Why engineering leadership matters more than ever | Manoj Mohan

0:05Welcome to Dev Interrupted. I'm your host, Andrew Ziegler. And I'm your host, Ben Lloyd Pearson. This week, I'm sitting down with the seasoned executive, Manoj Mahan, to discuss why AI should be viewed as an intern rather than a replacement for engineers. He shares his optimistic view of this new industrial revolution and breaks down his framework for deploying responsible AI at scale. First, we have some great news stories to cover today, starting with federal AI regulations, AI agents at the Linux Foundation, why scale is all you need is dead, and why whimsical tool names confuse everyone. Andrew, let's just start at the top with these new AI regulations.

0:44Yeah, so the United States president signed an executive order establishing a single national regulatory framework for artificial intelligence. And the goal of this is to override state-level regulations that can cause patchwork interruptions for companies that are expanding their AI domain into those areas. We're talking about creating an AI litigation task force to challenge state AI laws that go in against this executive order and doing so by withholding billions of dollars in broadband program funding. Of course, this type of broad action would require legislative action, but this is the work of many large tech companies as well to ensure that their road to an AI rollout is smooth and paved and not going to be stopped at every state along the way in terms of legislations and roll out confusion.

1:32So this is a really interesting order from the White House in regards to how the power makers that may be are shaping AI legislation within our country and enabling all of us to work with this tool. But Ben, what do you think about this type of broad sweeping executive order? Yeah, well, I've been expecting something like this to come along for a while now. We're kind of in a Wild West phase with AI right now. That reminds me a lot of the early days of the web. And over time, the web centralized around or consolidated around centralized providers and had more regulatory frameworks that were implemented on it.

2:13So I think it's totally reasonable to expect that AI will go through a similar process. And I also think that it makes a lot of sense to regulate this at the national level rather than having like a patchwork of state laws. You know, if you think about it, AI is trained on data that is created by people all over the world. The engineers who build AI tools live all over the world. The data centers that train the models and provide them as a service are also located all over the world. So when you're in this world where all the aspects of what goes into providing AI as a product or a platform is distributed across many geographies, it makes perfect sense that the federal government would regulate it under something like the Interstate Commerce Clause.

2:54But with that said, it seems like so far the approach of the executive branch of the US federal government is to lessen restrictions on AI rather than actually creating some sort of regulatory framework. I'm kind of personally not really a big fan of this, like calling the global competition some sort of race to something, because a race implies that there's some sort of end state that we're trying to reach for, like getting to the moon first, for example. And I don't really think that's like how AI is developing. You know, it's more like an ongoing progression that continuously improves at an incremental rate and just gets better and better and more robust.

3:31And, you know, my opinion is that we should really be more focused on raising the tide so that all boats are lifted. You know, that would be things like, you know, publicly available training sets. Like what if everyone in the world could build powerful AI models rather than having being forced to have centralized providers because they're the only ones that have access to the data that makes it possible. And I know we have lots of non-American listeners out there, but I think it's really important to cover because a lot of the legal frameworks that get established for things like technology in the US often get adopted in parts of the other world too.

4:06And it's kind of challenging to cover American politics right now because things change so dramatically week to week. But this is certainly something that we'll be keeping our eyes on as it develops because AI regulation will impact everyone that uses it, including software engineering teams. Indeed. All right. So let's talk about what's happening over the Linux Foundation, Andrew. Yeah. So the Linux Foundation, they've launched the Agentic AI Foundation or the AIF, was founding project contributions from Anthropic, Block, and OpenAI. And these are foundational AI projects, things like Model Context Protocol, Goose and Agents.md that are establishing the basis and the open source frameworks by which many developers and tools are adopting and advancing with these types of agentic coding frameworks and things that are available to us.

4:56And this foundation is also backed by Amazon, Google, Microsoft, Cloudflare, Bloomberg, you name it, a big player in the AI space. They probably have something to do with the new AAIF because this is all hands on deck to stabilize and secure and lay out a bright future for the frameworks and the protocols that will power AI and make it capable of scaling. This is the same exact type of coming together that we see earlier in like the Web 2.0 era and the cloud era of coming together to provide best-in-class standards that are open source for teams to build and roll out their computing systems on.

5:36The same thing is now happening with our AI tools. This is a pretty interesting announcement, especially, you know, I've worked a lot with all of these tools listed here. So to see them all come together into one cohort makes me really excited about what's next for them. Yeah, yeah. I love the Linux Foundation. Really cool to see a lot of companies that are, they kind of feel like a bunch of friends of the show, right, Andrew? Like, we're very familiar with a lot of these companies. But there's a lot of really cool projects that are being added to the Linux Foundation as a part of this. Like, Andrew, I know you're a huge fan of Goose.

6:07Biggest fan of Goose. You're right that these are all friends of the show. And so it makes me happy to see this kind of success for them. Yeah, yeah. I also really like what I see with agents.md, which is a project from OpenAI to provide guidance to AI agents on how to interact with repositories. And this one stood out to me because, you know, my personal approach to building AI workflows has been less focused on the prompts that I give to my models. You know, we've all heard this phrase prompt engineering. I actually think that's the wrong focus area. like context is the most important thing.

6:39So I think in a, yeah, I think in a, in a perfect world, your prompts are actually very simple. Like even as simple as like just a few sentences, but they should enable your agents to access all of the appropriate context. They need to make the right decisions. So, you know, the, the organizations that are being successful right now, like they're building a corpus that provides that context to their agents. And I think everyone should be focused on this. So it's, it's really cool to see the Linux foundation playing a role because this is how our agents are going to get more capable of responding to a wide range of situations.

7:12I'm definitely going to be keeping an eye on what's happening with this. And, you know, Andrew, who knows? Maybe we'll even try to attend the MCP Dev Summit next April. You know, that sounds like a pretty cool event. But also, yes, I definitely want to keep bumping into all things MCP and Goose and AgentsMD, of course, out in the wilds. I think you're spot on about the importance of context engineering and really what you have to come to bear with every AI project that you work on and bringing in your best practices. And that's what these tools, these frameworks are all about, establishing a really repeatable, successful framework for you to get the most out of your tools.

7:47So really exciting news, excited to follow this one further. And jumping into our next article here, we're talking about scale is all you need is dead. Yes, the movement behind this idea that you can just keep going and going and going on model size, compute and training data to just get bigger, better, smarter models. Well, this theory is really starting to not hold water for many leading AI scientists. In fact, at the recent Neuro-IPS 2025 conference, which is the largest AI conference in the world, many scientists are publicly acknowledging that scaling large language models is insufficient for achieving what we call generalized intelligence.

8:29And that challenges the previous dominant scale is all you need approach. And so you have industry studies from research at MIT, McKinsey, and BCG, also reporting that, you know, over 90 % of companies aren't realizing significant ROI improvements from their investments in generative AI. And maybe that's just yet, but maybe that will be forever. These are all questions that hang over everyone right now, I think, in using and adopting these tools, especially as consumptions and costs go up. And this really dramatically changes the drumbeat by which AI companies that are scaling, especially foundational models, can argue that, you know, getting more resources, more time, more compute, more power, more training data is ultimately going to yield a bigger, better product.

9:13In fact, many have called out that a lot of the work now is upon us and all of the stages in between. I would fine tuning is securing reducing the size and costs of the models. There's a lot of work to be done. So, Ben, what do you think about this article from Gary Marcus? Yeah, well, first of all, Mr. Marcus, you know, he lamented how difficult it can be being a contrarian on issues about this, you know, and I get it. Like, you see. But I'm sure you could relate with that, Ben, you know. Yeah, what's that saying? No, actually, I am. I am happy to call myself a contrarian. But, you know, the issue is you see all these glaring issues with society around you.

9:52And most people don't agree with you. And then by the time everyone catches up to your perspective, you've already moved on to the next challenge that you're contrary about. But what I really like, you referenced it, but I really like the part that references how many or why many AI initiatives fail. And I think it really boils down to this. Like if your AI initiative is focused on AI adoption rather than solving some sort of discrete problem, it's far less likely to be successful. In fact, the overwhelming majority of those types of initiatives fail. And it really, this article really reiterates a theme we've been hitting on constantly here at Dev Interrupted.

10:29You know, I'm looking forward into 2026. And I think that's the year that this incremental improvement from the frontier models begins to exponentially shrink. And at the end of the day, like when you really think about what an LLM does, they're really just like a facsimile of certain aspects of human intelligence. Specifically, our ability to combine words into a coherent whole. I believe that they're incredibly powerful, but I think we're officially reaching the state where most of the gains at this point are going to come from the application of existing technology rather than further incremental improvements.

11:04I think this race to have the biggest, most powerful model, like that's where the real bubble is happening right now. You know, and I and personally, like just as I've invested more time and effort into using and learning about LLMs and GPTs, I have found myself believing less and less that AGI will result from this technology. Like it really does feel like the gap gets wider the more you understand it. So great article. I really enjoyed reading this one. Yeah, and I like how you called out that the, you know, in 2026 being the year that the improvements for the models, they make them shrink, they make them smaller, more efficient.

11:40I think it even says something that we talked about this article instead of GPT 5.2. You know, it speaks to what we're all focused on right now. It's not yet another foundational model rollout and the transformative numbers on the benchmarks. We're getting to really, you know, a point where we're barely eking out changes on that type of leaderboard. and there's bigger things to focus on. It's a really great summary. Interested to see where this one goes. But I want to wrap things up jumping into our last article of the day. This is a really fun one about, you know, developers, software engineers losing the plot when it comes to naming their tools.

12:17And I read this article and instantly related with it in all of the good and bad ways. I saw myself in this article and I did not like it. But then I emerged from the other side being like there is a better way. So this article, it highlights a trend, you know, if you're a dev, you might be familiar with having very wacky or obscure tool names, library names that become kind of abstract and unrelated to like what you're coding. And the author of this article, he cites examples like Viper for configuration management, Cobra for CLI, or Melody for WebSocket connections. Like maybe the person who named it that, they had an idea or a clear analogy, but it doesn't translate when others pick it up and read it.

12:59And noting that these names that they have, they don't have a descriptive connection to their actual purpose. And this actually introduces a lot of cognitive fatigue to developers because now, you know, you're going through your project dependencies and you just have this big long list of random nouns and you're like, what is CASBEN or async with a K? Like you don't understand what they relate to. And this cognitive cost, you know, it can definitely cause issues with actually sharing and using a project, making sure you're using the right dependencies. I loved that ultimately this really called out that the naming schemes, they have to be a little more formulaic.

13:41You need to have a system. And this is something that I've kind of evolved with as well, because before, like my own project naming system, I would want to find that clever name. If you think I'm lying, go look at my GitHub profile and you will laugh at some of the repo names that I do have on there. And sometimes that paralysis would reach a point where like, I wouldn't even start writing code or start the Git repo until I had come up with that clever name. There's like a whole new level of paralysis, I think. So, Ben, what did you think about this article and all of the wacky names that we're giving our Git projects these days?

14:17Well, I have to think of my own personal example of wacky project names where a developer came up with a just a crazy name for an open source project that ended up getting our company associated with Ritual Sacrifice. by all the crazy internet conspiracy theories. So I definitely resonate with poor naming decisions being a problem from time to time. But I feel like this article is really particularly relevant in the AI era, even though it's nothing to do with AI. But, you know, think about it. Like we're all out here vibe coding now. Sometimes I don't really even know if a project that I've started with AI is ever going to have legs.

14:55So naming it is kind of like a wasted effort. Like I don't want to give up a good name on something that may not be used for very long. You know, I think sometimes it's like best to just keep the name like extremely simple and obvious to what you're trying to do with it. Oh my gosh, you went like the total opposite direction. You went like with like your new coding projects. It's like you're a pioneer back in the days. It's like my baby has to make it through the winter. It has to make it to its first birthday before we give it my grandfather's name. Like I love that that's like where you're at with these projects because that's how quickly they come up.

15:29and how you use them. But, you know, I think that there's like so many ways you're going to take it and apply it. If you get a systematic way, it actually just totally dissolves as a problem and it actually starts helping you. So now I'm like, I like the name stuff in a domain namespace kind of way. So like, let's say I have a bunch of projects for like agents, right? You might have your marketing agent, your copy agent. And instead of naming them that way, I would be like agent hyphen marketing, agent hyphen copy. That way, all of the agents live next to each other in the folder. And this is nice for me for scanning and scrolling.

16:02Makes collecting up my projects really easy. Makes naming them even easier. But the extra unlock here is that now they're nice and semantically organized for an LLM to understand too. So now when an LLM can pull up my list of projects, it knows all of the similar namespace things. Things that are working on the same kind of project or maybe employ the same type of framework. They live side by side in the folder. That actually is a huge unlock. So definitely always be thinking about not only is this confusing a developer, but is this confusing my LLM? Yeah. And, you know, and this reminds me of an episode we had quite a while back, actually, with Luis Alejandro Vega from Bloomberg LP about creating a culture of sharing internal developer tools.

16:45So as a part of their workflows, they help developers name and brand projects that they want to create and share with the rest of the company. And they come up with some really clever names. Like, I think my favorite one that I remember, they called it Logger, which does exactly what it sounds like. It pushes stuff to logs. And they had this fun little lumberjack mascot for it. So, like, you know, you can still have a very standard name, but have fun with it as well if that's what you want. So if you haven't listened to that episode, I definitely encourage you to go back and check it out. What a fun roundup.

17:18All I'll say at the end here is save the fun names for, like, your devices. You know, just ask my producer what all of my headphones are named after. That's where I put all the wacky fun stuff these days. But this was a great article. If you have a fun naming scheme for your projects or if you feel particular at this, I want to hear about it. I'm like a project organization nerd, so don't be shy. And now that we've gone through our news roundup, stick around because after the break is my conversation with Minoj. Engineering leaders are doubling down on AI, but how do you actually measure its impact?

17:51With Linear B's new co-pilot and cursor dashboards, you finally can. Linear B brings all of your AI coding assistant metrics into one place. Adoption, engagement, suggestions, acceptance rates, and connects them directly to delivery outcomes. See how deeply AI is integrated into your SDLC, quantify productivity gains, and understand where trust is growing. You'll turn AI data into real engineering insights. Visit Linear B to try it free and start measuring what matters. Today we're sitting down with Manoj Mohan, a seasoned engineering executive with experience leading large-scale software data and AI transformations at companies like Intuit, Meta, and Apple.

18:39And Manoj is passionate about empowering engineers and simplifying complex infrastructure, while also maintaining a focus on responsible AI and developer experience. And he's wildly optimistic about what the future of engineering holds. And we're really excited to sit down with him today. So Manoj, welcome to the show. Thank you, Andrew. Thank you for having me. And thank you for that warm introduction. Look forward to our discussion today. Yeah, likewise. So, you know, we're here at the Engineering Leadership Conference, And you're here on a panel today talking about AI is here to stay. Can you tell us a little bit more about that?

19:14Yeah, yeah. First of all, super excited to be here at the ELC annual conference. It's super energizing to be here and be part of the Linear B team and everything. So the conversation we are having in the panel discussion today is how are engineering organizations going to evolve in a post-AI era? Yeah. So that's the panel discussion we are having. I know it's kind of top of the mind concern for engineering community and things like that. So that's the conversation. And I'm happy to spread some light if you want me to. Yeah, let's park there for a minute and learn some more about that. So AI is here to stay.

19:53We talk about that a lot on the show. And it opens up a lot of opportunities for teams. And it shifts everyone into a new paradigm, a new way of thinking. So what challenges and opportunities do you see right now for engineering leaders? Yeah. So let me start with the forest view, right? So the common misconception that I hear from several folks is as AI is starting to generate more and more of the code, engineering leadership and engineers are becoming less relevant, right? That seems to be the resonating theme that's concerning to most folks. And in my opinion, I think engineering leadership is more critical than ever before as of today.

20:39And why do I say that? The reason for that is the onus of making a right judgment call, the onus of creativity, the onus of investing in right big bets based on criticality, all of this has to come from engineering, the rockstar engineers and the rockstar engineering leadership, right? And maybe an analogy that I can state here would be back in the 1800s when industrial revolution was happening. Everybody was freaking out that these big machines are going to take away all the jobs, are going to eliminate all the jobs. But if you look back at history, what happened is just the opposite of it. These machines actually spun up so many new industries, be it aircrafts, be it transportation, be it railways, that in turn created millions and millions of new jobs, right?

21:37So every transformation that we have seen historically always looks like it's an end or it's a doom. But if you give it its own test of time, it's going to translate into newer opportunities, the ability to do more meaningful work if you just let it carve out its path. So that's how I see it. I, you know, I sense a lot of optimism with how AI is taking over all of the mundane work. Yeah. Right. And one other analogy I could use is I see AI having the ability to generate code more along the lines of intern, having an intern in an organization. Right. Now, just because an engineering organization is constrained on capacity and say, if you feed in 100 new interns to that engineering organization, do you think it's going to magically solve all of those capacity constraints?

22:37Absolutely not. That's not going to happen. If anything were to happen, it is going to choke the organization even more. It's going to create collaboration bottlenecks. It's going to create innovation bottlenecks. It's going to create dependency bottlenecks and so on. Right. So I don't think the solution is that AI writing code is going to eliminate engineers. What is going to happen is it is AI is doing all of the mundane work and it's going to free up engineers to be able to do more meaningful work. Right. It gives them more space to innovate, to be creative, and to go after bigger, bolder bets so that we do more meaningful work in the longer run.

23:24So that's the optimism that I would love for everybody to think and thrive upon. Yeah. I like how you use the analogy of being back in the industrial revolution and factories being introduced. There's been a lot of transformative things that have transformed industries that we've talked a little bit about before and Dev Interrupted. That's a great example. Another one is like when the Macintosh came out and traditional artists kind of uprosed against the new computer-aided graphics. And the same thing for architects with doing drafting on a computer versus doing it traditionally on like a drafting table in a room full of architects.

24:00So in each of those changes, like you said, it opened up a new world. People had to step through a threshold. it's a bit of an uncertain time of what's going to be on the other side of the store but ultimately it's more jobs and more opportunities and more ways for humans to have a higher impact on the work that they do and it reminds me a bit of like earlier this year when everyone for the first time googled jevon's paradox to learn about remember when deep sea came out and the low inference cost caused the open ai to dunk down for a day because everyone thought oh if ai is so cheap that you can have it in all these different ways.

24:37And it's like, why does a company like OpenAI have its moat? But then what everyone realized is that when AI access became more available, it increased the demand for it even more. And so what happened is the consumption then went up even higher instead of then, oh, the costs that you or the benefit or the profit you gained from that going down. And that Jevons paradox applies also to engineers and engineering jobs. When everyone is an engineer, when everyone is building software, when it's at everyone's fingertips and grandma can vibe code, the need for engineering knowledge and engineering expertise is higher than ever before.

25:17Absolutely. And now all sorts of industries and companies and places where maybe before they had, maybe they had a technologist in-house. They're not a tech company. They had a technologist though, but now everyone within that company is a technologist in their own right. And that's like a really exciting time, but it's also a time of a lot of work for people like you and I, you know, to talk about it here at Devinterrupted, but also to take the stuff that we've learned, our foundational knowledge as engineers, and help train the world for this new era that we're in. Yeah. Yeah, absolutely. Very well said.

25:49You know, and every such transformation, there is a sense of skepticism. And for folks who don't understand the granular details that are happening, that skepticism just explodes into paranoia and things like that. Right. You know, the only way to approach any such skepticism is to drive more and more clarity, right? Clarity of, you know, of inculcating some knowledge within yourself and also trying to, you know, break down the barrier and, you know, create some clarity with, you know, with what is at stake and things like that. Yeah. So if you're an engineering leader right now, what can you do to build that clarity within your team to get everyone aligned?