The one where we vibe code holiday cards | Season 5 Finale

0:05Welcome to Dev Interrupted. I'm your host, Andrew Ziegler. And I'm your host, Ben Lloyd Pearson.

0:39tuning in for season six at the top of next year. Yeah, absolutely. It's a great opportunity for us to relax a little bit, have some fun, you know, do some cool things. But of course, you know, it wouldn't be a dev interrupted episode at this point without a news roundup. And we also have a gift for everyone listening to this episode, which we're going to post later. You'll find out about it towards the end of it. Listen to, you know, just listen to the end for more details. But I'll give you a hint that this episode does have a part two and it might be on a video platform. Andrew, where do you want to start?

1:11Should we should we just get right into the news or do you want to maybe talk about some reflections on this year with Dev Interrupted? The news? No, we have to we have to mix it up a little bit. It's our holiday episode. So we have to start with, I think, some reflections. And there's a few things I think that we can both reflect on. So, Ben, you know, what are you most thankful for when it comes to Dev Interrupted? Yeah, this is an easy one because, you know, we get to speak to so many really smart people about software engineering, leadership, like all these really cool topics. And of course, AI.

1:40And I think as AI just like completely transforms so much of what we do, I'm so thankful that we get to have all these experts come into our, you know, into our house or our space and share really awesome things with us because I've learned so much just by hosting these people. So yeah, I'm super thankful for that. What about you, Andrew? I totally agree. Really thankful for the guests, but I think what I'm most thankful for is just how lucky I feel. Because when Dev Interrupted, it's like I feel like I'm at the right place at the right time in our industry when things are changing and things are just so interesting week to week.

2:15And with really amazing people too, both you and people behind the scenes, but our amazing guest network. So all of these things together have just made an amazing year. So just thankful for how lucky I am to kind of be in a place where all those things are coming together. And what are you most proud of in the past year for both the show and maybe even yourself personally? I think what I'm really like the most proud of is how we've really turned Dev Interrupted in the last year into, I mean, I don't want to say a content factory because it kind of has negative connotations, but it's a system.

2:45Like we've really built a robust system for generating new content, for trying out new ideas, for experimenting with the podcast and our sub stack and all of those things. And it's really allowed us to iterate like a substantial amount in the last year. Like if you've been a long-term listener, you've probably noticed some format changes, particularly over the last 12 months. And, you know, I'm just really happy that we've been able to pull that off because it's enabling us to take on bigger challenges and do more interesting things. So really awesome to see. So what about you, Angie? What are you proud of this year?

3:18I'm just proud of like all of the energy that we were able to put into kind of automating away the toil of the things this year. That way you and I have been able to talk so, so much and our producer, Adam, as well about the strategy behind how to bring really great conversations and really amazing content to folks. And so I'm really proud of our ability to like iterate and get to that place where we can strategize that. And also just proud of us every single week for, you know, doing these new segments and covering things when it's easy and when it's not. You know, I think that it speaks a lot to like us showing up here and like what we want to do and why we're so passionate about it.

3:52And I'm just, you know, really proud of everything that we've been building last year. Yeah. And, you know, what are you looking forward to? Now we talked about what we're proud of, but like looking forward to next year, Ben, what are you most excited about? Well, you mentioned automating the toil. Yeah, we have done a ton of that using AI and workflow automations and all that stuff. And, you know, we're kind of at a point in the, you know, just in the content production industry where we are starting to feel those multiplicative improvements from AI. There's just a lot of toil that is just disappearing and all of that.

4:26And I think the next year, it's just going to get better. We're going to be able to have more automated workflows, better production systems so that we're just constantly focused on bigger challenges more and more. And it kind of relates to both of our last answer. There's just so much more like greenfield that I see to just build these like AI driven workflows to help us do better content and do it at a higher frequency. So yeah, that's pretty awesome. Like what about you, Andrew? What are you looking forward to other than more vibe coding sessions? Oh, I'm looking forward to more vibe coding sessions.

5:00I think that's top of my list because in the last year, I think what we've learned, you know, talking with so many engineering leaders, I go back to some really high profile guests from even really earlier this year, like Dr. Ashuri at IBM and how she said, you know, folks are evolving from being action doers to decision makers. So you see this elevation in the IC skill set of them becoming more managerial and like how they think about what they do. But you also have this really amazing revolution right now happening on like the leadership side of things that are vibe coding now every day, building out prototypes to help their team move faster.

5:36And I think in 2026, I'm really excited to explore more of those and learn about how folks are solving communication and collaboration problems with code as leaders and as managers beyond just using AI within code itself. So I think that's like something I'm really excited to explore with our audience and our experts because we got peaks of that this year. But I think we're going to see even more of it next. Yeah, so stay tuned. Next year is going to be another great year for this show. I'm really excited to see all the stuff that we have in the works. But yeah, let's dive into the news now because that's what we love to do every week here.

6:11So this week we have some pretty awesome stories. So we'll start with MCP agents need real safeguards. We have killing stale engineering process dogmas and big tech sours on remote work. Andrew, let's just start right at the top with MCPs. And I know you love to talk about this, so I'll let you lead this one. Okay, so this is an article called It's Not Too Late to Roll Back MCP. is from Theory Ventures blog, and we've covered some of their work here in the past. This article dives into why, you know, AI agents are currently limited to making easy decisions within domains. And in things like software engineering, we have these pre-existing safeguards that were there before AI showed up.

6:54Things like Git and code review and database transaction layers that make decisions more reversible and less consequential. But then you get AI on the scene and you get MCP. Now you have applications deterministically calling all of these tools, and that is less easy to reverse. So this article calls out how MCP has a lack of system boundaries. It lacks like trust and support between intersystem kinds of tool calls, you know, and it makes it hard for difficult to prioritize the work they do. And it makes it impossible to safely reverse high consequence decisions. Like, that's why you don't see MCP transferring money out of your bank account, right?

7:34Because while it's easy maybe to reverse ACH something, it's less easy to reverse an agentic decision as part of something else that transferred your money away. It makes the problem a lot harder to tackle. And so this blog post tackles how we need to extend MCP to support these distributed transactions, develop a standardized criteria for people to like orchestrate the outcomes and the decisions and the tool calls that agents make. Because right now MCP is operating on a blind trust layer and that's just not going to scale, which is what this calls out. And I agree with this article. I think there's a missing decision management layer in the mix here.

8:12And I like how it framed agents work right now as having a lack of reversibility. I think that speaks to maybe some people's hesitation around using MCP because ultimately, like somebody has to be culpable for certain actions. And so we're still not at a point, I think, where an AI can fully take the culpability. And there's a chart here as well that I'll let our audience check out when they read this article. It shows these domains of decisions on like a vertical and a horizontal axis. And you can kind of see why wiring money to a stranger is like highly irreversible and highly consequential versus maybe some of the other decisions that AI can make for us.

8:48Ben, what do you think? Yeah, I mean, you and I both know how MCPs can both be like amazing and terrifying at the same time. Like I kind of think of them as almost like hooking up a nuclear reactor to your software. On one hand, they make it really easy to connect GPT to services that provide data and tools that extend their capabilities. You know, we both experienced this firsthand, actually, just recently with Linear B. We launched this unblocked campaign where we actually use Claude connected to the Linear B MCP to prototype an end of year celebration for our customers who use Linear B AI to increase their team's productivity.

9:27So, you know, it's a really powerful tool on one hand. But then on the other hand, you know, MCPs can kind of explode in your face. Like if you have poor configs that can lead to like overfilling the context window, you know, it's not deterministic. So you can't rely on it to produce the same result twice. And it needs a lot of context and guidance to do some things like the way that you actually want them to do it. So I really like how, you know, this article breaks down this onto two spectrums like you mentioned. You have high versus low reversibility and high versus low consequences. I just love two-axis spectrums like that.

10:03And decisions that are low consequences and highly reversible are what the author calls the comfort zone, where MCPs, like you said, they can make decisions, but they aren't able to move decisions from a state of discomfort to comfort. So if something is in the comfort zone, they can do it. But if it's in a discomfort zone, you probably shouldn't trust it as much. Yeah. So I think like the key takeaway here is that MCP gives agents capability, sure, but it doesn't give them safety. And safety requires orchestration, validation, rollback semantics. These are very subtle layers of the protocol that have not evolved yet.

10:40So I do think that there's an opportunity for people to implement MCP in a more forward-thinking way. I think that's actually what this article is really digging at, is that we have to come together as a community of builders that use this protocol to evolve it ourselves. And of course, you know, this article also shouts out how MCP is part of the Agentic AI Foundation now, New Lakes Foundational Organization, founded alongside many other foundational AI tools. And so if you're reading, if you're building it, anything with MCPs, definitely check out this article. Yeah, absolutely. All right, next story.

11:10Your job is to deliver code you have proven to work. What's this about, Andrew? Yeah, so this article really hammers on what we just talked about a moment ago, is that software engineers are responsible for delivering the code that they produce, whether they generated it themselves or someone else made it. Just like how a moment ago we said how the agent can't make a decision for you that's outside of your comfort zone. The same thing goes for engineers. you can't ship code that an AI wrote for you that you don't have awareness and understanding of. And so this article really hits at the importance of understanding your code, but also using techniques like manual testing to make sure you understand how your code works along the way.

11:52And really calling out the danger in the rise of AI coding tools like CloudCode and Codex CLI just makes it really essential for engineers to be heavily and aggressively reviewing the code that they create and put out into the world. And ultimately, they remain responsible for keeping it tested and working and fitting for the reasons that it's created in the first place. What do you think, Ben? Yeah, so this article closes with what I think is an incredibly important point, and that is that the human provides the accountability. Like, it's perfectly fine to have AI generate code for you, but the code is always your responsibility at the end of the day.

12:29And I actually recently heard from an engineering leader who said their organization banned the phrase, Claude wrote this, because too many of their developers were using it as a way to offload responsibility. And AI will continue to get better at writing code, which means that us humans, we can focus on other bottlenecks to ensure that code working properly is one of those things that we can focus on. But I think one common example is that automated testing is more important than ever. And all PRs should bundle testing improvements with them. One great example of this is Coinbase. Actually, they've been making the news recently, and we've covered this, about how their CEO claims that more than half of their code is generated by AI.

13:14A big part of that is that they apply additional rigor to AI-generated code. So that includes a strict requirement that all new code must have 100 % unit testing coverage. So these are the types of things that engineering leaders should be thinking about right now as we you know, culturally shift into using these tools more frequently. So our next article is about the five engineering dogmas that it's time to retire. Ben, what is this article about? Yeah, so the article challenges five common engineering practices, things like reliance on external packages, mandatory code reviews for all changes, using two to four week sprints, feature flags, you know, stuff like that.

13:55I cite some examples of actually one that I really loved of some NPM incidents where, you know, a package that gets 160 ,000 weekly downloads is just a few like four or five lines of like sloppily written code that was written as a part of like someone's learning experience. And, you know, the author of this article argues that these dogmas should be regularly re-evaluated as they can all introduce inefficiency or complexity if they're applied rigidly rather than tailored to your team's context. So, you know, you know me, Andrew, I always love contrarian takes. You know, I really love that example that they had of, I think the package was called is even.

14:35And it just like uses, it imports another dependency to like determine if it's even and then just outputs results. like results. It imports is odd and it just returns the opposite of is odd. Yeah. Yeah. That makes you wonder what is odd imports, you know, and you know, and I have actually have a good friend who he works for a software development consultancy. And he once told me about this client they onboarded where he spent the first week searching through hundreds of dependencies that their last developer added to the stack just to find all the modules that could be replaced with a few lines of code and there was a surprisingly large number of those modules so there's a lot of real problems that are being outlined in this article yeah there's a bunch of ones in here that i thought were interesting one of them the first one about like uh not using agile that one like or waterfall that one like made me shudder it's like yes like obviously the engineering dogma of like if you're still a huge enterprise and you're trying to build things with waterfall i think you're going to encounter a lot of problems in today's speed and also having closed tool chains, like just using the same tools over and over and not evaluating new ones.

15:38Definitely a dogma that I think should be retired. There was one in here that was interesting about like team heroics, right? About like counting on the one individual to rise up and to like do the most and to like produce the most output for the team. This one was interesting because I think that's less of a dogma and more of just a phenomenon of workplaces and how people interact with each other and their outputs and i think that like uh in the age of ai those folks are actually going to stand out even more and go even faster um and produce even more code and be even louder and so i think it's really interesting how the dogma maybe needs to retire but that itself is more of a reflection of the workplace that we live in but what do you think about um like the heroics that that this thing kind of talked about that definitely stuck out to me something that i think stuck out even more though was the examples they had of mandatory code reviews for all pull requests because this is something that we think a lot about as part of our work for Linear B.

16:36And just to open the hood a little bit into Dev Interrupted Operations, we just wrapped up an interview with Ori Karin, the CEO and co-founder of Linear B, about his predictions for 2026. And that episode is going to come out early next year. But a preview that I can give now is we're entering this phase where the increased volume of AI-generated PRs will necessitate that we stop treating all PRs as one size fit all as many organizations do today. You know, some pull requests need very deep scrutiny. Others simply need to be merged because the risk isn't big enough to spend a lot of time on them.

17:11And I think one of the big challenges that engineering leaders will face in the next year in particular is automating the process of determining the difference between those types of PRs. Yeah. Understanding that like not all your code is created equal. They need different levels of care and attention is a really important discovery to make because right now, like everyone is focused on creating code, the code generation. We're less focused on these other parts of delivering software that are encountering bottlenecks from that process. So it's about understanding the kinds of code that is getting created and the level of attention that it needs and splitting things out into lanes.

17:50Otherwise, you're never going to get out from underneath all of that. You're never going to get the velocity. you're looking for. So definitely a really amazing interview. I highly recommend you all check that one out. That would have me nodding along the whole time. All right. Now I want to talk about something that is really important to me, and that is remote work and how big tech is falling out of love with it. So, you know, major tech companies like Meta, Amazon, Google, Microsoft, they're all shifting away from supporting remote and hybrid work to doing things like mandating like three days in office, for example, and noncompliance with these policies often leading to job loss or just reduced opportunities for growth, those types of things.

18:29Gallup did a study of over 400 ,000 US employees recently and found that about a quarter of them work fully remote and a little over half work hybrid, leaving only like less than 20 % that actually work fully on site, which is pretty astounding. And then in the background, there's also things like Microsoft's chief people officer stating in a memo in September of this year that But, you know, when people work together in person, more often they thrive, reflecting a broader industry trend, you know, sort of towards these policies. And, you know, I'm just going to come out and say I don't agree with that statement from this Microsoft.

19:06And I think return to office mandates really are just a stealth layoff strategy. You know, these companies are really just trying to make it so that enough people want to leave so that they don't have to go through additional rounds of layoffs after, you know, the COVID era hiring spree that all of these companies went on. And, you know, I once worked for a very large corporation that we experienced attrition levels so high at one point that there actually were rumors swirling that they canceled like a round or two of layoffs because of the attrition, which is I guess that's great for like people who just want to collect a paycheck.

19:41But like it was just frankly demoralizing for everyone else. So, yeah, Andrew, I kind of suspect you probably agree with me on a lot of this, too. So what's your take? Like, yeah, I mean, honestly, whenever an executive stands up and they say like, oh, we need to be in office collaborations better, we thrive, it's for our culture. Like that take is always feels false to me. It feels tired. I feel like that's just not the right way to look at it. And ultimately, you have to look at their incentives and why. It's like big tech wants people back in offices because they already spent all of this time and money and effort building and owning these massive campuses and building out huge real estate portfolios.

20:16portfolios. And that's ultimately a cost item that they're trying to justify an expense. And then it has the added benefit of being a helpful, like you said, stealth layoff tactic for managing that because all companies that have this problem have that problem too. So I think that they're kind of like a yin and a yang right now. It's kind of like a blender that we're all going through, but I never believe this when an executive says that as someone who's worked in in-person and remote collaborative environments, I don't think that spontaneous collaboration is what people are thinking about or doing more.

20:45I consider myself a very high output individual. My output is much, much higher when I work remotely simply because that spontaneous collaboration is actually protecting me. No longer can people pivot to me and need things to me spontaneously that break me out of my focus and pull me out of the things I do and ruin all of my output. So for people who say like, oh, it's for spontaneous collaboration, I always read that as, oh, like you need someone else to do your work. And so that's how I always interpret the words underneath these kinds of things. I thought this one was funny. I'm not going to be able to unsee that now whenever I see that phrase.

21:24But, you know, and I'm with you. Like, I personally worked at least partially remote for my entire career, like going on 18 years. In 10 of those years, I have been fully remote. And, you know, the quality of life improvements from that are just absolutely massive. And it would take huge incentives to convince me to go back to an office, particularly considering that I moved away from a big tech hub a few years ago to raise my children. And, you know, and if these companies were offering like big incentives to come back, you know, I wouldn't be so pessimistic, you know, if there was like childcare and shared transportation services and food services and health benefits and like all these things that they could provide to make the office experience more beneficial to the employees.

22:05But I think stuff like three days a week in the office just like really make no sense whatsoever to me, especially for these large companies. Like when I was working for the big tech companies, you know, everyone I worked with was distributed across multiple offices, sometimes a dozen or more, and all the meetings were still remote. So the spontaneous collaboration was usually just with the random friends I made at lunch. Yeah. And I don't think that, you know, requiring people to work from an office is universally a bad but for these large companies you know i don't think it's going to create really a noticeable productivity improvement all right we got one more story and that's the cloud fair radar for 2025 what is this andrew yeah this is the 2025 year in review from cloud flare basically the ruler of the internet as if you might know if anything goes down it's always because cloud flare just kidding cloud flare everything is built on it and so they have amazing insights into how the internet has grown and evolved in the last year.

23:01And one of the things that stood out is global internet traffic has grown by 19 % in 2025. That makes a lot of sense to me, right? It's like we have more AI agents, we have more crawlers, more bots on the internet. And among these threats as well, like security threats remain significant. 6.2 % of global traffic was mitigated by Cloudflare as if it was malicious. It was going to be harmful traffic that's nearly 10 % of all traffic in an environment where it was growing 20 % year over year and then 5.6 of analyzed emails that went through Cloudflare were identified as malicious which is up from last year which was at 3.9 percentage point so pretty pretty sharp increase on both like the actual surge of internet traffic but the dangers that come along with it so this is just a friendly reminder to you to definitely be always aware and vigilant and protecting yourself online.

23:52And I think that honestly, this number is going to go up even more. There was another one here that was fascinating about Snapchat overtaking X and user engagement, as in like users reengaging with the domain over time, which is a really interesting signal about shifting social dynamics. I never really thought that Snapchat would have gotten a surgence over X, but... Well, in our favorite social media platform, LinkedIn also now outranks X. So maybe it just speaks more to the decline of maybe just speaks more to x yeah in the fall of x yeah but i also learned about a new attack vector from this report and that's the harvest now decrypt later attack vector which i had never heard of oh yeah yeah but the idea is that there's malicious actors that might be collecting as much encrypted data about you as possible so that they can decrypt it later once quantum computing advances but you know one of the charts in here showed that now more than 50 % of web traffic is protected from this attack.

24:49So that's pretty cool to see. And yeah, there's just AI bot crawler traffic is just kind of all over the place. That's sort of the last big thing that I learned from this. A lot of swings, a lot of changes, things sort of like switch on a dime, it seems like. So yeah, it's pretty good report. There's just a ton of data. There's no way we could cover it all in here. So I definitely encourage all of our listeners to go check it So there's one more thing in there I want to call out that I thought was amazing. Is that Roblox user, like new users continuing to surge by 30 % in the last year, which is obscene when you already consider how large the user base of Roblox is.

25:29And Roblox is the standout for me. Roblox is the company to watch. It's actually, I think, it was like Zuckerberg's bet on the metaverse was a generation too soon. I think when the generation that grew up on is playing Roblox right now grows up, that's when the metaverse and the VR and augmented reality is all going to start to matter. But I thought that was fascinating to see it still growing so strong. So another one to keep an eye on. Yeah, the Roblox dominating the metaverse was fascinating. I never really expected that. But, you know, I just wish it could have been Minecraft. And now that we're at the end of our news segment, Ben, is it time to reveal our little surprise?