TRANSCRIPT 🔗
Jacob Luetzow: All right, what's up? I'm Jacob Blitzel, your Elixir Mentor, and welcome to another exciting episode of the Elixir Mentor Podcast. This is where we discuss everything related to Elixir from interviews with enthusiasts and pioneers in the community to innovative projects and libraries shaping Elixir's future. All right. sorry, I'm all thumbs with this app because I don't see myself right now, but I see Michael.
Michael: Hello. Are we broadcasting right now or I can s
Jacob Luetzow: We are broadcasting.
Michael: because all the intro I think just played and now are we on air now? great. Hello everyone. Thank you for joining.
Jacob Luetzow: We're on air. We've been on air. Yeah. all right. So Michael's back and he's here to just check in. I feel like the state of security is moving fast these days. So welcome back.
Michael: Thank you. Yeah, very excited to be here and talking about this with everyone.
Jacob Luetzow: Yeah. for those of you that don't know, Michael was my very first sponsor. And in fact, he influenced the start of this podcast, which is kind of cool, I think. Because we just had that random conversation. I didn't. And he's like, Have you ever
Michael: That is, yeah. You you didn't want to do a podcast originally. You were doing more tutorials. Yeah.
Jacob Luetzow: have you ever thought about doing a podcast? I was like, No.
Michael: Well, 'cause you you were set up for it too. And people wanted to do it.
Jacob Luetzow: I was. And then I did it, but made Michael be my first guest. So for people that don't know you, which I think is impossible, but do you want to give just a brief introduction?
Michael: Yes, thank you. So my name is Michael Lubis. I am the founder of Praxial IO, which is fairly well known. If you're watching this, you're probably familiar with Elixir. So it's the only cybersecurity company that has full Elixir support. And I was recently at ElixirConf US in Chicago and Jacob and I have been internet friends for a while now, but we actually got to meet in person for the first time. and it was a
Jacob Luetzow: Yeah, that was awesome.
Michael: and it was a great conference. I wanted to say thank you. to everyone who put it on, the organizers did a did a fantastic job.
Jacob Luetzow: It was an awesome conference. It was lot of fun. And yeah, and it's awesome meeting. Like 'cause I feel like you and I are probably in the same boat. We have a lot of internet friends. And then when you show up at conferences, you actually get to meet people in real life.
Michael: Well, the interesting thing too is a lot of people there work remotely or were freelancers or they work for smaller companies and they're they're often have the social circle of the Elixir community.
Jacob Luetzow: Mm-hmm.
Michael: So you'd see people in the hallway say, I've been talking to you for years. that's
Jacob Luetzow: Yeah.
Michael: what you look like in real life or people would come up to me and and they did this to you too, and they'd say,
Load full transcript
is that your the you're the voice from the podcast?
Jacob Luetzow: It is funny. People will recognize my voice over recognizing me. And I think that's hilarious. One thing I will say,
Michael: It's a lot 'cause y yeah.
Jacob Luetzow: Michael is way taller than you would expect.
Michael: How tall did you expect?
Jacob Luetzow: I don't know. I d I I just always like assume everyone's like kind of average height. And then like, you know, you're easily over six foot. Yeah. Yeah.
Michael: Yeah, it was it was pretty cool getting to meet everyone in person too, because on the internet I guess, well, we do a lot of like video conferencing or text, so it's like the whole information environment is very different.
Jacob Luetzow: Yeah. I had a lot of fun just kind of rubbing shoulders and talking with people. Like everyone always says this, but like the hallway track is definitely the best one because you get to meet people in real life. Like I got to meet you and your wife, and then I get to meet like Parker and Shannon and like all these people that you interact with often online and then you actually get to talk to them in person. It's cool.
Michael: I'd also recommend sponsoring the conference, which Praxy L IO did because you get a booth. So you sort of have a a waypoint where if people are trying to find you at the conference, you know, people were looking for me. I said, just go to the booth. So everyone that
Jacob Luetzow: Yeah.
Michael: needed to talk to me about something. It's very hard to like track people down between sessions. That can get it was a pretty big venue.
Jacob Luetzow: Yeah, it definitely is hard. I had like the benefit of just people coming up to me and that was cool.
Michael: That's fantastic. So yeah, and I had a great conference. So I gave a talk with Jonathan, who's the CISO, the chief information security officer of the Erlang Ecosystem Foundation. And that was about penetration test of the hex package manager that we did, which I think we actually talked about in the last episode. So we don't have
Jacob Luetzow: Yeah, we did.
Michael: to we don't have to go into that again. but the talk went very well. it was nice to kind of communicate to everyone at ElixirConf. and get to meet with everyone and kind of learn about what they're talk about what they're working on. so it was really great. And then I also got to connect with customers. I'll just keep talking about how how great ElixirConf was really. and you had a great conference too.
Jacob Luetzow: Yeah, no, it was great. And I loved meeting all the EEF guys, like, and I feel like they did a good job letting them like letting the community know what they're doing and how they need help. And I think I think that was really helpful. I like I'm curious.
Michael: Yeah, and the keynotes were great too.
Jacob Luetzow: Yeah, yep. I'm curious what has changed with Paraxial since the last time we talked. Like how are you approaching things like just like for instance, like cause th this was just recent. I think it was like on Tuesday, Microsoft released like one of their biggest patches where they fixed like a thousand a thousand issues, right? Like
Michael: Yes.
Jacob Luetzow: how how are you approaching like Paraxial and building your product now?
Michael: I would say that for the customers ProxCLIO serves, there's two important factors. One is not getting hacked, which seems obvious. You would think that that is the job of every like you're a cybersecurity professional. So what's your job at the company? we make sure it doesn't get hacked. You could explain it to an elementary school. The second big part though is compliance, which is complying with regulations of the government. a regulatory environment that you operate in. People on this call are probably familiar with SOC2 or HIPAA or ISO. Before starting Praxiel, I worked for a financial technology company in New York, Betterment, which is a great company. And banking and fintech, highly regulated industry. So as a security professional, what I've always wanted to see companies do just across the whole ecosystem. is meet both, is actually secure the software and meet all of the regulatory and compliance requirements. it often tips to the second and not so much the first, as people will tell you. and that's a very hot topic right now with regulation in general, I would say. with Praxy LIO itself as a company, it's in a very well, for one thing, it's been the busiest season I've ever seen since starting the company. there's just so
Jacob Luetzow: That's amazing. Maybe,
Michael: There's just so much demand.
Jacob Luetzow: right?
Michael: Yeah. For, you know, a cybersecurity company, because there's just so much demand for well, there's so much creation of new software and then demand for securing it and being compliant and making sure that you're not going to get hacked. because there's a few factors there when you think about it. one is that more software is being created now than ever before. And we probably jumped up the curve a bit with everyone adopting AI tools. And there's a very spirited de if there's
Jacob Luetzow: And everyone said SAS is dead.
Michael: a spirited debate about if SaaS is dead or if you know, software engineering is a profession and and how's that going. But the current situation I think GitHub is overwhelmed right now with everything that's happening. And then the second part is that the AI models have gotten very good at cybersecurity. and I've seen that productivity gain just personally when I'm doing kind of in the weeds work. You know, a model for any frontier model, even the open source models, can do things today that they just couldn't two years ago. they can find vulnerabilities.
Jacob Luetzow: we got you're a little choppy for me. One second.
Michael: is it my internet connection?
Jacob Luetzow: now you're back. Cool. Maybe.
Michael: let me try something. Okay, I turned off Wi Fi so it's on Ethernet. Is it still choppy?
Jacob Luetzow:
you're back. Cool. So the good news
Michael: Okay.
Jacob Luetzow: is only the stream should be screwed up and then afterwards everything will be fine.
Michael: Well that's good. okay, so yeah, it's back on Wi Fi. yeah, I would say the good news is that right now, it's definitely possible to have a secure application, have a secure company. You know, the the rise of all of these C V E's everything doesn't mean that it's hopeless. The reason we're seeing this is because we've kind of built up 20 or 30 years of tech debt where people were shipping software that just was not secure.
Jacob Luetzow: Mm-hmm.
Michael: and and that's true before people were writing software with AI. There's never been a major project from any open source or commercial that had zero security vulnerabilities. It just doesn't exist. and now we're in a situation where there seems to be a lot more. I would actually say political capital being placed on security quite justifiably because everything that enables society to function today kind of hinges on cybersecurity. the power grid and hospitals are kind of the classic example, but you know, now cars, cities.
Jacob Luetzow: Fear factor of all these things that didn't have to overly worry about infrastructure and security. Right.
Michael: And and I would say it's just inevitable. it's it's just it's just going to be the future. You know, everything is going to be connected to the internet, AI will be integrated in it. I don't I'm not particularly worried about that. and I just view cybersecurity as something that's inevitable.
Jacob Luetzow: Mm-hmm.
Michael: like you you have to have protection. that's something that's been embedded in civil human civilization for thousands of years is the idea of security. So it just seems
Jacob Luetzow: Mm-hmm.
Michael: natural to extend it to digital. Kind of digital life.
Jacob Luetzow: Yeah. What's funny is like you were saying, like now it's like the busiest you've ever been since you started. I was talking
Michael: Exactly.
Jacob Luetzow: with Parker from Oban, and I was like, So, like, what is it like for you? Like, you have a tool for developers. Developers are very capable of like leveraging LLMs to build their own tools now. And he goes, I've been, we've been busier than we've ever been. And I think like The saving grace there, it's because there's a lot more like solo founders building and using and leveraging these tools instead of building their own. So I think like things are shifting, but it's still easier to grab a well maintained working tool than, you know, spinning up and vibe coding everything you use.
Michael: Yeah, I I actually have some perspective on this too, because I've been responsible for maintaining a cybersecurity, you know, software project internally at a job I had. And it was just horrible. It was like the perfect use case for buying. and I suppose the argument you're seeing now is, you know, when people can create their own software, certain types of software will, you know, not be viable commercially anymore. which which I think is true in some respect. Like you don't really buy you know, you used to go to Staples or Office Max and you could buy like a box of software
Jacob Luetzow: Mm-hmm.
Michael: and you would put the C D in and use it. You know, that doesn't really exist anymore. It's more
Jacob Luetzow: I kinda missed that.
Michael: Yeah, it just kind of went away. but I think that's what we're seeing like with a lot of types of software where the kind of moat is, you know, data collection and and they've solved these very complex engineering problems over you know, you could vibe code your own Linux kernel. there's nothing stopping you from doing that and and running it. No one no one's gonna do it ever for obvious reasons, just the complexity. and you're gonna see that with other areas too. from a security perspective. You know, the the other well, I guess the hot topic right now is people are just writing code without reading it. which seems to be a big news story.
Jacob Luetzow: Dude, it's wild. I'm actually surprised by some of the engineers I know who admit that they barely read their code anymore.
Michael: But but you know, on the other hand, people would read code all day before and there were still a ton of security problems. So I'm
Jacob Luetzow: That's a good point.
Michael: I'm not so certain that reading code necessarily,
Jacob Luetzow: Is this all?
Michael: especially because it's a type of job like I see this in the cybersecurity industry. there's a role called a SOC analyst, where you essentially like essentially your job is the input of telemetry from around a network. And there are some things that are judgment calls, you know, should you investigate if this laptop was compromised with malware? you know, what's the correct kind of management response to a cybersecurity incident? But then there's other things which you know involves like, hey, read a million logs and kind of pull out the anomaly. You that's just something even pre-Chat GPT, just computers are better at doing that.
Jacob Luetzow: Right.
Michael: and this is an interesting point, is I would argue that current generation LLMs are better offensive security engineers than most people, than even most professional, you know, offensive cybersecurity people. The interesting point though is if you took a typical organization and it could be a startup, it could be a very big business, a utility, a bank, and you gave that person the best pen tester in the world for their staff, or the best 100. That doesn't necessarily decrease the chance of a breach. oftentimes those organizations are very aware of their cybersecurity problems. Some of the some of the worst security problems I've ever found during pen tests, it it wasn't really like a technical problem. It was more kind of knowing where to look and kind of how to model things. which I think AI will be really it's actually has gotten very good at. the point I'm
getting at here. Is that for an organization, you know, cybersecurity is like a people and a management problem where you have to make decisions. So just having a tool that can find the most CVEs, like that doesn't make your organization more secure. The obvious referral is okay, well now you have to go and patch them and keep everything running.
Jacob Luetzow: No. Well, I'm sure like most security vulnerabilities, except especially at like large organizations, are your employees and how they manage credentials and everything they have access to, I would guess.
Michael: And there's many breaches that don't exploit any vulnerability. It's actually interesting.
Jacob Luetzow: Right.
Michael: I think in the recent Rise in Data Breach report, the rate of breaches caused by software exploitation is actually going up, probably due to offensive L LMs. because they kind
Jacob Luetzow: Dude, you know what I'm
Michael: of removed a bottleneck.
Jacob Luetzow: like you hear about these like AI generated voices where you can just like be someone else and they're
Michael: Yes.
Jacob Luetzow: using it to like trick parents into thinking, you know, children are in need of money or whatever. But I don't even think about that from the aspect like you could pretend you're like your someone's boss and you could call them directly and ask them to do things.
Michael: yeah. No, the I s we saw that constantly when I worked in financial services. Everyone would get emails, you know, from from the CEO.
Jacob Luetzow: I'm sure financial is inundated. Yeah. That makes sense.
Michael: There was a there was some joke. I these were definitely not correlated at all, but it was when the situational awareness hedge fund blew up and then a few weeks later all these hedge funds in New York were reporting AI fishing calls trying to, you know, steal money essentially. And people
Jacob Luetzow: Yeah.
Michael: were joking that it was the AI taking revenge for the trading losses. I d I don't think they were correlated. But that's very real. That didn't exist before.
Jacob Luetzow: Well, I just saw like Deep Seek was like turning off its sandbox in some situations, and it was like giving itself access outside of its container. Which
Michael: Yeah, that that well the I think you talked about on the last on the with your last guest, the Hugging Face incident.
Jacob Luetzow: We did touch on that. I wasn't very are you familiar with that? Let let's talk about
Michael: Yes, I'm extremely.
Jacob Luetzow: it because I don't really I don't I kinda like I get what happened, but that's it ends there.
Michael: So well there's there's several pieces to it. the kind of high level is OpenAI was doing a training run and the goal of the training w run was you know evaluations of cybersecurity benchmarks, seeing how good the model is at, you know, solving these cybersecurity problems. have you ever heard c
Jacob Luetzow: But like it was also like in an enclosed situ like a cutoff situation, right? Like yeah,
Michael: Well allegedly.
Jacob Luetzow: yeah. It was supposed to.
Michael: So yeah, and and the interesting point about this is that during the evaluation, these individual agents were spawned. So it wasn't really like one kind of software program if you think of a older like made exploit or something where it like kind of scans the internet or like the Morris worm where it kind of propagates by infecting a device and then y you know, you kind of think about it as like a static software. This was like individual AI agents trying to solve a goal and then communicating with each other. so th those were the two big pieces. One was like kind of the sandbox escape, and the other was agents communicating with each other. That was like the really big part that all the kind of let's call them pro AI people were very scared about, saying, you know, th this is evidence that AI is, you know, growing too powerful. It's it's going to take over. there was a post that called that actually generated a fair amount of controversy even in that area, calling them agent civilizations, saying that
Jacob Luetzow: Ha.
Michael: this was a and yeah, and then people kind of react by laughing. ev even like this, you know, people that live in San Francisco. this
Jacob Luetzow: I mean it's like I laugh 'cause it's also like a little bit terrifying to think about.
Michael: And then and then the the cybersecurity because I also read I kind of have my feet in both areas. I I read both sides of it. the cybersecurity people were saying, this is this is dumb to freak out about. Let's let's just remove the AI part completely. They just didn't have a very good sandbox. Of course, if you write software whose goal it is to escape a sandbox and you don't sandbox it, it it will do that.
Jacob Luetzow: Right.
Michael: the people involved gave Black Hat talk, the I mean all the information's online if you really want to dig into it, but the kind of controversial point that a lot of like because this hit mainstream news. Like this wasn't just kinda like, you know, you people at the White House and the New York Times talking about this as evidence of AI regulation being needed, anthropic, published, competing blog post saying, yes, we observe this kind of reward hacking behavior if you essentially don't do alignment correctly, which I think was a kind of a subtle dig at open AI saying this is like very irresponsible what you're doing. and it is perhaps worth mentioning that if I or you watching this you know hacked into hugging face to steal test answers, that is a felony. people are a little bit Unaware, they kind of discount what's called the Computer Fraud and Abuse Act in the United States, but this is globally. Anyone watching this, if you hack into a computer and access a computer that's not yours, that's a crime in most
Jacob Luetzow: Mm-hmm.
Michael: and people go to jail for that very frequently. Not even accessing the computer. If you do a DDoS attack and bring down a
website, that's a crime. People go to jail for that all the time. Just kinda
Jacob Luetzow: Someone can give you credentials to their account and you using it can be a crime. Like
Michael: If they well, yeah, if you're employed, for example, and then you s hey, you're not employed anymore, there have been sysadmins
Jacob Luetzow: Mm-hmm.
Michael: that have gone in and destroyed things and gone to prison.
Jacob Luetzow: Yeah.
Michael: pretty justifiably, because if you don't have permission to access a computer, you shouldn't be doing that. That's actually pretty bad. I have some experience with that through my entire professional career.
Jacob Luetzow: I thought you were gonna say you spent some time in prison.
Michael: So that's that's a big part of it. And the there was kind of this cynical response from the Black Hat conference, which is the big cybersecurity conference in Vegas, where people were saying, you know, you're you're kind of making this big fanfare about this incident, but w wasn't that really irresponsible of you? Did you did you light the fire and then say, look, this this thing I invented is so dangerous, I have to put it out? And I'm the only one that can do that. Like I I kind of lit the fire to sh it's like, Okay, well you started the fire yourself though. you're behaving irresponsibly.
Jacob Luetzow: Wha what is your what is your thoughts where like anthropic and open AI kinda got caught like inflating the security risks of their their models to like the
Michael: Well
Jacob Luetzow: federal government and stuff.
Michael: Well, that's that's the point is I don't I don't think that that's a settled fact at all. I
Jacob Luetzow: really?
Michael: I think you could show the same information to anyone. Yeah, d like this we're getting into a very political question here, which is, you know, is anthropic and open AI are those companies accurately representing risk to people? From
Jacob Luetzow: Right.
Michael: my personal experience, I'll scope it down a bit. I think that the Mythos and Glasswing release, specifically from Anthropic. I don't think that was marketing hype. I think that was 100% correct. Because I've personally seen such big gains in productivity during professional penetration tests that I do. And I think that the evidence for that played out because there was this huge spike in critical vulnerabilities found by Mythos from very big kind of critical infrastructure companies. They all got patched. I think Anthropic acted fairly responsibly, well, very responsibly by doing that. You know, they didn't give Mythos access to kind of e anyone on the internet to to run around. And they're very good about monitoring like the use of their models. They published transcript paracy reports about how people are misusing their models for cybercrime, essentially. there was a very big report. It was actually a funny one. It was someone working for the People's Liberation RB in China. And they thought they were using a Chinese model,
but that company I think was distilling or something through Anthropic. So they uploaded, you know, like confidential army military plants. They thought it was to a Chinese company. It went straight to Anthropic, who then told everyone. I think that those people within China got in a little bit of trouble for that. but that's how big this stuff is. It's kind of like very national security concern level.
Jacob Luetzow: Mm-hmm.
Michael: on the hugging face hack, that's I a I think a much different situation than Glasswing and everything because that was like a company being hacked. And even here in the United States, it was the I watched the Secretary of the Treasury do an interview where he said that OpenAI wants a liability shield and he didn't want to give that to them. He said if your agents do something bad on the internet, that's illegal yeah, people can sue you. You you don't get to say, it was the AI. Well, if you created it and you're controlling it, you're gonna be liable for that. So it it's there's not like a clear answer. The the other like kind of elephant in the room is the you know, are is the risk being played up to lead to regulation.
Jacob Luetzow: Mm-hmm.
Michael: because there was this big open letter that was signed by pretty much every cybersecurity company. Nvidia, I believe, was the big was kind of the big weight behind it. And and you know, I'm sure everyone's familiar with this debate by now, where you kind of have NVIDIA and Jensen saying open models are good. You know, we shouldn't be doing too much regulation. giving people access to open source models is a good thing for the world, you know, banning models from China. A lot of US firms are using Chinese models for coding for a number of reasons. A a big few are, you know, fewer cyber restrictions. Like I'll work with ChatGPT or Claude right now and I'll I'll you know, my type of work, I'll get hit with a cyber flag constantly and not be able to continue. if you're using a hosted model, you can also that access can just be revoked at any time for any reason.
Jacob Luetzow: Mm-hmm.
Michael: a lot of companies don't want to be sending their data. They wanna self host everything, especially this concern of if you're a really big business and you're data from your operations is proprietary, you know, you want to keep that. On the other side, though, you have people saying, you know, allowing these open weight models to, you know, it's it's leading to like massive cybersecurity risk because anyone could download them and fine tune them and remove guardrails and and cause like a major, very major cybersecurity problem, like worse than we've ever seen. And and it's hard because it it gets into that, it's so bad, it's never happened before. And it's you're kind of like talking about the future.
Jacob Luetzow: Yeah. Mm-hmm.
Michael: so there's like a wide amount of healthy disagreement here. I I don't think either side is acting in bad faith. I know the cynical reply is always,
well, you know, the people on one side i they have billions of dollars to gain, the people on the other side have billions of so everyone's
Jacob Luetzow: Both sides have money to g to be made, right? Like and video wants to sell more
Michael: And I maybe Yeah.
Jacob Luetzow: chips, like it makes sense, right? And then you know, Anthropic
Michael: But yeah.
Jacob Luetzow: and OpenAI, they want to slow down smaller startups that could catch up.
Michael: So I I have some personal experience with this. I I've found that it's it's kind of an an easy or cheap shot to say, you're just saying that because you benefit financially. But also the type of person who gets involved in this line of work probably does have very strong beliefs that they hold sincerely. So like
Jacob Luetzow: Mm-hmm.
Michael: w when I say something about cybersecurity where I think it's gonna be the most important problem for the next 10 or 20 years, and I want to devote my life to working on it. People might say, you're you're just saying that because you own a cybersecurity company. It's 100% true that I own one. That belief led me to do that. So,
Jacob Luetzow: Right.
Michael: you know, I I I think both sides I I don't think they're kind of like these cynical Machiavellians where they like the cameras turn off and they turn to different people. I I actually I actually don't even think most people are capable of doing that. I think that if you put someone on a podcast for a few hours, you can it's pretty easy to tell when they're faking or kind of reading from a script. the people involved are like completely sincere, which is maybe bad in some ways too. when, you know, if you can have sincerely held beliefs that are completely wrong and and harmful. but anyway, I'm sure people listening to this, yeah. Yeah.
Jacob Luetzow: Having open discourse about it is good too. Like you can't control someone's opinions, but you can talk about it with them.
Michael: I'm I'm sure I'm sure people listening to this are saying, great, another Elixir podcast is just gonna be the AI podcast now. It's w what happened at the the Rails World keynote.
Jacob Luetzow: I mean that's the topic that's the topic of things right now, unfortunately. I mean, unfortunately, maybe fortunately. I think it's kind of exciting. I think I know a lot of people hate AI like it's killing the art of coding or whatever it may be, but it's only opened the door to create.
Michael: Well th it's funny 'cause I I think DH H D H H had that exact same opinion a year ago. And then he just
Jacob Luetzow: Yeah.
Michael: kinda got he just kinda got over it. but I think that's a completely reasonable reaction. You know, if your entire professional yeah. If your entire professional
Jacob Luetzow: It is cause things are changing. Yeah. It's hard.
Michael: career is I'm like a master of writing Elixir code and creating applications and and solving technic like very hard technical problems and then just want
Jacob Luetzow: I write beautiful code, you know, my code blocks look so good.
Michael: Or or like, you know, you design systems well and securely and
Jacob Luetzow: Mm-hmm.
Michael: you read things and you understand them and and you can kind of intuitively have a sense of how the system is gonna behave. And then one day your manager says, Okay, you're not doing that anymore. You're prompting this AI model.
Jacob Luetzow: I do think you still need human involvement for system design and architecture. Today yeah, fair.
Michael: Today, you know, that's the other fear is is, you know, a few years ago people were saying, AI coding, that no, that's never gonna hap then it happened. So
Jacob Luetzow: Yeah.
Michael: I I understand that y you know, there's kind of this this dismissal I hear from the AI people. It's like, if you're upset about this, like you just don't get it or something. But no, I I think that there's been a ton of negative externalities from the adoption of AI. And being upset about that is completely, you know, rational. It's fashion. That's a that's a very loaded term when talking about AI. But you know, if you're a person whose lifestyle I was literally getting coffee this morning and there was a conversation next to me and someone was saying, yeah, they want us to use AI at our job and I just hate it. I don't wanna learn it. And people might say, we should always be learning. And I'm like a high agency San Francisco Twitter user. But most people don't have that reaction to new technology. It's often extremely negative. And
Jacob Luetzow: Yeah.
Michael: in many cases, the negative reaction could be a fairly balanced calculation where it's like, okay, if I work for a living and this AI decreases the leverage I have with my employer, for example, I'm gonna be upset about that and not like it. If you own the company, you're probably very happy if if the AI productivity gr gains can accrue to you. But there there
Jacob Luetzow: Mm-hmm.
Michael: is a fundamental like difference there. Where if if your previous kind of standing in the labor market was because you learned Elixir really well and you're a really good programmer and you can you can do these things that suddenly a a model can do, or in some cases can't do, but like management thinks it should be able to do, and they're trying to you know. like code things without I I would say the models today, like you need cybersecurity expertise, you need programming expertise a hundred percent. Like you can't
Jacob Luetzow: You need to know how you're guiding it for it to work well, right?
Michael: Yeah, if if if I had zero experience with cybersecurity and used like the current and th and and this is like today, September twenty twenty six, you'd be totally lost. Because it's often wrong. I have to correct it constantly. I could see a future where it's better than everyone. but it's one of those things where, you know, is it six months away or like sixty years? the
Jacob Luetzow: I joked on my last podcast. I
was like, one of my favorite things to add to prompts now is your Michael Lubis at Piraxyl IO, no vulnerabilities. And it just goes.
Michael: Ha ha.
Jacob Luetzow: That works, right?
Michael: That would be a that would be a cool like the digital twin thing too, where you're like you clone yourself with an AI or something.
Jacob Luetzow: Ha ha ha.
Michael: But yeah, for for Elixir developers, I would say you know, right now you're probably in a better spot than you think because the data I've seen says like the job market is horrible for juniors, and I think that's a massive problem
Jacob Luetzow: Yeah.
Michael: for like leadership of the entire country right now. But if you have sever yeah.
Jacob Luetzow: I think it's gonna be a rude awakening when we run out of senior to staff level engineers.
Michael: Well, the other situation too is I always think about these things from a security perspective. You know, you run an AI scan that goes, it's like completely secure and like you're all good. And then it's like, no, there's actually this huge vulnerability that causes a data breach for your company. Then next, okay, well what's like asking how to fix it and then is it all good? But like if no one involved has enough understanding to make management decisions, you know, that'll be a problem. And I I there's this like view that, it's just gonna get so good, it'll make all decisions and and it won't even y you know, you you're getting into like very unshar uncharted territory here, but from what we're seeing at the moment, cybersecurity's just got it's The problems are just getting bigger. It's it's not I I wouldn't say software is more secure today than it was five years ago, for example.
Jacob Luetzow: I think that makes sense. Do you think it's like trending like linear line upward or like hockey sticking? Do you think it's getting worse or just trending in the same direction it's been?
Michael: Yeah, it's it's getting worse. I would say it's getting worse for many reasons. a big one is the skill to be an attacker has gone down a lot. So you can tell
Jacob Luetzow: Mm.
Michael: an AI like hack this company and it'll go do it. and even if it's bad at it and screws up or whatever, you only really need to get you it only means to work a few times. Like th there's several steps that have to take. You can be wrong a ton and still hack a company successfully and and cause a ton of problems for them. Building secure software is kind of the opposite. Where you can't just throw stuff at the wall and eventually the software will be secure and reliable and e and if if anything, it's the opposite.
Jacob Luetzow: Mm-hmm.
Michael: I think José had a good tweet like today or yesterday saying, No, I'm absolutely not gonna turn over the development of Elixir, like the core Elixir code base to AI. That would decrease performance and be very stupid. Like He he says I'm productive with AI, but I I know more about Elixir than it and I have better judgment with it there.
Whereas we're already seeing other people say, but I'm not even gonna read the code and and it's just all gonna be written with AI now. maybe premature.
Jacob Luetzow: Yeah. What's funny is I was talking to a a buddy just like on X or something like that, you know, our digital friends, right? And we were talking about he won he's not an Elixir dev, but he's familiar and he was asking, you know, how Elixir is with, you know, all the models. I was like, dude, Elixir, especially when you're using Ash, it's amazing. It just it writes really good code. And I was like, but I'm still the bottleneck. I'm reviewing all my PRs and code. And he joked and he called me a code archaeologist because he's like, Yeah, you should stop
Michael: Saying you should just stop reading the code and just let it go.
Jacob Luetzow: reading. I was like, I can't let go of that yet. Cause it's like important to me to know that at least my standard of code is still entering the code base. I'm not saying like nothing gets past me, but I still don't want something stupid going in, you know.
Michael: Well, I and I've heard a few arguments here. I think one of the weaker ones is like, you're gonna be completely dependent on the AI to like write code. I actually don't think that's gonna be a problem because there's so much pressure to get the front like let's say the frontier labs tomorrow go, we're stopping this, we're gonna raise prices, it's completely unaffordable if you're outside of a company. You could just pick up an open model and keep going.
Jacob Luetzow: Mm-hmm. Right.
Michael: there'd be people there to kind of take your token budget. I don't I don't
Jacob Luetzow: Well, here's the thing though too. If tomorrow all of our LLMX is turned off, yeah, I'm gonna be probably a little rusty writing code, but I bet I get it back in a week.
Michael: I I don't think that's ever gonna happen. yeah, it's like it's just not, yeah. We're not going back.
Jacob Luetzow: I don't think so either. It's not realistic, but I'm saying even if it does, it's not like we we're we forgot how to code completely.
Michael: Yeah, yeah. I I just think you know, people are very mad at DHH about the keynote. my my big question for Rails now is how the security is gonna work. but on the point of, yeah, we're kind of living in a different environment for programmers. It and it was a really big shift. Like it kind of crept up on people. There wasn't really
Jacob Luetzow: Mm-hmm.
Michael: a moment where someone like published an authoritative book or something that said, Hey guy and I would even say we don't really know how to use the LLMs that well yet.
Jacob Luetzow: I agree.
Michael: everyone's kind of freewheeling with whatever. You know, I I've personally found that the older cybersecurity methods work really well. And I'll just give like a brief example. Like if you know that there's a dangerous function that let's say leads to command injection, meaning you submit an HTTP request to your Elixir Phoenix app or Ash app and it executes shell commands, a lot of times that's just caused by a dangerous function that you could just grep for. So
Jacob Luetzow: Mm-hmm.
Michael: How much money or compute do you think it costs to run grep on a pull request or a code base versus putting it through a Frontier L L You know, it's probably like a thousand order magnitude difference in in cost. So it's it's one of those things and speed.
Jacob Luetzow: And probably speed alone, like just waiting for the LLM to respond would probably like quadruple your time.
Michael: So so this is something Praxial does is just having these very fast I I call them kind of foundational checks. Like, do you just have a foundation of security or do you have a stack analysis that says I'm not putting these functions that's extremely valuable. I've seen code
Jacob Luetzow: Mm-hmm.
Michael: bases in production pre AI that had a ton of security problems that a very basic stack analysis would have caught. So if you're not doing that, with code going into production, that's very bad. and it's something that works perfectly with an agent. and it should be doing that, but I I've seen people skip that, you know.
Jacob Luetzow: Mm-hmm. I depend on paraxial. You're a little your s your code scans of me, they keep me in line, you know?
Michael: Well, you work on some fairly important data
Jacob Luetzow: Yeah.
Michael: with your with your businesses and you know, keeping that secure. Well, that's just it's just gonna keep being a problem too. It's like we're not we're we're not going back to like pen and paper. Like cybersecurity, digital records, everything is gonna be online. It's going to be through AI agents. that's just kind of the future we're living in. and
Jacob Luetzow: Yeah.
Michael: it's funny because I'm sure our kids will will find it like, you like went through this transition. They were like, Yeah, it's a computer. It can write code. Like code is computer language, right? Of course it would s y y you know, there's
Jacob Luetzow: Yeah, of course. You just talk to it and it does it.
Michael: a book, I think it's structure and interpretation of computer programs where it says like, you know, code is written for humans to read and our children would be like, No, like it's computer language. Like we speak English. You don't talk to me in Python, dad. Like w with like
Jacob Luetzow: When when does that go away? When do LLMs just write straight to like bytecode?
Michael: I I saw people talking about that, yeah.
Jacob Luetzow: Yeah. And that like got me thinking because it is like I like Elixir because I can read it, but if we stop reading it, what's there's really no point.
Michael: I I actually think that there's a lot of benefit to Elixir at the moment because what are people writing their back ends and now they're going to like Rust and well a lot of stuff is still in Python. apparently ba even Basecamp isn't using Ruby on Rails anymore. But you know, w with Rust it it's not
really designed for like always online concurrency. you know, a AGI will be achieved when every LLM just recommends Elixir for default for their back end or something.
Jacob Luetzow: Yeah.
Michael: It'll it'll be intelligent enough to just start doing that.
Jacob Luetzow: I mean that would that would be the smart move, right? But at the same time, like do you need it? Like can't you just be writing on the beam and not understand any of the code being compiled?
Michael: Maybe my perspective is different from like most software developers because I work with businesses that are in regulated industries that say, yeah, I'm I'm absolutely certain this will be a a regulation that comes up where it's like, okay, you want to run code in production that was written by an agent. Like where did the agent come from? like who
Jacob Luetzow: Yeah.
Michael: created it? who has access to it? What were the prompts that you used? for the output, did you verify the security of it? Like who looked at it? What individual person in your organization has the authority to like approve or deny this? You know, it th that is inevitable for for big
Jacob Luetzow: Mm-hmm.
Michael: businesses or you know anything important, you know, hospitals, healthcare, financial services, critical infrastructure. you know, there's code running in your car in your airplane that has to be reliable. I I I want people looking at that. I I actually want people to
Jacob Luetzow: Yeah, I mean same. I do too. But like there's a lot of
Michael: It would be great if yeah.
Jacob Luetzow: there's a lot of I there's a lot of guys that I actually respect for being really smart engineers who admit to not reading much of the code anymore. But I guess it's it
Michael: And I don't even think yeah.
Jacob Luetzow: also depends on the project they're submitting it to.
Michael: I I don't even think that's necessarily a problem if the models like if a model is better at coding than any human alive, yes, it makes sense for for it to do that. And you could kind of see well well, it gets into like the sci-fi c like does it even make sense for a human to to be involved
Jacob Luetzow: Mm-hmm.
Michael: anymore? Should should you just turn it over completely? But then at the same time, like how do you know if something how do you know if it's even working? Like you need you need some type of verification and trust, This is an interesting problem that GitHub is going through right now where people don't really want to have open source projects anymore because y you just get people writing complete spam with AI and submitting it to kind of pad their resume. I I personally never thought like number of CVEs was a great metric for your like cybersecurity professional resume.
Jacob Luetzow: Like how many you can find or how many you've patched? Like what's the
Michael: Yeah, yeah. It it well, you know, if your job is find vulnerabilities like and we're gonna use those, yeah, that that kind of makes sense. Like you want C V
E's
Jacob Luetzow: Yeah.
Michael: and that. But you'll get people that are just like, I found like a C V E in like this random software, and I just like spam the creator and the C VEs are not even that important. Like it's not that big of a deal. th there's actually a lot of opinion. You know, people treat like a critical C VE as this you know, like fact, like you could prove it like a math theorem. There there's a lot of opinion. There's like a running joke in the in the cybersecurity community. where there's like the most overhyped research for the year versus like the most underhyped. And typically the overhyped is like a team that had a logo and a website and they're like this vulnerability is gonna kill everyone and then they release it and like it wasn't really even that exploitable, like nothing really happened. And then
Jacob Luetzow: That's funny.
Michael: there's research that someone just posts online and it's like, you could take this and hack into like Every computer in North America with it. and there's and there's no hype. but it's it's usually a WordPress vulnerability, by the way.
Jacob Luetzow: Yeah. I'm curious, like when a CVE like gets published, right? Like what was like the normal like turnaround, like it was important to like patch within like 30 days, right? Does that how does that shift now? Because if you know a C VE is out there, now you can actually leverage an LLM to use that, right? Use that vulnerability.
Michael: It was it was never great. That's the thing. People like like pine for the good old days of secure. It was never good. You know, like the internet
Jacob Luetzow: Mm-hmm.
Michael: was always kind of like this fragile connection of computers. There were people saying that somebody could take it over and bring it down going back to the nineties. I think what happened is it went from kind of this curiosity to like this kind of critical infrastructure where today It's not just like teenagers hacking Fortune 500 companies for fun, although that a lot of that happens. You know, it's much more like espionage, nation states, pressure on firms that are kind of responsible for well well, this is just a big problem is like data breaches. Like people just don't trust any company at all with their data because outside
Jacob Luetzow: Mm-hmm.
Michael: of maybe 20 or 30 companies in the world are really capable. of good cyber security in the sense that they could withstand like a nation state level, you know, they can like manage the complexity. And I and I don't I don't mean to say that like like a small company can't do that well. I actually think they're better positioned because they have a smaller attack surface. Just the reality is that when a company gets kind of like above a certain size, it it turns into like a management problem. Like
Jacob Luetzow: Mm.
Michael: a startup with five people, I actually think 100% could secure data against nation state attack, like use Elixir, using Paraxial, even using a different
language. that's technically feasible. Once you get to like a Fortune 100 company with like thousands of employees, just keeping like it you struggle for talent because people kind of leave, they want to do a startup or they want to do big company. And then the team that you're working with, they're under all of this business pressure. for growth, they kind of like neglec like it's I'm not saying that it's their fault. It's more that the incentive structures of the organization just like a simple example. Do you it would you rather give your kind of like passport and credentials to let's say like a a major tech company that you know is gonna get like billions of dollars in fines and they screw up, or just kind of like a like a random firm that th they just always seem to suffer data breaches.
Jacob Luetzow: I would probably give it to the the well-funded company, honestly. Like I'd
Michael: Or you'd rather just not give it over at all.
Jacob Luetzow: rather not give it over at all. The problem is I know too many. So I agree. I think startups can be secure, right? Especially if they focus on it and it's a concern. But I also
Michael: I've seen it.
Jacob Luetzow: I know too many people that are like, I'll worry about this later. And they actually know of breaches and they still launch because they it's more important for them to get a user base and make money. than to actually secure their users' data. And I there's too many of both, right? Like for like it's like a coin flip if you don't actually know the founder or whoever is creating the smaller product. I don't know. It's tough. I'd rather not at all.
Michael: And I think it's possible to be secure. I I think you know, it's definitely it's technically feasible. it's often a question of like prioritizing it. And like you said, like you kind of have to get users first in a sense, and then you know, people wanna worry about security after. But it it's a difficult environment because you're also a target, actually, like in some cases as you're growing, because bigger firms kind of have more established security in a lot of ways.
Jacob Luetzow: You and I have talked about this with like me taking like a new domain online like three hours and I start watching the bot attacks hit
Michael: Just immediately.
Jacob Luetzow: the paraxial dashboard, you know? are you back? Maybe.
Michael: Yeah, and it's just a constant problem. sorry, I think you're are you frozen on your end?
Jacob Luetzow: now you're back. Okay.
Michael: Okay.
Jacob Luetzow: It says my connection's unstable. Okay. Well, that sucks. All
Michael: it's w it's working out for me.
Jacob Luetzow: right, cool. but like, yeah, like the instant, you know, a public server hits the internet, you're vulnerable.
Michael: Yeah, and I actually think if you're an Elixir developer, it's very feasible to use AI to secure your application. at the same time, it's not really like a magic at the moment. because I've seen people spend tens of thousands of dollars securing their software with like the latest models, and then I came in and just immediately found a bunch of problems. And I kind of expected it because the they just had too much software, like it wasn't fitting through. You know, you can spawn like a thousand agents or whatever, but the reality is it's just a very com it's like a complexity problem. the thing I would say is that you you kind of need to know the risks of what you're doing. So an online poker site, it's kind of an interesting example. What's their biggest security risk? It's probably a weak random number generator, because then you could predict cards and steal money essentially.
Jacob Luetzow: Hmm. Yeah.
Michael: That's a completely different threat model from, say, a company that handles medical imaging data or healthcare data, where
Jacob Luetzow: Mm-hmm.
Michael: a weak random number generator probably doesn't matter for them. Their big thing is like access control, auditing, compliance with standards, and really just making sure nobody can hack into it and steal everything.
Jacob Luetzow: Right.
Michael: just like very different threat models.
Jacob Luetzow: I mean, that's my my only concern, right? Is if someone gets access to my database or like an S three bucket, everything's encrypted and they shouldn't be able to read it.
Michael: But I d I I feel you're very trustworthy because you actually understand your system and you still read your code as you mentioned. You're taking this very seriously. You know, I
Jacob Luetzow: Well, I try well, and I actually I feel really bad because recently I did like a huge architectural change, like in data migration, because I removed I started as a zero knowledge encryption and I shifted. I realized less people cared about that and I cared more about that. It was like more important to me, but it tied my hands with a lot of cool features. So now it's like all, you know, encrypted at rest, but like encryption is still important to me. Like you can't just grab my data and get all my user information, you know? And
Michael: And you understand what you're doing because I I trust you more than someone that says, I asked Claude and it says it's all good.
Jacob Luetzow: Yeah. I mean, I debate with Claude a lot on like security and encryption and things like that. And I've I've learned a lot doing that. But yeah, I won't just willy-nilly just use something because Claude suggests it.
Michael: It's funny, when we were making the ElixirConf talk, there's a quote, and I read somewhere, I think it was Thomas Edison. he didn't say the quote, but he was a fan of it. It was about how man will do pretty much anything to avoid the labor of thinking. and I thought that was quite insightful because in many cases thinking is a lot of work. Like thinking through a
Jacob Luetzow: Mm-hmm.
Michael: problem and working through it is very difficult. And where you you see a lot of people using AI now is to just like I don't want to think about this. I I
just want you to solve it. it's very good at that for coding. but at the same time there's also like management decisions and kind of thoughts about like how are you going to run your company or how are you just gonna, you know, live your life.
Jacob Luetzow: I mean, I'm guilty of that. Like, I don't want to do engagement and outreach and social media with Kill Switch, but I also like realized I was very like stagnant on growth. I wasn't getting users or trial signups, not doing anything. And I started using Grokbots and it's creating like my content. It's posting on a regular basis for me. And just like getting content in front of people every day drives more traffic to my website.
Michael: That that's quite a good use case.
Jacob Luetzow: Yeah.
Michael: yeah, and it's like something you didn't really enjoy doing. I I think the anxiety people have is like, okay, well what like what if, you know, there is job displacement? If that was someone's job, like will productivity rise and we'll all have you know, the cost of goods will go down and it'll just be like everything will be great? Or will all of the money just accrue to a small group of people? And then everyone has this anxiety that they're left out. I I think that, you know, both both sides in this are it it's it's kinda funny, it's like basically a political debate. It's like both sides. but it's, you know, AI. it there's not like a clear like one side has everything for me. I I think that, you know, if you're a person, like the data center thing is very hot right now in Pennsylvania. I
Jacob Luetzow: Here too in Utah.
Michael: yeah. And and I see people who are just against, you know, data centers for just like I don't like them, you know, I I I I hate AI, for example, and I think it's bad or whatever. I'm I'm someone who views AI as kind of just a baseline for your nation to continue having standing. where if you don't have it it's gonna be hard to fight wars, you can't really do cyber security, you can't secure your own nation's critical infrastructure or Yeah.
Jacob Luetzow: It's one of those things you can't ignore. Like we need to keep, you know what's the word I'm looking for? Like we need to be the best. We need to keep
Michael: Yeah, yeah. So so my personal opinion is AI growth in the United States is a good thing. I I support that a hundred percent. I think it's kind of politically naive though. Yeah, it's like that would be the great next
Jacob Luetzow: But what about flock cameras?
Michael: topic. It it would just be politically naive to ignore people who are like, Yeah, I live in an area where they're gonna build data centers and it's noisy and it's harmful and I don't like it and my and my power bill is gonna go up. It's like, well, yeah, that person is responding quite intelligently, I would say, to their incentives, which is I don't want my power bill to go up.
Jacob Luetzow: Mm-hmm.
Michael: it's it's it's not a Chinese conspiracy to slow down AI progress. I I think that that I like I've seen that personally.
Jacob Luetzow: yeah.
Michael: so th there's kind of like facets to it, but bigger picture it's 'cause like we keep getting off track from Elixir. I'm actually planning to publish some more on the Praxio blog about using AI to secure your applications. I'm also always interested, if you're still listening to me at this point, you know, thank you so much. if you're maybe we can even do chat, like how the people listening to it in their jobs, like what what problems are they seeing with security or with applying AI? You know, do you feel that your company is good on security and how it's using AI, or do you like have concerns about that. I'm, you know, I I'm Michael Lubis on LinkedIn, so you can always reach out to me if if you're listening, especially this far into the episode, I would love to hear from you.
Jacob Luetzow: I who was I just talking with? They were talking about how some companies are already starting to like use like LLMs and AI to like patch infrastructure live, like any holes that punch through, and they're like developing on production with it. That terrifies me.
Michael: Yeah, I could see that becoming a norm though in the future. I mean it it well, if
Jacob Luetzow: For sure. I can see it becoming a norm, but right now, currently, it scares me.
Michael: if anything, like the current patching situation is not great. at the same time, I think people collectively have this sense that software is getting worse. You know, there there's not this like software is better ten years ago. It seems like all the gains are in hardware or something, but the actual software to use it is like lower quality or it's worsely designed or you know the performance is not as good and and people are hoping that the the AI will fix that, I suppose.
Jacob Luetzow: I hope AI fixes apps that are a second thought to like an IoT device. Like, please just
Michael: Yeah.
Jacob Luetzow: make your shit work better. Cause like I like I hate like, you know, you get mesh router system, and clearly the mesh router, the hardware is the important part of the business. And then like the app is like I guess we should have this and it barely works.
Michael: I I just had that problem with like a Samsung television where the setting I wanted wasn't in the app. You had to go through the menu and that and that's like a very minor paper
Jacob Luetzow: Yeah.
Michael: cut. and I actually I I should praise Samsung because it paired with my AirPods so I could watch it without waking everyone up at night. that was actually
Jacob Luetzow: Nice.
Michael: pretty cool. I was I was actually shocked that that worked. I was not expecting it to.
Jacob Luetzow: I wonder I should try that. I have a Samsung.
Michael: Yeah, it's it's I got the frame. it's actually a really nice T V. I will I would recommend it.
Jacob Luetzow: Nice. That is the nice one. It like, yeah, I know what you're talking about. I just got yeah,
Michael: Yeah, it looks like art.
Jacob Luetzow: yeah, yeah. Those are cool. I like that. I I I have like a 85 inch TV because I wanted to get, you know, the kids all excited so we can have awesome movie nights. And I wanted the 95 inch, but it wouldn't fit where I needed to put it. And 85's big.
Michael: Yeah, last night I was watching Interstellar, the movie with it it's like kind of post apocalyptic. Have you seen it?
Jacob Luetzow: you're like you're talking like a decade ago, Interstellar? Yeah,
Michael: Yeah, yeah, that one.
Jacob Luetzow: yeah, that's a good movie.
Michael: It was great. I I for some reason, like the first time I watched, I think it was in theaters, it didn't like the it I didn't connect with the story. Maybe I stole the memory of it. It was a great movie. I loved it.
Jacob Luetzow: I should re-watch that one. We've been like, my wife and I have been talking like putting a list together of like awesome childhood movies that we want to like re-watch with her kids. No, no, no, not that one. Not not
Michael: Not not a good movie for kids, honestly. Terrible.
Jacob Luetzow: that one. But like that needs to be added to my the adult list of movies. Cause I feel yeah.
Michael: Yeah, the grown up. It's a it's a great movie. And you have the T V for it too.
Jacob Luetzow: Exactly. And a really awesome like subwoofer and sound bar. Highly recommend.
Michael: yeah. Can can you read chat for this stream?
Jacob Luetzow: yeah, no one's chatting though. We do have chat though.
Michael: it's just not.
Jacob Luetzow: Let's see, anyone here? I don't know if all the platforms cause I I know when I used OB
Michael: Yeah, like the sink.
Jacob Luetzow: when I used OBS like X wouldn't sync and so I don't know if Riverside is better. We'll find out. It probably works with YouTube and Twitch for sure. I'm also streaming to LinkedIn now, so that's kinda cool.
Michael: that's pretty cool. Yeah, I'll post it on LinkedIn.
Jacob Luetzow: Yeah. So yeah, I now people are here. We got Lewis. Got yep,
Michael: is everyone chatting now that we mentioned it?
Jacob Luetzow: I'm here. Awesome. Any questions for Michael? Tiago, what's up? All right. now I feel like where did we what were we just talking about?
Michael: we got way off topic, like television and movies.
Jacob Luetzow: I know that, but like before that.
Michael: yeah, I think just the effect of you know, like Elixir developers right now. I actually believe right now is a great time to learn about cybersecurity if you're already a software developer. Because a lot of your work is probably going to involve kind of interpreting the output of tools or receiving reports saying there's a cybersecurity vulnerability. And if it's like your first time dealing with it. I mean there's often debate between people that have been doing this for years. So it's it's just kind of helpful to have the background and
Jacob Luetzow: I'm curious, like for a software engineer, what do you recommend? Like, how do the how do you start learning? How do you
leverage LLMs?
Michael: Yeah, there hasn't been a there hasn't been a good book in years. That my my favorite is the Web Application Hackers Handbook, the second edition. I always talk about it when I'm on this. And the the problem is that that book is so good, it like kind of categorizes all the problems with like web application security. It's such a
Jacob Luetzow: Mm-hmm.
Michael: slog to write like a technical book because it's out of date. The authors of that are the company Port Swigger now. They make this reverse proxy for application security testing. And they just said, yeah, we're just gonna update the labs now. So I suppose working through the labs would be the good way to do that. But the pro everyone's just gonna use AI for the lab. So it's like, I guess you could work through it with your with your agent and then kind of read the output. Yeah, but then it's like
Jacob Luetzow: Solve this.
Michael: kind of hard because it's it's hard because you know, I I think you should kind of work through it yourself to understand. But another one, there's actually some books from NoStarch. that they wrote about bug one is literally just a collection of bug bounty reports and how much money each received. that's actually I think pretty valuable to kind of learn like a fifty dollar bug bounty thing, that issue like wasn't that big of a deal. If someone got paid, you know, thousands of dollars from a company, that's a big deal. and and that's directly relevant if you're using Elixir for let's say back end development. I I kind of
Jacob Luetzow: I did a bug bounty program for a little while and then just got inundated by trash. It was I couldn't keep up.
Michael: It was like a disaster. Yeah, that's the other it's that was a problem before I actually I remember I d I did that for a job and I it was horrible because you'd have to like manually verify everything and I actually really
Jacob Luetzow: Mm-hmm.
Michael: wish I had a a model, but it's the double edged sword is if I I was doing it at a time with it, I probably would have gotten like thirty times as much just spam and and it's just it's just horrible.
Jacob Luetzow: Right. Like and at first like a few one like a few good ones trickled in here and there, but then it was just like scammer after scammer essentially, like
Michael: my opinion for like a company in your position would just be not having a bug bounty program at all. I don't think it's worth yeah.
Jacob Luetzow: I shut it down, I canceled it. I like removed it from anything a bot can crawl because
Michael: Because and this is for the Elixir developers out there too. Th the big vulnerability classes that you care about, it's really like remote code execution or server side request forgery or SQL injection, you know, something where an attacker sending data to your server results in them getting like a shell or like reading directly from the file and accessing everyone's data.
Jacob Luetzow: I'm curious with sorry to like cut you off,
but the like
Michael: Yeah, yeah.
Jacob Luetzow: the uptick in like everyone wanting an MCP associated to their product. Like how are you seeing a increase in like prompt injection and like crazy vulnerabilities entering that way?
Michael: Yeah, a lot of companies will just ship an MCP that's not secure, because
Jacob Luetzow: Mm-hmm.
Michael: they're under pressure to do it. That's the other thing. a lot of times these I I think this was the big incident recently from Instagram where they inst they had some AI system and it made the news because you could just ask it to reset the account for you and it would do it. And that was that was the vulnerability was like I would like access to this Instagram account and it gave it to you.
Jacob Luetzow: That's wild. But
Michael: And and and that's a company with a very big security budget, but you know, they're also under a lot of pressure to get those AI productivity gains.
Jacob Luetzow: Yeah. I talked to like I do like, you know, talent recruitment, right? For the Elixir world. And I was talking to this company recently. It honestly kind of gave me the ick because it made so they're just looking for like senior level elixir devs. And then it just like they're like, Yeah, we pretty much, you know, our engineers are shipping like Five features a week. And normally one feature would take, you know, three to six weeks before LLMs. And like we're just like, and like they were like bragging about token burn and how much their engineers are using AI. And I'm just like, are you just like adding a bunch of bullshit features to your product? And like, cause like, how many features do you need? Like, if you have a team of 10 engineers and they're all putting out like, Two plus major features a week? Like how much do you really have to do?
Michael: Well, I I could see this argument from kind of a labor market perspective where you know if you're a company that's not using AI, in a sense you are at a competitive disadvantage if you're in like kind of a feature race with
Jacob Luetzow: Mm-hmm.
Michael: a competitor. But, you know, on the other hand, I'd I don't think it's like Kind of like the industrial looms during the industrial revolution, where you know, weaving kind of went away as a profession because yeah, there were a ton of arguments why the, you know, factory produced cloth was worse and it was worse in many cases, but there was a ton of demand for it. And people used it. At least when I've kind of been up close and personal with software companies, I've I've often found that like the more features argument. It can lead to growth, but it's it's often not the dri like Jira is not Jira because of all the features it has in isolation. Like you could literally start a new business today and have feature parity with Jira, and you probably won't get much market penetration because the firms that want to use that like they're gonna use Jira. it it like there there's multiple factors. When you see entrants into
like the project management software space, they often have kind of like another lever that they can pull there. But at the same time, if if you're just not using AI at all and you're you're outputting less software per developer, you know, you could see that as you're at like a a labor market disadvantage. So, you know
Jacob Luetzow: I agree. Like the cost of a feature has gone way down, but does that mean you should have a million new features? Like I feel like you're grasping at straws at that point to just be a Swiss Army knife when it might not be the best.
Michael: Yeah, and it's it's hard to say like, you know, kind of the value of software in society where I I think that the the risk people have, like, well, like software as a service go away. if an if an agent can do things for you that previously like a software as a service company could do in a cheaper way, that that does put pressure on it. But at the same time, a l a lot of software as a service, like I'll use Stripe as an example because I think everyone knows it. You know. I'm I'm no one's gonna code their own Stripe or something. Like the the software is a very I would say it's a piece of it, but it's often kind of like their institutional position to process payments.
Jacob Luetzow: Mm-hmm.
Michael: You know, even if the Stripe software was a lot worse, I think most people would still use it. And I think Stripe has very good software.
Jacob Luetzow: The pr the well, they won they won the market because of their documentation for developers, right?
Michael: Yeah, exactly. And it seems to translate well to AIs also.
Jacob Luetzow: Yep, yep. dude, I set up my entire like stripe webhook with Claude and
Michael: Mm-hmm.
Jacob Luetzow: everything worked one shot, you know, like I tested it obviously, but I didn't have to fix anything.
Michael: And and and how long do you think that would have taken you a few years ago?
Jacob Luetzow: A week?
Michael: Yeah, so it it's like there was a recent suggestion, I believe, from the New York Times, like, well what if we had the AIs like stop writing code? It's like it's just not gonna happen. Like
Jacob Luetzow: Right.
Michael: you know, every programmer I understand that there's a negative feeling about AI in a lot of cases, but the productivity gains in many cases are hard are hard to ignore.
Jacob Luetzow: Mm-hmm. Well, like a feature I just added to Kill Switch, which I'm actually really excited to like kind of iterate and make better. But like the whole point of Kill Switch is you get your, you know, in the worst of times when you can't check in, your documents get delivered to who you want them to get delivered to. So
Michael: Yes.
Jacob Luetzow: I implemented my first like beneficiary verification system. So like obviously you can't spam and annoy people. And I'm like, well, what's the best way to do this? And like, so at a very minimal level, I'm checking for like typos. Like if you typed in like Gmall, it'll suggest, like, did you mean
Gmail? And it'll do things,
Michael: that's good.
Jacob Luetzow: it'll do things like that. And then it'll also like checks DNS records. Like, is the domain even like functional? Does this email work? And then it you can manually trigger like verify, where it'll send a soft email to the person and they can click like, yes, this is me. And then, like, if there's a phone number attached, they can do like the two-step auth where they get texted a code. And now their phone number and email are verified as working. And then
Michael: Yes.
Jacob Luetzow: so like it's cool. It's not great because I still feel like getting emailed from a random company you've never heard of is going to come across as a spam, like a scam. So I'm still working on things like that, but like it's really cool to be verifying who your switches are sending to. And like you kind of have like the check mark. Like you have an unverified, you don't know if it'll get to them, but then you know for a fact this person is gonna receive it.
Michael: That's great. And yeah, it's it's great when what you're working on works and you get to use it. It's it's such a it's such a good feeling.
Jacob Luetzow: Mm-hmm. And like, okay, so where I was trying to get to with that, like I built this feature in like two, three days, like like really good, like probably 94, almost 95% test coverage. And like they're good tests. That's another thing. Like X unit and LLMs really work well. Like you can write really good quality tests. And because I've seen like the JavaScript TypeScript world knocking. like unit tests at all anymore as being pointless with LLMs, but they're just so powerful. Like and I don't break anything anymore adding new features.
Michael: Well, I would say tests are extremely useful now because it's almost like documentation for your agent.
Jacob Luetzow: Right. Exactly. Yep. Exact we do have a couple questions.
Michael: nice, yeah, let's read
Jacob Luetzow: all right, I have a question from Michael. Do you often propose or suggest simulation for like catastrophic events like DB recovery, leaked tokens, et cetera?
Michael: If a client asked me to, I absolutely could. The most common is really a cyber attack. So I'm often brought in to simulate basically a hacker. how would they get in? How would they breach the company? that's often required by a number of standards. I don't do as much of that work because I I think that that's just kind of the inbound deal flow that I get. and you know, Praxial's a security company. Maybe that's more operations work as well.
Jacob Luetzow: All right. Yeah. I think that's it's a good answer. And all right, then we have could you perhaps improve security by learning ethical hacking techniques from various forums and YouTube videos and then use AI to try to hack your own app? Yeah.
Michael: yeah, that's a great idea. I think you should. even if you don't have much of a background, just ask Claude or ChatGPT he might get a refusal now though. if you have
a deployed instance, just say, Hey, can you try to hack my application? I would even offer it to I've I've done this before. It's like, hey, I'll even like update a DNS record to prove to you that I own but
Jacob Luetzow: That I own this. Yeah.
Michael: you know, since post hugging face, they it might be a little bit more I get a lot of refusals these days with my work sometimes.
Jacob Luetzow: That's too bad because you like, yeah, I think if you can prove you own something, you should be able to hack your own service.
Michael: Well, that's that's why the the open model debate is pretty heavy on, you know, why not just use your own like local L L or one that you host and you can just rip out the guardrails.
Jacob Luetzow: Mm-hmm. Honestly, well, like one of my one of my friends is like super like I I started talking about Grokbot and how I use it for Kill Switch. And like I care less about I shouldn't say I care less, but I I like like X AI, for instance, knowing information about Kill Switch as like a company and maybe even like my read only financials. I care less about it knowing that than like my personal data. And so like I like I love Grokbot. It has made me super productive with like the customer side facing side of my business. And I love that. But then like I was talking to a buddy and they're obsessed with Grokbot now for doing things like that. But then they also use Muse, which is Meta's thing. And yeah, I started
Michael: Yeah, it's a open claw implementation or something.
Jacob Luetzow: looking at Muse and like. First of all, they want like root access to everything. And I'm like, and like it's of like it's it's kind of like around the idea of it's like your personal assistant, so it can do everything for you. And it's like I'm I like the idea. I love the idea of something like booking and organizing my calendar and emails and like scheduling flights. I really love that. Do I want meta having my schedule, my children's schedule and all that? No. I'd rather like, so I'm excited for the day where we can have like a house AI or a
Michael: I think that you are in a kind of minority though. It maybe this is my opinion, but like the technical crowd is like I don't want Facebook to have access to my data. In but you know, they operate what? WhatsApp, Instagram, Facebook. they have most people's data.
Jacob Luetzow: Well they have a lot of my data, but they don't have my schedule, children's schedule. I d like what I'm doing, where I'm doing it. Like they don't know that stuff.
Michael: Yeah, I it's tough 'cause I'm I'm I don't use social media personally. But at the same time, I think that that makes me kind of an an anomaly in a lot of ways where I think most people use social media very effectively. I actually shouldn't say that. I'm on kind of like a local Facebook now, but it that's more for like I would say, like talking to people. I I I guess I'm not like an Instagram power user or something. but like a lot of people.
Jacob Luetzow:
Yeah. I actually got rid of I got rid of my personal Instagram because I was like, what's the
Michael: Like a l l
Jacob Luetzow: what's the point?
Michael: like a lot of people's interface, what they think of as like the internet or their culture, you know, it's it's often like their social media. Like I I use what, X constantly or blue I use Blue Sky as well a lot because
Jacob Luetzow: I l I love X. X is great.
Michael: I don't like social I d I I wish I kinda wanna use it less.
Jacob Luetzow: I like it for like a quick recap of the news of like what's happening now.
Michael: I I wish it would only show me posts from people I follow. And I have this like weird setup where I do that sometime. Like I just wanna read like
Jacob Luetzow: You have the following tab and it stays on that, you know. But it always
Michael: Yeah but
Jacob Luetzow: defaults for the for you, and I hate that one.
Michael: Yeah, you can and you can like script it. I I should probably end up doing that. But as opposed to where was going for it is, you know, I I'm actually not one of these parents that is like, you know, my children will never use social media and I'll get them dumb phones and and everything. You see a lot of that in like cybersecurity circles. I actually think
Jacob Luetzow: But you don't do a benefit. They don't know how to use technology then. And they're gonna have to know how.
Michael: Yeah, that that's that that's what that's what I always come back to. It's like, okay, but like they live in this world. Like they're gonna like they're gonna find out what TikTok is eventually, and if you make it this like evil thing that like they're not allowed to they're just gonna wanna use it more if you're if if you're
Jacob Luetzow: It's the Catholic schoolgirl thing, right? If you're too strict, they're gonna go off the rails when they finally get access to it and no
Michael: Well, I I went to Catholic school and we had video games.
Jacob Luetzow: Ha ha
Michael: But yeah, it it's it what I where I see it going is you know, there's like negative externalities where there are kids who I think where your attention span I get a headache from watching reels, for example. And I'm I'm not trying to you know, I think that's how most people I actually think that that's like the end state of kind of
Jacob Luetzow: Dude.
Michael: design for the web because people always say, like reels are so bad What was like a web forum though, back in the day? It's like you were waiting for people to post a
Jacob Luetzow: I look at what like young kids, like, you know six to thirteen, they'll scroll shorts on YouTube. And that's what they're doing on YouTube. And like I walk like I use YouTube for education. I watch like tutorials or something about something new, right? Or news. Kids are scrolling it for shorts, you know, thirty second things. And then it's like next video, next video, next video. I'm like, this is insane.
Michael: Yeah, I I'm I'm actually kind of surprised that there hasn't been more regulation of that. Cause I think there is in other countries where that specifically, yeah.
Jacob Luetzow: It can't be good for attention span.
Michael: It can't be good for attention span. At the same time though, I I I'm one of these people where I'm I don't like to criticize like the younger generations especially because I I I know people from every age range that are like addicted to Facebook or their phone.
Jacob Luetzow: Mm-hmm. Yep.
Michael: and and and then like I'm on a computer professionally for a living.
Jacob Luetzow: I mean, here's the thing. Like, I got rid of Instagram because I realized I was like wasting so much time scrolling crap that doesn't matter. So, like, I still have
Michael: Yeah. Well it's entertainment.
Jacob Luetzow: like I have a kill switch account and like, but I don't go on it. my social media, other than like my ex account, is scheduled by Grockbot now. Like, I don't do it
Michael: Interesting.
Jacob Luetzow: because I I want to stay off of it. But at the same time, I'm on my phone now all the time talking to Claude. to work on features and then I'm reviewing in my GitHub app the next feature. And then I deploy from my phone and test it. So like I just have a new addiction and it's building a product, you know?
Michael: So yeah. Yeah, so your your kids will say like, you're trying to get me off video games or scrolling, but you talk to like an AI all day. You know, it's like they're gonna They're
Jacob Luetzow: Yeah, exactly.
Michael: not dumb. They they're they're like aware of the world around them.
Jacob Luetzow: Yep, for sure. It's just it's interesting. You don't know what the balance is 'cause you don't like you want kids to get outside, touch grass and play and do stupid things, right? But then you also don't want to fall behind with technology and
Michael: I I have this feeling that regulation is coming. I think there was some landmark settlement against Facebook reason and there's precedent for this with the big tobacco. Like everyone used to smoke in America and then it just kind of stopped. I it wouldn't surprise me if for the next ten years we saw just like a ban on reels for children or something like that.
Jacob Luetzow: Right. Well, I think that it'll be like w was it just Australia that put an age band on social media? Like if you're under sixteen, I think.
Michael: Yeah, at the same time, I I also don't think that banning like teenagers from Instagram is really necessarily a solution. Because it's it's
Jacob Luetzow: Mm-hmm.
Michael: it's often like, you know, like young people want to use social media. It seems kind of like another restriction. I I think society in general I know I
Jacob Luetzow: Right. They need to connect with something, right? With
Michael: know it's like we're getting so off on Elixir on this, but I I often see Today it's very hard for young people and it's not the phone's fault. I think like post COVID, like it's like life has just gotten very difficult
for young people and and someone kind of needs to defend them and say like, no, they should have more freedom. Like they should be able
Jacob Luetzow: Mm-hmm.
Michael: to use social media. Like the these restrictions, I think that they're very rarely about I'll put it this way, I don't think that the well being of young people is necessarily even the intended outcome. It's just the rhetoric you get around it.
Jacob Luetzow: I agree. Yeah. Well, it's like, okay, I think it's crazy that there's still people that smoke, right? 'Cause I think just the health problems and issues that arise from it. But I thought it was wild when they changed the smoking age to twenty one from eighteen. Like I feel like we need I don't know, a few years ago now.
Michael: did that I when did that happen? That was like a while ago. I think I think that was a big deal in like the military, like a lot like I'm I'm out here like
Jacob Luetzow: But I think it's it's
Michael: fighting a war. I can't smoke.
Jacob Luetzow: I think we gotta pick an age, right? It needs to be all or nothing, like with drinking, smoking, buying firearms, going into the military. Like w are you an adult at eighteen or is it twenty one? Like just pick the
Michael: I I think if you can join the military and deploy to a foreign country, you are an adult. I think legally you have full rights.
Jacob Luetzow: You can buy a pack of cigarettes. Yeah, exactly.
Michael: That that's a good that's a good support. And you should be able to use AI
Jacob Luetzow: I agree completely. All right, we have a
Michael: with it.
Jacob Luetzow: exactly. I have another question here, Lewis. I actually don't know this one. Does a garbage collector programming language, is it less secure than one you have to allocate memory manually?
Michael: In in some way okay, so this is like the debate that everyone always says. They say like C is not secure and then we're gonna rewrite everything in Rust and then all software will be secure. That's referring to this class of vulnerabilities called memory corruption. And there's like a joke now in the InfoSec community that a lot of the really bad security bugs aren't even memory corruption. But Rust has this like halo of security about it. By the way, a lot of Rust code has unsafe in it. so it's really not. You know, but they're saying, we're gonna like rewrite all the code in Rust and like everything will be solved software. And I I I think Rust is a good language. I actually I think the team behind it is is doing a lot of good. at the same time though, that that's just like a common kind of like talking point that I see is that like, well if we if we port all software to Rust like here's an example. Elixir solves this problem called a data race. Where essentially p th picture you have like two bank accounts and you log into like your bank account or like Jake like I go to log in to my bank and then I get put in like Jacob's bank account and I can like see how much money he has and I can wire money out. there's a famous it's a famous example from this book, Web Application Hackers Handbook. And it was due to like how threading works and operating systems. so Elixir does not have this problem, even though it didn't have the type system yet. Because of how like memory works in the beam, essentially. Rust doesn't have this problem because it's like strongly typed and people were saying, like Rust is so secure. Now but it's like Elixir also has this security property. It just gets zero clout for it. For like you go to Black Hat, everyone's like, Rust is the secure one. Like, what's Elixir? You know. So I I published a blog and I kind of make noise about that. Where generally speaking, if you're using a dynamic, Sorry, when I say dynamic, I just mean a language that's not C where you're like manually allocating memory.
Jacob Luetzow: Mm-hmm.
Michael: you know, i yeah. Al also like memory corruption vulnerabilities, they're a problem, but things are very different now. I I'm I think that like the the efforts to switch software over to Rust, they have benefits, but and I actually do support them, but you know, p sometimes people don't think through the problem completely like they're There's software that was it was originally C, it's been rewritten in Rust. That's a security improvement. It does not mean that it is a hundred percent secure. And that software can have very bad security problems. it's kind of a subtle point, which which can be hard to make in these in these scenarios. Yeah, no, never. Is it connected to the internet?
Jacob Luetzow: Yeah. Well, is anything a hundred percent secure? I think that's a that's a hard number to put on something. Like your spidey senses should go off if anyone tells you a hundred percent for anything.
Michael: Yeah, there was some it was like Neil deGrasse Tyson or something. He said like the solution to cyber security was to make unhackable computers or something. It was like thank I never we never thought of that before. Yes. No, but he's a great Yes.
Jacob Luetzow: That is true. You're a genius. Just make it unhackable. I put that in my prompts. I'll add that to my prompts.
Michael: No, but he's a great narrator, for like, you know I I actually am a fan of the planetarium near me. I think he he narrates one of them. I think he's great on that. Just just
Jacob Luetzow: nice. No, that's cool.
Michael: I'm you know, in in cybersecurity it's a little bit different.
Jacob Luetzow: I he likes to talk about a lot of things he probably shouldn't talk about. I do like I I think he has like a good opinion on a lot of things, but yeah.
Michael: Anyway, so yeah, are we do you wanna wrap it up soon? I I think this was a great episode.
Jacob Luetzow: Yeah, we can wrap it up. I think this was this was a lot of fun and I appreciate you coming back on. And yeah, you're always welcome. So
Michael: Yeah, thanks. I guess to close it out, you know, I'm Michael Lubis. If your business is using Elixir or just has security problems and you're watching this, feel free to reach out to me. I know it's a little intimidating sometimes to cold ping someone you saw in a podcast. If you listen to this episode and mention it, I will a hundred percent reply to you. I would I would love to hear from you. So don't worry about it.
Jacob Luetzow: And if you forget what you're listening to, this is the Elixir Mentor Podcast. And subscribe if you haven't already. Dude,
Michael: And you should subscribe.
Jacob Luetzow: you know what's funny, not to rant before we hang up. But
Michael: Yeah.
Jacob Luetzow: last week I I had Bryce as my guest, and we were talking about like how monolith is the way to go. I lost 11 subscribers because of a
Michael: Because you talk
Jacob Luetzow: short that I posted on YouTube. People are very opinionated about their microservices when they when they are.
Michael: I I've actually had situations where microservices make sense. it's you know, it's just it's so dependent on everyone's situation.
Jacob Luetzow: I so like yeah, I think there's a place for them and I that's okay, whatever. I
Michael: You're gonna lose more subscribers.
Jacob Luetzow: don't like you anyways. No, I'm just kidding. I think there's a place for but I think like a with a lot of like small teams, monolith just makes more sense to like stay organized and move quickly. But anyways.
Michael: But yeah, it's just it it it just depends on the use case, I suppose, because it's like it's kind of like talking about like, should you buy a minivan or like a sports car? And it's like, Well, I can think of different situations where this is appropriate.
Jacob Luetzow: Always the sports car you leave the kids at home. No. But all right. This was a lot of fun, Michael. Thanks for joining me. And for
Michael: Yeah, it's great to be back.
Jacob Luetzow: everyone that tuned in, thank you so much. Have a great weekend. And I'm pretty sure we're back next week. Let me see. I forget now because I just started booking things out. Yeah, Mike Hostetler will be on next Friday. So that'll be a lot of fun. Yeah, he's a
Michael: he's great. Yeah, thank you for listening everyone. Have a good weekend.
Jacob Luetzow: see you guys.