The Ruby AI Podcast
The Ruby AI Podcast explores the intersection of Ruby programming and artificial intelligence, featuring expert discussions, innovative projects, and practical insights. Join us as we interview industry leaders and developers to uncover how Ruby is shaping the future of AI.
The Ruby AI Podcast
NVIDIA's Rise, Rails Security Windows, and AI Model Reliability
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Rarely does a single episode cover so much ground so quickly, and this one delivers on nearly every front in Ruby and AI development news.
Kicking things off, Joe Leo reveals he now serves on the RubyConf planning committee, with the conference heading to Portland in May. That personal update gives way to some genuinely fascinating news: Evil Martians built an open-source benchmark called Lehman's to test AI coding agents on Rails, only to discover those agents were actively cheating. Printing API keys, raiding Git history, consulting other models with web access. It raises an uncomfortable question: if AI agents game benchmarks, how much should we trust them in production?
Naturally, the conversation covers a lot of territory. Rails pushing twenty-one security fixes sparks debate about whether the three-year support window fairly pressures teams to upgrade. Meanwhile, a Ruby 4.0 performance regression is slowing Rails route helpers by up to four times, pushing boot times from 15.5 milliseconds to 76.5 milliseconds (not a small hit for large apps). DHH's claim that programming languages barely matter anymore also gets a thorough and skeptical examination.
Tune into this episode for a lively, real-time reaction to the headlines shaping Ruby and AI development right now.
Hey everybody, welcome to another episode of the Ruby AI podcast. I am one of your hosts today, Valentino Stoll.
SPEAKER_01Yeah, hi, I'm the other guy. I'm Joe Leo. Valentino, I wanted to mention right at the top, it looks like you're growing out the beard in anticipation of winter. Is that what's coming?
SPEAKER_00Yeah, that and Santa.
SPEAKER_01Oh yeah. Yeah, yeah. He's also coming. No,
SPEAKER_00it's mostly winter.
SPEAKER_01We're going to at some point we're going to start putting these, I think at least these headlines between the commits episodes on YouTube. And I think you're really cutting an impressive figure today. When this hits the internet, people are going to say, Oh, this is a good looking software engineer over here.
SPEAKER_00I was hoping it would be more gray so that I could become a gray beard. I'm not ready yet. You're not ready
SPEAKER_01yet, but there is always time. There is always more gray. Don't worry about that. I'll tell you that from experience. I had the show on last night. My daughter was listening. She wanted to hear, she wanted to hear dad do a podcast.
SPEAKER_00That's funny. How did she like it?
SPEAKER_01She immediately started laughing. A few minutes into it. So Lucy is. I don't know. I mean, I was laughing with her. She's 10, going on 11. And so the next thing she started doing was mimicking my mannerisms. So whatever kind of verbal ticks I have, now it's all in my head because she's been parroting me and like making me realize the stuff that I
SPEAKER_00do
SPEAKER_01when uh when I'm on the show. So if you hear me talking very self-consciously, that's why.
SPEAKER_00That's really funny. Hello there.
SPEAKER_01Yeah.
SPEAKER_00Yeah. For those that uh maybe are joining new or aren't familiar with our new format, we occasionally do these news briefs where we run through a list of articles that get sent to us last minute, just freeform comments on what we think about it.
SPEAKER_01Yeah. So you're getting our raw reactions to today's news. Except for the very top story. Everybody knows the top story. So yeah, I'm gonna have an opinion on that. Uh that's probably not so fresh. But the rest of them, I have no idea. Your guess is as good as mine, what I'm about to say.
SPEAKER_00I forgot to start the timer, so we have four minutes to kill here.
SPEAKER_01Yeah,
SPEAKER_00what are we calling this again? Between the commits.
SPEAKER_01Between the commits.
SPEAKER_00I I really enjoy it.
SPEAKER_01Me too.
SPEAKER_00Maybe I can just skip ahead on this timer. Uh oh, look at this.
SPEAKER_01While you're doing that, I'll tell you that I am now in the planning committee for RubyConf. And so we are gonna have a RubyConf next year. We're excited about that. I think we talked about that with Marty a little bit. Coming your way in May in Portland. Hope to see all of you there.
SPEAKER_00Nice. Uh that's exciting.
SPEAKER_01I know.
SPEAKER_00I haven't been to Portland in in a year now.
SPEAKER_01Yeah, me too. It's been about 15, maybe more years. Wow.
SPEAKER_00But
SPEAKER_01I really enjoyed it.
SPEAKER_00It's gotta be a whole new place, right?
SPEAKER_01Yeah, yeah, I guess so.
SPEAKER_00Let's jump right into it. First up, you know, NVIDIA. I feel like NVIDIA has been on just like such a wild roller coaster.
SPEAKER_01That's for sure.
SPEAKER_00You think it's leading all of AI development one day, and then the next day you wonder why it even exists in some cases.
SPEAKER_01Yeah, and it almost didn't make it.
SPEAKER_00It almost didn't make it
SPEAKER_01to being all the way now to the most valuable company in the world by some measures.
SPEAKER_00The big headline is, of course, Jensen's exposure to Chinese secrets or exposure of secrets to China via a plant. What is your take on this?
SPEAKER_01The news item that I wasn't aware of here was that these three House Committee chairs have sent Jensen Huang a letter asking him to come in to a closed door briefing on November 18th, saying, Hey, we need to determine to what extent has U.S. national security been placed at risk because of your philandering. My first reaction was, we really are idiots. I sometimes don't know why we are trusted to run anything. It doesn't shock me at all. Big, powerful men taking a lover outside of the marriage is not so surprising, is it? And I guess you could do all that and not expose national secrets, not send chips to foreign adversaries. People do it all the time. But in this case, it seems pretty damning. Nvidia comes out with this flat denial saying, well, Jensen, of course, he meets with executives all the time, which I'm sure is true. And we're confident that if the US government believes that any person is a threat to American industry, it would let us know. Well, when the House of Representatives sends you a letter saying, hey, you got to come into this closed door meeting because we think you expose some secrets, that's America letting you know. That's exactly what's happening. What I'm curious about, I have some opinions on this too, but I'm gonna ask you do you think that Jensen Huang's job is safe?
SPEAKER_00As safe as it's always been. I think that he's been tried to get kicked out. This isn't the first time, you know.
SPEAKER_01The
SPEAKER_00same for Elon Musk. I am sure that the board of any company he's ever been on has tried to get him out any number of times, right?
SPEAKER_01Possibly.
SPEAKER_00I imagine that this is an ongoing battle. Historically, Jensen's pretty put together. He's responded to incidences like this before. It seems like without any effort. I think if you're willing to play along with the House and all of these committees, that is sufficient to protect whatever your corporate strategy might be. I just think about how frequently, you know, even Meta is in Capitol Hill and
SPEAKER_01being
SPEAKER_00interviewed and, you know, shamed, and he's still there.
SPEAKER_01Yeah. I mean, there are a lot of differences there. The first being that Meta is an exceptionally well-run company with or without its founder. Yeah, the fact is Jensen Huang's a celebrity. He's out there with his leather jackets doing his talks that have become very popular, millions of people watching him, and it's a big impact on the stock. Which, by the way, that's ultimately going to tell us, right? Stock drops too much for too long and Jensen will be gone. But I just don't think that'll happen because he is the brand. He's a big selling point for NVIDIA. Like it or not, I mean, I don't like it, but I don't think that's going to change here. I think even if they determined, hey, you know, you actually did give up a lot of secrets, hey, actually you've compromised national security. I don't think it's ultimately going to matter. I think he's still going to stay in that seat unless or until there's some damage to the stock.
SPEAKER_00To your point of meta being a well-run company, at what point does an NVIDIA just take advantage of all of Jensen's hard work and be a well-run company? Yeah, I guess I'm with you. I don't really know the answer to that. And maybe that means that it's not fully, you know.
SPEAKER_01This is Tuesday. News came out widely over the weekend. Stock dropped yesterday, but it didn't drop dramatically. I haven't checked today, but I bet it's gonna come right back. That's what I think.
SPEAKER_00Sell, sell, sell.
SPEAKER_01Uh let's move on to later. Maybe if we have time, I'll tell you about my friend who shorted NVIDIA for all of 2024 and lost his. Oh
SPEAKER_00no, oh no. All
SPEAKER_01right. Next item Rails shipped 21 security fixes in one release, and the 8.0 series just aged out of support. So there's a critical fix, arbitrary file, read-write access in active storage. There's high severity bundle, there's strong parameters bypass, there's a denial of service path, there's a bunch of stuff here. The post that came along with this for Rails said we strongly recommend upgrading as soon as possible. Well, yes. It also notes flatly that security support for the 8.0 series ends on November 7th, 2026. That's three days ago. What we want to talk about here, the discussion question with 8.0 support ending the same week as this release, is Rails security support window getting too short for teams that upgrade slowly, or is upgrade or accept the risk exactly the right amount of pressure? V, your thoughts.
SPEAKER_00Everybody should feel the pressure to upgrade, to be honest, now more than ever, until we find the right solution to this ongoing agentic security battle. And I feel like from this point on, there will probably be a daily patch that you're gonna need to upgrade in some form or another. And I would expect that going forward until you get your loop in place, to be honest. Is this a short window? Yes, especially if you're a bigger company, but it's only gonna get tighter.
SPEAKER_01Right. You
SPEAKER_00hope. If it doesn't, then it means that there's a lot of stuff falling through the cracks.
SPEAKER_01I think that this question, I mean, AI generated this question, so there's nobody to blame here. But I think the question is slightly ingenuous because upgrade or accept the risk is not really the right framing. I mean, you're accepting the risk whether there's an end of life or not, right? This is Rails just saying, hey, you're on your own at this point. We've given you the tools to upgrade and you haven't done it. I agree with you that business practices have to evolve here. There are too many threats that are being discovered in real time that are serious threats. I mean, the kinds of things we're talking about here, you know, possible denial of service paths, arbitrary file read-write access, these are serious.
SPEAKER_00Yeah,
SPEAKER_01you know, this is not just a small thing. Active storage is not a small thing, everybody's using it. The proliferation of these kinds of security findings and the ability for anybody who wants to be a bad actor to attack these surfaces means that our processes need to evolve. Like we need to be upgrading. You're right, probably every day. Now, you mentioned something interesting about doing it every day until your loop is in place. What is the loop that is in place for you?
SPEAKER_00Subscribing to many different security channels as they affect your software. So any number of varying kinds of software have their own security channels and they publish a feed for you to listen to updates to. That would be number one. Yeah. And I hope that you're already doing that. But the loop aspect of that is then reacting to that and then having something that proactively applies the fixes that are fixes to the things that are announced and then merges it so that you close the gap from the time that it is announced publicly to the time that it's in your code base. And a lot of these security channels, too, are not public, where you have to like subscribe in some fashion via email or or account if you are like a sponsor to Rails as an example. Where you can get some closed channel security updates on top of the public ones to kind of get advanced notice. I recommend first finding that and then finding how you can proactively react to those fastest using agents.
SPEAKER_01Having a dedicated agent that scans the messages as they come in, proactively go out, submit the patch in the form of a pull request, and then make it easy. This is the other piece, make it easy to bypass scrutiny and get into your code base, which I think is also probably a challenge for people as we know from talking to people that PRs are just piling up. And this one can't just sit on the bottom of a pile somewhere.
SPEAKER_00You have to uh weigh the risk, right? Is it okay to assume liability for that risk staying open for a certain amount of time? And what is that what is that time? Is it worth merging something maybe that you need to figure out how to resolve later?
SPEAKER_01Yeah.
SPEAKER_00Or is it something that you can fix retroactively after the security aspect is resolved?
SPEAKER_01Yeah, I agree. And I I also think to Rails, the Rails core team's credit, they're saying you can't wait. You know, this really has to get in there.
SPEAKER_00Yeah, and we're gonna only find that more and more. So finding that right loop where you can automatically merge and approve security things sooner than later, I think is something that every team should be focused on.
SPEAKER_01I
SPEAKER_00was hoping GitHub would have an answer to this. I haven't seen one, but maybe they have it and you just have to pay for it.
SPEAKER_01I haven't seen it either. I think they're still trying to just keep their head above water right now with the increased amount of attention that the application is getting.
SPEAKER_00I know I just ran out of my free tier credits from next up we got Ruby 4.0 performance regression makes every Rails app route helper up to four times slower to load. This is a little bit unfortunate.
SPEAKER_01Yeah.
SPEAKER_00A bug report was filed end of October here that found that defining a method inside a module gets progressively more expensive in Ruby 4 for every single class that includes that module. So a pattern that's everywhere in Rails is like this, since every named route adds methods to helper modules, included in every controller view, mailer, etc. This particular benchmark found that defining 2,000 methods with 500 including classes, it took about 15 and a half milliseconds on Ruby 3.4, on Ruby 4.09 got about 76.5 milliseconds. So that's like a
SPEAKER_01That's pretty slow.
SPEAKER_00That's pretty slow. I guess the question here is should a regression like this and something every Rails app hits on, every boot get backported on principle, even though Ruby core doesn't normally backport these performance fixes?
SPEAKER_01Yeah, the question should the a regression like this get backported? Yeah, I do think it should get backported on principle. That's really slow. Look, a performance issue that only hits every class that includes a module, that's like saying, well, it's only it applies to every sentence in English that ends in a period. That's a lot of code. A lot of classes are including modules. So it's everywhere. This regression is really significant. And I would say, you know, 76.5 milliseconds is an intolerable amount of time to wait for anything with a route helper. So I do think they're doing the right thing here and saying, let's get this backported. I don't know why there's a policy against backporting performance fixes. I guess the idea is that you want people to upgrade in order to continue getting all of the good performance improvements. But I think it's fair to treat this as a bug because for at least large applications, it's slow to the point of being unusable.
SPEAKER_00Yeah, all the more reason to use the next version of Ruby.
SPEAKER_01Well, yeah, that's the counterargument. You're right.
SPEAKER_00You know, we take for granted all of the improvements that are made for performance when you upgrade versions of Ruby, but to get it where it also just behind the scenes makes it slower, it's not something that you want seen advertised for sure. I wonder if it bounces out, if this is just a net positive over time.
SPEAKER_01Yeah, so that's the thing. It would be one thing if you found something that sped up the application quite a bit, and you said, okay, we're gonna put this in Ruby 4.1 or 4.2. And hey, if you want that, upgrade. But this, you know, in Ruby 3.4, you had 19 milliseconds, and then you upgrade to Ruby 4 and you have 76 milliseconds, that's really slow. So I do think they're doing the right thing. I understand what you're saying. You know, upgrade if you want better performance by that same token, you might say, Well, let's just put performance regressions in all of our outdated versions.
SPEAKER_00That
SPEAKER_01will really for let's
SPEAKER_00how do we encourage people to upgrade? Uh yeah, yeah, exactly. So we're gonna
SPEAKER_01start by taking a page out of the Rails book.
SPEAKER_00I'm suspicious Apple does that.
SPEAKER_01Right. You think so?
SPEAKER_00Oh, yeah. My phones always get slow when the new model gets released.
SPEAKER_01It's a fair point. Yeah, slow and the battery starts dying in a couple of hours. The two of these news items combined with Rails saying, Hey, we're just not gonna support Rails 8.0 anymore. There's too many security issues, you gotta upgrade. Ruby actually taking a slightly different tack and saying, Well, okay, we kind of messed this one up in Ruby 4, or at least in Ruby 4.09. And so we're gonna backport it. We're gonna make sure that people that are using it are okay. And it's because there are two different issues here. Performance is something that can be improved in place, but security is something where you really got to get people on newest versions, or you're putting your own community at risk.
SPEAKER_00I would prefer slower code to insecure code.
SPEAKER_01Yes, I would too. Okay, the team that benchmarks AI coding agents, that's evil Martians, they benchmark AI coding agents on Rails. They had to build their own anti-cheating jail to keep the results honest. So let's see what this story is about. Evil Martians published the second entry in their Agents on Rails series on November 3rd. That's one week ago. The story of building Lehman's the open source Ruby harness that actually runs every trial of the Rails Foundation's coding agent benchmark. So, in their words, every model must get the same conditions and no agent should be able to escape its sandbox or fool its grader. But getting there took several rounds of hardening, mostly in response to agents actually trying to cheat. The incidents that forced each new guardrail, one model printed out environment variables, including a live API key. One found compiled bytecode sitting in a build cache and literally announced jackpot. One searched its own unreachable git history looking for an earlier attempt at the same task, and one used its own API access to ask a different model with web search for the actual upstream source code it was supposed to be implementing rather than doing it on its own. So the question here if a benchmark needs a literal sandbox jail to stop models from cheating their way to a good score, what does that say about trusting those same models output in a real unsandboxed code base?
SPEAKER_00I feel like there's no good answer here.
SPEAKER_01Let our listeners be the judge of that.
SPEAKER_00This makes me think of benchmarks in general. Like randomness matters, even just setting the right seed upfront because you're doing so many iterations. How do you calculate that each thing is truly in isolation and not just like performing better from like an internal system cache perspective of how you're running it? I imagine they improved it.
SPEAKER_01Yeah.
SPEAKER_00From my experience running and thinking about all these eval benchmarks, there's always a way to game them, no matter what.
SPEAKER_01Right.
SPEAKER_00You're never going to be able to stop that. Hugging Face accepted that. And I think that they stopped running leaderboards for a while because of that.
SPEAKER_01Oh, that's an interesting point.
SPEAKER_00The evals are important, but trying to get them right is almost impossible.
SPEAKER_01Well, let me see if I can take a similarly impossible stand here because what I want to say is it's not cheating, it's benchmarking. The benchmarks are always gameable. Although what I'm trying to say is really, really contradicted by an AI agent yelling jackpot when it finds compiled bytecodes sitting in a build cache. What I think is that one has really nothing to do with the other. Yeah, evil Martians, if they're going to continue to try to steward this initiative, and I think it's a good one, benchmarking the AI coding agents on Rails is one thing. There's the results themselves, but to me, the bigger thing is that you're putting the industry on notice that we're Ruby on Rails developers, we're watching. We want to see you cater to this community. So I think it's important work that they're doing. I don't think that these little AI agents are all running around trying to, you know, trying to copy each other's work and steal bytecode from one another in any way reflects on their ability to do agentic development for your software application. And I think even more to the point, it doesn't really matter. What does it say about trusting those models output in a real unsandbox code base? Look, we are trusting it. That is what's happening. So I don't really know if that question is going to move the needle in any meaningful way. It's certainly very interesting to discuss and to see how these agents work in this approximation of the real world, but it's not the real world. In the real world, we're all trusting these agents all the time, like it or not.
SPEAKER_00I hope that they're focusing on the right ones. I haven't looked at this in a while.
SPEAKER_01Yeah. But
SPEAKER_00yeah, some are just not worth pursuing.
SPEAKER_01Some models, you mean?
SPEAKER_00Some models, but also some tasks.
SPEAKER_01Okay. Yeah.
SPEAKER_00Especially coupled to that. I imagine like running a security type eval check on haiku. It's not like a great use of your time or tokens.
SPEAKER_01Right. Yeah.
SPEAKER_00You kind of already know what that outcome will be. It does seem like they're going in the right direction of at least labeling it in the right ways and applying the grades to where they need to be. Yeah, I'm looking forward to seeing more.
SPEAKER_01Yeah. Me too.
SPEAKER_00I don't think any of this jailing is gonna be an issue.
SPEAKER_01It's not in the real world. Like I get why you need to do it. You're trying to make your benchmarks honest, but I don't think it really applies to real-world development.
SPEAKER_00Yeah, like we've been talking, not everybody has all of their agents jailed anyway.
SPEAKER_01That's kind of my point. These things don't really get anybody to do anything more careful.
SPEAKER_00Maybe though.
SPEAKER_01We're really accepting a lot of risk.
SPEAKER_00Maybe having them jailed and showing how much that they fix and protect you could help encourage people to get them jailed.
SPEAKER_01Oh, that's a fair point. Maybe. I don't actually think so. Maybe I'm just a little bit pessimistic today. What's next?
SPEAKER_00A working developer's take here. Reviewing AI generated code might be pure overhead with no actual payoff. This is a very controversial stance here. But David Bryant Copeland published an essay on November 3rd last week, arguing that in a professional setting, co-review of AI generated output carries real cost and delivers little benefit, and walked through each of his reviews' three traditional purposes to make the case. One on mentoring, he says that it doesn't work since the A doesn't learn from review comments the way a junior developer would. Another on approval, saying that it doesn't really happen either, since compliance reviewers mostly check the change description and tests, not the line diffs, often leaving bug finding as the one purpose left. And his claim here is that in quotes, it is unlikely that an experienced developer would find many bugs in AI written code through review alone. And his sharper concern being a long-term one. And again, in quotes, this is short-sighted thinking, especially as developers lose the skill of code review entirely. And his alternative presented here isn't just stop checking the code, it's moving the checking earlier and invest in custom static analysis and better prompts instead of after-the-fact review. One real gap here worth raising is that the essay barely mentions tests at all and where they fit in. Question here being is code review of AI Apple genuinely theater, or does skipping it just mean the bug surface later in production instead of a PR?
SPEAKER_01I don't know whether or not it's theater. So I haven't read this article. We know David Copeland. I have not read this particular piece, but I will after the show. This is the part that moved me. It's short-sighted to continue doing AI code review, especially as developers lose the skill of code review entirely. That is direct. It's a little bit bleak, and I think it's a hundred percent accurate. I just think that right now, in a lot of ways, we're trying to hang on to things that used to make us feel like software engineers. When the truth is, the definition of what a software engineer is evolving and has been since these code gen tools first started becoming usable. And I don't know about the custom static analysis, better prompts. I don't know about the solutions or the alternatives that he's positing. I'd have to read the article, but I can say with some confidence, yeah, we're trying to roll the boulder uphill right now with AI code review. And I think that it is uh not long for the world.
SPEAKER_00Yeah, there's a few things in here too, I don't necessarily agree with in practice, at least not from what I've seen. Especially the argument among the AI not learning from review comments. I mean, there's a way to implement that.
SPEAKER_01Yeah, you're right.
SPEAKER_00And get those learnings captured and have them applied to every future review.
SPEAKER_01Right.
SPEAKER_00As well as these AI reviews are only getting better over time. From what I've seen, catch everything.
SPEAKER_01Yeah.
SPEAKER_00The stuff that they don't catch that you would want catched uh is. Really, just bike shedding at this point for the most part. And the cases where you might think that you'd catch something from a review, often there are times where you would miss something.
SPEAKER_01Sure.
SPEAKER_00And I think that we're kind of hitting the balance of that. I wonder more what do you think about what that means once humans aren't reviewing it anymore?
SPEAKER_01I think what you just talked about with AI learning from review comments, I think, yeah, so you make AI do good reviews and then leave comments for the AI. I still think that there's a lot of important work to be done for software engineers. I just don't think it is in reading line by line the code that's submitted by AI agents. I just think that that is not long for the world. That's my personal opinion.
SPEAKER_00Either way, I think we could both agree. There's a real gap for sure.
SPEAKER_01This is interesting. I want to get more into this essay and uh learn a little bit more about Copeland's take here. DHH on why he thinks the language you write in stopped mattering, in his own words. Ugh, oh God, we're still talking about this. DHH gave a talk in Rails World in September, and we're still paying for it here in November. He has this quotable line. He was published an email interview with DHH in FOSS. Quote is they clearly don't matter in the same way they did when we were writing everything by hand, unquote, on whether high-level languages like Ruby still matter now that AI writes so much of the code. V, let's skip through the rest of this. Let me just ask you does it matter to you that the guy who wrote Rails and is, you know, still nominally the head of Ruby on Rails doesn't really care what language you use to write code anymore.
SPEAKER_00I think everybody can agree he's no longer writing Rails.
SPEAKER_01Yeah.
SPEAKER_00Or Ruby. So does that matter? I don't know that it does for me personally.
SPEAKER_01Yeah.
SPEAKER_00I'm happy to use Amachi every once in a while on an old machine. If he wants to continue pursuing that, great. Look forward to the improvements there. But I certainly don't look to him for improvements on Rails
SPEAKER_01or
SPEAKER_00Ruby for that matter anymore. So I don't know.
SPEAKER_01This is funny because I think this goes back to our Jensen Huang argument before. You know, he's the face of the franchise, which means that what he says matters to some degree. But I think it's interesting that your take is really the only take that I've heard from many other Rubyists where it's like, eh, DHH can do what he wants. I'm still gonna write Ruby. He doesn't have the same weight that he used to in the Ruby on Rails community.
SPEAKER_00You're not gonna use Ruby for everything. You know, some tasks are better purpose in very specific languages.
SPEAKER_01Yeah, of course.
SPEAKER_00And that's fine. Ruby itself is a language, it empowers very specific shapes and flows of code that you almost can't get in other languages. I prefer it for that reason, especially like the meta programming aspect of things.
SPEAKER_01Yeah.
SPEAKER_00The performance gains are also answerable. So which do you prefer the shapes of and the flow of better? I feel like it comes down to taste again.
SPEAKER_01When you say the performance gains are answerable, do you mean that in response to you know rewriting hay in another language? You
SPEAKER_00gotta take the trade-offs. A framework brings a lot of things just for free.
SPEAKER_01Sure.
SPEAKER_00And that's why you use it. There aren't frameworks like Rails anywhere near it. Until those things are made, and maybe they are getting made. I don't know how I feel about the quality of those things, right?
SPEAKER_01Yeah.
SPEAKER_00And long living of them. I've seen and witnessed Rails perform very well for large organizations, enable them to move quickly and adapt to change. Yeah,
SPEAKER_01you and me both.
SPEAKER_00Rails gets upgraded behind the scenes. You get those community contributions for free as you upgrade and move with the community. I feel like a lot of that isn't being talked about and just being talked about performance gains.
SPEAKER_01Right.
SPEAKER_00Which is just one tiny little part of building an organization around it.
SPEAKER_01I think you're right. Scalability is a big one. We've seen Ruby on Rails scale along with organizations. We know how that works and that it supports not just the application, but the organization itself as things get bigger and bigger. I think that that is one really compelling aspect. People have been saying this a lot. Like I'm just thinking back to the Rails world. You know, convention over configuration is actually a big deal. Rails as omicase is actually a big deal. Putting some constraints around your development actions and your development team for that matter helps you to be creative within the boundaries that you want your team to be creative. So I think there's really a lot to be said for Rails, leaving aside its performance, which, you know, by the way, is not so bad these days. It's pretty good.
SPEAKER_00Also takes away from the one person founder idea that Rails was built for. Somebody like Anthropic can be like, oh, let's rewrite our whole messaging protocol in Rust. It's like huge success of a rewrite. And a lot of people not realize that there's a whole team of people supporting that infrastructure change.
SPEAKER_01Yeah.
SPEAKER_00Other experts that are making sure that continues to grow and optimize in a uh fashion aligned with the rest of the organization.
SPEAKER_01Yeah.
SPEAKER_00I feel like it is not a job for one person.
SPEAKER_01Yeah, for sure.
SPEAKER_00And as soon as you start expanding that view, I feel like you get a smaller and smaller capability in your hands when you're not using a framework.
SPEAKER_01Yeah.
SPEAKER_00Long live Rails.
SPEAKER_01All right.
SPEAKER_00Next up, we got Instagram's new Mac app lets an AI agent assemble your rough cut from a folder of raw footage. So Instagram launched its edits for Mac on November 5th. Free multi-track, automatic captions, color grading, watermarks, a whole litany of features here. The headline feature isn't the editor itself, it's the agent integration. Matthew Samari told the Verge uh an AI agent can use the editor's own tools to assemble a rough cut. You hand it a folder of footage and a brief, and it takes a first pass, then you open the timeline and finish the edit yourself. The real limitations here are you can't yet edit the same project across your phone and Mac, and the music rights still need checking before reused elsewhere. The question we're presented with here is is agents assemble your rough cut and you finish it the right division of labor for these creative tools? The same shape as how a lot of us already work with these coding agents, or does creative work lose something when the first pass isn't yours?
SPEAKER_01Let me answer your question with another question. What even is Instagram? I don't have Instagram, I've never had it. Uh, I think Def Method might have an Instagram account that's been dormant for like eight years. I think the question here is does creative work lose something when the first pass isn't yours? Yeah, it kind of does in writing. I know that. I don't know about for video because I don't know what the goal is. If it puts together a rough cut of the videos you took while you're on vacation, does it really matter? I mean, it's kind of nice to have a tool that stitches something together, I guess, if that's what your goal was. And if you're trying to do something truly creative or monetize it in some way or make a film or something like that, then yeah, the creative work will probably suffer. But then if it does, then you'll probably not choose that path. You'll probably do something different. I'm like, whatever. Do whatever you want. What's your take?
SPEAKER_00I'm a little worried.
SPEAKER_01Yeah, are you? Okay.
SPEAKER_00Maybe I don't do scroll as much as I used to, but it does just become a watch, especially as more AI planned video content comes out, it starts to become very similar. Makes me not want to watch it as much, I feel.
SPEAKER_01Yeah.
SPEAKER_00So maybe it is a good thing.
SPEAKER_01Because then you won't watch as much. You
SPEAKER_00know, it's back to like, are we handwriting code or is somebody like gracefully designing it behind the scenes? And I hope somebody's designing it. And for the same case here, I hope somebody is putting some design effort in the video flow. Whether or not they do it up front or at the end, I don't care.
SPEAKER_01I understand what you're saying, Valentino, but I also think that your quality tends to rise to the top. So you're right, this thing probably will create many more videos that are not particularly fun to watch and are not particularly creative or interesting. But that doesn't, I don't think, in any way diminish the standout work that will be done with or without AI's help by truly creative minds. And I think that there's a subtle difference here between agentic development, where yes, people can be creative, but we are trying to solve a business problem most of the time. And something that is truly creative, that is not always trying to solve a business problem, but is making art for art's sake. And I think I will more comfortably rest on my ability to discern what is truly good and truly suits my taste from the slop. And I don't think making it easier to make slop will negatively impact the quality or the quantity of the truly good stuff that I say.
SPEAKER_00You know, we had Rick Van Michelin on from OMA. We got into what makes art art.
SPEAKER_01Yeah.
SPEAKER_00And how is AI influencing all of this? I feel like it's very related here. The takeaway was artists use AI and it doesn't necessarily take away from the art because it takes the artist to get the output that it delivers. So what shape that takes, I feel like it's MySpace, right?
SPEAKER_01Yeah.
SPEAKER_00Get a bunch of videos that become a MySpace content generator. We'll just see all kinds of fun flavors in higher def.
SPEAKER_01Right. Okay. That's the end of the list. That's our show for that. That's it. Thank you everybody for listening. I think there were some really interesting stories that we covered today. I'd like to hear other people's takes on it. So get in touch. We are available. Write to us at news at therubyaipodcast.com or message us on LinkedIn. That's where we post a lot of clips of the show. Or track either one of us down individually and tell us what you really think. We'll be happy to get your feedback.
SPEAKER_00Hey, one other thing. Uh we made a thing. We'll be announcing it soon. There will be a way for you to connect to the Ruby AI Podcast MCP server and have it influence your agent sessions. I have personally gone through the pain-sinking process of tearing apart our feed and transcribing everything and splitting it up so that you can get cited sources. Uh
SPEAKER_01all right.
SPEAKER_00Dropping into your sessions as you code away. I'm excited. I've been messing around with it and it's been producing some really fun results. Tune in if you want to drop us a note and get a free gift link. Let us know.
SPEAKER_01Yeah, I'm excited about this too. And join us next week where we'll be speaking with Andy Kroll and probably the economics of putting on a conference, among other things. Among other scary, scary topics here in 2026. Thanks everybody for listening.
People on this episode
Podcasts we love
Check out these other fine podcasts recommended by us, not an algorithm.