It’s these things, they’re built on top of each other. [0:34:05] 

‍

Richard Rodger:  [0:34:06] You guys are right in the middle of it. I’m so jealous; that’s awesome. That’s an awesome, interesting place to be. [0:34:12] 

‍

Matthias Wagner:  [0:34:13] Yeah, the timing worked out very nicely for us. The – I came from Metta, where we used machine learning for all sorts of stuff, even stuff that we consider mundane, but I saw the impact. A favorite example I have here is, at Metta, wherever you open – it’s called a people picker, like if you want to invite someone, there’s a list of your friends. 

‍

This isn’t just alphabetically. The first section is a machine model that is really good at finding the most likely people that you would want to invite or add or send this to. And that’s much more effective than having to go through an alphabetical list, or having to search for these people. This is a small example. 

‍

I knew that there was – there would be tons of opportunities to use machine learning and AI for hardware design. Now large language models I didn’t quite foresee having this impact in 2019 when we started the company. But then when it happened, it was pretty clear quickly how that would integrate, what the benefits- [0:35:12] 

‍

Richard Rodger:  [0:35:12] Totally applicable. 

‍

Matthias Wagner:  [0:35:13] Yeah. Just think about-

‍

Richard Rodger:  [0:35:14] 100%. 

‍

Matthias Wagner:  [0:35:15] -the use case of – every semiconductor out there, microprocessors, whatnot, they come at thousands of pages on PDF datasheets, where all the specifications and requirements and whatever, the application notes and whatnot. And I’m unsure you could read all that; it takes a lot of time. It turns out that a large language model is incredibly good at extracting information you’re seeking from that. Incredible, right? And within split seconds. And that is an incredible use case. In hardware, and in electronics design, a lot of time spent on researching; a lot of time is spent typically on Google. [0:35:52] 

‍

Richard Rodger:  [0:35:52] Reading the sheets. 

‍

Matthias Wagner:  0:35:53] -trying to figure out – yeah, reading the datasheets; finding the data sheets; finding other components; replacements for things; finding cheaper ways. And now you can have a conversation with the AI agent offering, and within split seconds you can retrieve information and create associations that otherwise would have taken you weeks or months to do. [0:36:12] 

‍

Richard Rodger:  [0:36:13] Isn’t – which is amazing, but isn’t there another problem, which is sourcing the components? Are you guys going to integrate with manufacturers that provide components, so that- [0:36:23] 

‍

Matthias Wagner:  [0:36:25] We already do. We have a real time index of the components in the world, the stock availability at different distributors, the pricing at different order quantities. All of that we have,. and all that information, we’re finding – we’re coming up with new ways to integrate that information into the AI agent, so the AI agent can make a chart. [0:36:46] 

‍

Richard Rodger:  [0:36:46] That’s what I meant. The AI – when it – it doesn’t suggest a component that isn’t available, or is only available in small quantities. Or you can only order 10,000 at a time, which is not suitable. [0:36:57] 

‍

Matthias Wagner:  [0:36:58] And you can leverage the – with AI, the thing is, you can train or fine tune the model. And that’s an incredible capability, but it’s very slow and very expensive to do. So, you can’t do that for something like pricing that changes every hour, of stock availability. But you can fine tune it on general trends, that components from company A are always cheaper than B. Or the other way round components company B are always higher quality. 

‍

These general trends you can definitely code, so they can be used to make recommendations. You want a more reliable thing; it’s known that Texas Instruments makes a more reliable version of this component you’re looking at. Or that this other manufacturer from China makes a cheaper version. 

‍

So, you can work with that. And we’re working on keep improving that, and keep enabling it to access as real-time information as you possibly can. The datasheets example is good, because datasheets don’t change so much, and that was a good first win for us to get data in, because it’s normally static information. [0:38:01] 

‍

Richard Rodger:  [0:38:03] Yeah, Matthias, really cool stuff. I’m going to end with a macroeconomic question, totally change. [0:38:10] 

‍

Matthias Wagner:  [0:38:11] Hit me; hit me. 

‍

Richard Rodger:  [0:38:15] The world is going to strange places these days. Do you think that electronic component manufacturing will get more distributed? It got very focused on China and places like that. And do you think it’s – do you think that will change now? [0:38:29] 

‍

Matthias Wagner:  [0:38:33] Yes, it will change. There’s two layers to it, but the first one is the actual semiconductor fabs. There, we’re going to see now these moving onshore again, to North America, but also Europe, because – especially for this stuff, GPUs, which is critical – considered critical infrastructure now. 

‍

Also, because it’s leverage for AI. And that’s also probably the more high-end fabbing of semiconductors. There’s a lot more bread and better stuff that’s already fairly distributed. And it’s also easier to spin up fabs all over the world. There’s for good reasons a good focus on this high-end, cutting edge of fabbing. That’s the first layer, semiconductor fabbing. 

‍

And then the other one is the fabrication and assembly of the printed PCB boards. And here too, China had for the longest time – not an advantage necessarily in the manufacturing of the board, but in the assembly of the board. Because that’s a human intensive piece. And it’s impossible to compete on price, doing this onshore in Europe or North America, because labor costs are so high. 

‍

The only way to bring that onshore in a competitive way – for things where the cost matters, which in some cases it doesn’t, but for the cost matters – is through more automation. And automation exists in this field; it’s just you have to design with automation in mind. If you think about where these components are of microscopic size, they get off these picking machines that vacuum sucked, and there’s a needle, like a blunt needle for your buds. That’s tiny; that sucks it up and grabs it. 

‍

For that all to work reliably and at scale and fast, the components have to have a reasonably flat surface so they can be picked, average availability. Then the components have to not be placed too tight with each other, so these pickers have easy access. So, there’s lots of opportunities here to improve that, and if that’s done. And a big leverage is in the design tools like Flux, to enable that, to enable us to do that, then we’re going to see a lot of that coming onshore. [0:40:55] 

‍

Richard Rodger:  [0:40:59] Yeah, okay, interesting. So, that fits into what you guys are doing as well, because you can support that activity as well. Will we ever get to the point – sorry, second last question. I have one last question. Will we ever get to the point where 3D printing becomes 3D complete manufacture, where you can produce a radio-controlled toy car? You can 3D print a lot of components now, but the electronics inside it are still – you can’t 3D print those yet. [0:41:32] 

‍

Matthias Wagner:  [0:41:35] You definitely can; we definitely will. There’s a lot of progress happening. 3D printing too has been slower than we all thought, but the potential is still here. You see now 3D printing being – in a lot of industries being adopted for mass manufacturing. That means that the quality and speed has – and cost has got to where that’s competitive as a solution. I know a company up in Canada, Pentium Design, they don’t even call them 3D printers anymore; they call them micro-factories, because that’s what they are. 

‍

That’s come a long way now, and us seeing an integration of 3D printing with electronics, it’s going to happen. There’s going to be an in-between step, where we don’t even have that yet for a factory to do that. It’s typically still injection molding in one place, PCB manufacturing  and assembly in the other, and then it comes together. 

‍

And I’ve seen startups working on that, where they’re working on these self-configurable manufacturing lines, where you have this module. One can be a 3D printer; the other one is an assembly arm, the next one… and they can on a job-by-job basis reconfigure each other, create this customer’s assembly line and then produce the product packages and all that. 

‍

And that’s definitely going to happen in a few – timewise – for some categories of products, maybe you could argue it’s already happening. For other categories, we’re maybe five years out. For some categories, it might never happen, but you probably for the bulk of it, I could imagine that we’ve got a 10-year span, especially if you think about the consumer electronics goods, the stuff you buy on Amazon, that you are on a 10-year horizon; you can see a drastic change here. [0:43:36] 

‍

Richard Rodger:  [0:43:38] So, it’s still going to be exciting. That’s awesome. I’m so glad we had this talk, because I’ve been feeling a little bit cynical about the future of physical technology, but there’s all sorts of cool stuff going to happen. That’s awesome; that’s amazing. [0:43:54] 

‍

Matthias Wagner:  [0:43:54] Yeah. We – I always tell people about – think about how Annakin Skywalker made his own protocol droid. That’s the future I want to live in, to enable that, to give that kind of leverage people. And to get there, that’s totally possible. We just have to build more abstraction layers and democratize more of the knowledge and abstract that away. [0:44:15] 

‍

Richard Rodger:  [0:44:17] Which takes us all the way back to the start, democratizing technology; looking after your engineers; having good engineer relations. Matthias, thank you so much; this has been fabulous, really cool. I’m so excited. Thank you so much. [0:44:28] 

‍

Matthias Wagner:  [0:44:28] Thanks for having me, Richard. 

‍

Richard Rodger:  [0:44:29] Bye-bye, bye-bye. 

‍

Matthias Wagner:  [0:44:30] Bye. 

‍

Endnote

‍

Richard Rodger:  [0:44:31] You can find the transcript of this podcast and any links mentioned on our podcast page at Voxgig.com/podcast. Subscribe for weekly editions, where we talk to the people who make the developer community work. For even more, read our newsletter. You can subscribe at voxgig.com/newsletter, or follow our Twitter @voxgig. Thanks for listening. Catch you next time. [0.45.00]