AI Is Rewriting How Software Gets Built | Agents Of Dev Ep. 5

Control. This is agent dev. I'm in position.

Copy that. Dev. Stand by for go.

Standing by. Hey everybody. Welcome.

Thank you for joining us on the Agents of Dev podcast. I'm Mitch Ashley. I lead software lifecycle engineering practice at Futurum, uh, practitioner product development.

Lots of different things in software in my career. And I'm joined by co-host, of course, Brad Shiman. Hey, happy 2026.

Well, welcome to the, uh, the next year, the next era of what's gonna happen in AI Next. Yeah, thanks. Thanks a lot, Mitch.

And, uh, yeah, starting off slow, uh, clearly, uh, we're off to a very even slow, slow pace start to the year. Fantastic. Feels good, man.

But I sure to cut you short, so let folks know who you are. Oh, yeah. VP practice lead for data intelligence, analytics, and infrastructure.

Also here, futur and like Mitch, I'm also a practitioner slash analyst slash whatever, uh, tech TechnoEvangelist de uh, despair and, and Reveler, uh, at the same time. I think the term for that is geek, but Okay. Yes.

Which we both wait, not, not the sideshow geek. Let's clarify. Oh, yeah.

Not the and Bailey type, but, yes. Mm-hmm. Very much so.

Well, you know, so lots of things we can talk about. Uh, it's an exciting year and, and we're in the midst of our predictions and getting some things published out that folks will see. Uh, we're both gonna be on Predict, uh, which is a conference Predict 2026, that conference that Textron puts on, that's on January 15th at 9:00 AM Eastern.

com, you'll see, predict right there, uh, easy to sign up for, and a bunch of us from the analyst community are, uh, presenting. And there's other great presenters that are of course, too. So, um, you know, the few things happened over the holidays.

There are a couple things leading, leading into it. One of them was, you were quoted in an article that was making the case that, you know, this Python thing is nice, but it's gonna get taken over by Java. That's really what rules the world.

And, uh, so you, you data scientists go keep playing with Python. If you wanna go, the rest of us will will use Java. I think it's a, I'm maybe in a little tongue in cheek, but that's just essentially what the article said.

Not your quote in it, but what's your thoughts on that? Yeah, I, I, I feel the same way now, and even more so than I did when, when I responded to that, um, request. And, and it's that, um, you know, I think to believe that that Python and Java and C and TypeScript, et cetera, you know, are going to continue as domains of, of expertise that will drive ultimate buying power in the enterprise, I think is a mistake.

Uh, I think that, and I think you, you and I are gonna chat about it on today's episode about trying to, to, you know, readjust how we view software development and how we view the, the tools that we use to build software. Mm-hmm. A language is a tool.

It's something you use to imperatively say, here's my recipe, please do the recipe. And some, you know, languages are well-suited to certain tasks, and Java has been to, uh, backend development. And by and large, because it has a terrific, you know, uh, mechanism for managing large scale projects through its o you know, object oriented programming and through automatic garbage collection.

So you don't end up with the things you run into if you're coding and sea at scale, for example. Um, but it, it's just a language. And at the end of the day, what drives language use isn't so always the, you know, how well suited it is to the environment it's running in or to the task it's supporting, but instead to how low the friction is that developers experience in using the language, how quick they are to solve problems with that language.

And as I started to say a second ago, you know, the further we go with Agentic development, I feel like the less important that becomes, because, you know, we're, we're basically seeing a collapse of all of the layers of abstraction that we have built since we were writing in, you know, uh, assembly language. Are we not? Yes, yes.

Our programming, those dip switches on the front of the computer. Let's go back to that. Yeah.

Well, you know, I can remember when Java was taken off and a lot of excitement about it, and it certainly has done extremely well, but I think that's more of a bias of well, and my world, whatever my world is, it's all Java. So the rest of the world, and it is a dominant language. No, no, no doubt about it.

In an environment, there's a lot of great strengths, you know, GRE, the runtime that Oh, yeah. Executes on the libraries, all of it's very good. But also Python came along for some very good reasons, and one of it was its usability and easy to learn and very extensible, whether it's PyTorch for machine learning or, you know, whatever library you wanna use.

Um, it's extremely extensible and people are writing some low level things in, in it as well as, you know, applications and everything else. So I, I think if anything is, if Java grows, one, you could argue cobalt modernization to Java is probably gonna be one of the biggest growth areas. Well, that, that's, IBM actually coded a or trained a model to do just that task, that one task.

Exactly. And that's not a slight to Java in mainframe or whatever, but I think that's probably the biggest growth opportunity, at least right now. Um, so I, I agree your point, and this is something we're gonna get into.

And so, um, Christmas, uh, Andre Pathy put up a post on X basically saying, I'm not sure if I know what I do, what I'm doing anymore with this whole AI using it for development. And I've sort of kind of reached the point of I keep falling back on essentially what I knew how to do already, and I feel like I'm not really totally leveraging what AI can do for me. That's my summary of it.

It's probably an inaccurate, but it gets a little bit of the, of the spirit of it. And he described it as it's kind of like aliens landed on the planet and dropped off these tools, and we're supposed to figure out what to do with it. Yeah.

So I think it was one of those moments, and, uh, I don't wanna be the old guy too much here, but I've, I've seen this myself when, like in the cloud, when everybody described it as, uh, it's just somebody else's computer. It's just like all the, you know, it's no different. We've been doing that for years.

And then suddenly you realize, well, when you remove a constraint, like, I can have how many servers available to me in, you know, seconds, minutes, okay. Didn't know I could do this. Yeah.

Maybe I can do this. Exactly. And that's kind of, we we're in that point.

I think that's, you know, well, this will be, it'll be one of my predictions about this year is kind reaching this tipping point with AI and development where yes, you can do things the way you have been doing it and still be the developer you've been, and kind of use it as a tool. You can also, not saying you go and divide coding, which is Andre also coined that, but, um, it's developing software in a different way. And I know you were experiment with some exp spec driven kind of stuff over the holiday that had caused you some questions about, Hmm, this is changing.

I am rethinking my rethinking. And maybe maybe Andrea is like, um, going in post vibe, vibe computing, uh, you know, like postmodernism, I, I don't know. But yeah, I, I'm with him and I, I felt this actually, and I will, I will like put my stake, my flag on the moon saying that, uh, back in 2022 when chat GPT first came out, I said, and WR wrote down the words, you know, it is as though we've been hand, you know, we've discovered a UFO.

Um, we know that it works. We don't know how it works. So we're joy riding around in this machine that we dunno why it does what it does, but by God we love doing it.

It's like this alien technology and Superman that humans don't know what to do with, right. Right. And so we get in trouble and we, we, we do, you know, learn, we learn how to, to run with those tools and how to use them.

And I think you could say that with every major advancement, you know, the, think about the punch cards used to program looms, uh, back in the turn of the, the century, uh, not this century, the one before, um, to, to automate, you know, something that we didn't think was a, and do we think that it would work beyond looms at that point? No, today. Yeah.

Right, right. So it's the same thing. And like you said, uh, it, it is forcing us, I think, to reevaluate and to live in sort of a space of uncertainty and a and a shift, much more shifting foundation.

And I feel like somehow generative AI and agent AI right now is, is simply one of many, uh, unsettling shifting aspects of the reality that we inhabit as a species on this planet at this time. You know what I mean? Sorry.

Good, good point you were making though. And that is, it isn't one thing that's changing. It's many things that are changing, not only how we develop software, how we think about Yeah, the economics of software, the economics of it, the who can do software, what are the skills required?

Um, how much can you do without having a lot of skills in developing software, you know, using ai, how far can you take it? You know, do you believe all the Instagram and TikTok videos or, you know, there's some limitations to those. Yes.

Maybe those aren't, you know, all the cases in the world that can be solved that way. But, uh, but we learned through failure, we learned through, you know, discovery. Uh, and those two go hand in hand.

Do they not? They do. I, I believe, um, you know, the fellow who discovered America, uh, Christopher Columbus went to his deathbed denying the fact that he discovered America.

Yeah. Well, um, and Alan Kay, who's an Apple fellow, he was at Zurich Park, and, uh, one of the things he coined was the dyna book with the idea for the laptop. That tells you how bad far back it was.

He had this saying of, I don't know who discovered water, but it wasn't a fish. It's like, that's right. We're just swimming in it.

Yeah. That's, it's so, so ever prevalent. And I feel like that's what change is of like, with this.

It's, and like, like you said, Mitch, it's, it's, um, it's uncomfortable and, and, um, uh, for everyone out there, Mitch and I were chatting just before we got on the air about this idea of spectrum development. And, and it is being touted right now as a means of making vibe, coding more enterprise scale and, and a trusted companion instead of, uh, skunkworks. You know, why did you do that?

Now we need to port it to something we can manage. And, um, it was, it, I I had a revelation, uh, actually recently about that. And it was, um, it, it's terrific at basically following the scripts that we've known for the last 60 some odd years.

It, it thinks in that paradigm, it's trying to apply a paradigm to, uh, this new way of development that that paradigm itself may no longer be relevant. I mean, there are aspects of it that are always gonna be relevant. It's like a good practice.

But to impose a, a way of thinking about software within an environment that, you know, doesn't recognize that boundary, um, sometimes isn't the best thing. And I, I ran into that headlong and, and I was, I, I had specked out a, a step change, uh, and it, we had a good plan, you know, with check boxes and everything, and, um, got about halfway through it and it occurred to me that, um, I, I wanted to take, take on a task that was not dependent upon it, but was, uh, impactful to it. And it, when I did that, it, it absolutely fell over hard.

Oh, interesting. Interesting. And, and was like, no, we need to finish this.

I'm like, no, no, we don't put that aside. So I, you know, had to basically tell it to just take a pack up shop, do all the commits set up its log. 'cause it's a nice, a nice facet of spectrum and development is that every time you make a change, it will basically not just do the commit with a lovely, you know, commit message, but also log the change in a very programmatic way such that it becomes a part of the living memory for your age agentic system layer, which is chef's kiss.

Um, but it, it was like so inflexible, uh, in, in how once one, once it had derived its spec driven plan, it couldn't, it couldn't deviate from it. It couldn't adapt to new ideas, new thoughts, and those new ideas. By the way, were sort of, you know, arose from the fact that I was using an agentic tool to do this if I was writing code, you know, manually, I would've just seen it through and then gone back and started another path.

Well, it's interesting. I mean, 'cause if it, my first reaction when I heard, oh, spec driven development, my one thought was, well, shouldn't we be, have been doing that all along? But Okay, alright.

I think that was part of it. We should know what we're gonna build before we build it. Yes.

Agreed. Yeah. But there is, you know, I have this, I have this belief of 50% of planning and designing is doing, you can only plan and define.

So, mm. I like laptop. Yeah.

You've got the other part of the learning of do is actually doing it. You know, I'm in the hands on with it. And it's interesting because the ID tools that have this design mode or requirements mode, development mode, uh, whatever they may term it, essentially using a different, they're switching context in a major way.

This is the new system prompt for design or for specs, and we're gonna work on specs. And Beth, I'm guessing that's part of why, you know, you get stuck in the mode of spec and then you switch over into a different system, prompt development. Well, it's, doesn't know how to go back and iterate and, and change to design and adapt, you know, it's, it's been told to follow the spec and maybe a little bit too firmly.

So maybe those things should be agents that work, but we together. But, you know, do we wanna impose those controls and constraints? We do, but we also want it to challenge and say, you know, now that we're doing this.

Yeah. Now that I see what you're really trying to do as we're building this. Okay, there, there's something, here's what I would do.

You know, it's like you and I would be sitting down developing on a, on a project and say, that was a great idea where we started, but actually given what we know now, let's do this, let's change Yep. And change the architecture a little bit. You know, add, add.

Don't do that functionality first. Let's do this. That was a mistake.

We go back and fix that. Yeah, exactly. I used to have this thought belief of it took about three tries to get the software right.

Of what you're trying to build. The first two were just sort of practice attempts. By the third attempt, you're kind of on the right track.

Um, so it, I I'm not, when I heard about the spectrum development, I get it. I'm not, I'm not a believer as in, oh yeah, that's the answer. That's the new paradigm.

That is not the paradigm we're shifting. That isn't what Andre is talking about. Mm-hmm.

This is the no, we're developing software not by just writing spec, not by inspecting code that gets generated or, or tweaking it and modifying it. It's, we're letting code happen. What we're directing is the process of how it gets created and all the inputs going into that.

And so it change, it's the changing of the job of the developer. Yeah. Yeah.

I mean, I know I've, I've harped on it before. Um, but, but, uh, I firmly believe, and I think Andrea is pointing to this, that what we're seeing is the death of syntax and the emergence of context. And, and like you just said, you know, you, you need to be able to describe the problem, and that's why, you know, it is speculated.

Uh, and, and so commonly seen right now that companies are looking for people who know how to describe problems. They know how to articulate that problem. They know the, the domain, they have the domain expertise to solve that problem, but they may not know how to code, but they know how to manage this, you know, swarm in many ways of, of agents and subagents that might be tasked with, with solving that problem.

So it's fascinating. So I'm curious your thoughts about this, what doesn't sort of jive with me about spec driven, that, that it's the answer to everything is software development is so incremental. It's one of the reasons why I thought DevOps is such as such a good idea or agile was 'cause you're not trying to tackle everything at once.

Right. Um, because frankly what product spec didn't change along the way. Right.

I think none of them, one is not what it starts out as. Exactly. So, so sort of finding it as a linear process again, and maybe I'm overstating that a bit, but still, it, it sure feels like it's, it's, that doesn't seem that's how we would create software.

It's not how I would design a product. Agreed. Even if it wasn't gonna be in software.

But what are your thoughts? Am I crazy at I don't wanna to No, I don't think so at all. And like we're talking about with the spaceship, we're driving around in, you know, this is all about dealing with ambiguities.

And when you're, when you're using a deterministic, you know, paradigm to, to run something that's probabilistic, you know, it's, it's not, it's gonna be a mismatch. It's gonna create a lot of, not just friction, but, uh, I think, you know, paradigm mismatches wherein, you know, people will be disappointed with what they build because they're just trying to turn, you know, waterfall into a new version of waterfall, you know, for the development paradigm with due tools, you know? No, don't do that.

And so, like, like with our example of, of spec development getting frustrating, you know, wouldn't it be great if you, you could have, you know, a swarm of agents and subagents that were each tasked with particular roles and, and, and specialties and skills that could work with you to say something to the effect of, you know, Hey Brad, I I know you just, uh, upgraded to the newest, uh, embedding model, uh, and I see that you, you know, are using, uh, a number of, um, uh, sorry, uh, you know, how you have, uh, the length of the, the string that you use, um, for, for that embedding, you know, is just a set of numbers. And I'm, I'm using a high dimensionality that's gonna cost me an arm and a leg. Maybe you shouldn't do that for this use case.

I would love it if, if there was, you know, a system that was probabilistic, that watched what was happening, not engaged all the time, Hey, this is gonna be high, a high cardinality field. Right? So let's sort of switch approaches here on how we're doing the database 'cause of that.

Right. Right. And maybe I had thought of it and forgot it.

Maybe, uh, I didn't think about it at the outset and so didn't put it in the spec, you know? And so I, I do believe that, you know, uh, we do, we have to live with in this realm of uncertainty and not just live within it, but embrace it and, and use it itself as a tool that uncertainty to, to allow us to solve problems, um, that we know about and that we don't know about. Um, you know, it's the unknown unknowns as we know.

Yeah. Are, are the, the problems. Well, speaking of those, uh, unknowns, and one of the unknowns that I, I think about is I also don't believe the IDE is the ultimate destination of what we're creating here for how to build software.

I think it's an, it's, it's what Andre's talking about, right? It's the extension of how we do things Now. It's the incorporation of, of literally extensions in, in ides.

And it's part of why you see this dual modality of I'm talking over here and these two windows and with the AI and uh, here's my code window where I'm kind of living in the world I used to live in only. Um, and, and I've described it as I think the next UI look something like StarCraft, you know, where we're managing resources and agents and tasks. And maybe it's not that fun.

I'd love, I'd love, oh, I know. Maybe it is a game gamified environment. You know, wouldn't that'd be awesome.

Find the best gamer, turn them into a software develop. Hey, we already kind of do that new, new job title. Yeah.

So there, there's a, there's a tipping point where we sort of cross over into, okay, there's a new paradigm, you know, is there another Kubernetes out there for ai? Mm-hmm. The control planes.

Is there something like that that's going to emerge out of either open source or maybe when the vendors become more dominant on the kind of the control plane for managing and governance and security guardrails, all of that. Yeah. You know, I keep looking for signs of those kinds of things.

So, you know, we're gonna make a shift in some of those directions. Too many of 'em already. It's the, it's the swimming and change.

Yeah. It's, it's the orchestration layer, uh, that where a lot of the money is going right now mm-hmm. From every, every model maker and every AI platform player.

And, um, and every tool maker as well. And oh my goodness. And, um, yeah, the reason why there's so much invested in it is that it's an area of difficulty and opportunity.

'cause they do go hand in hand. And, um, will there be like a Kubernetes idea that comes out of that? I would hope so.

Um, because I, I, I'm, I'm a proponent of open source all day long, okay? Mm-hmm. Especially with an FO in front of it, so free and open source, uh mm-hmm.

And, um, yet, you know, it doesn't preclude us from having a dominant player just ask the Apache iceberg team, you know? Yeah, that's a good point. Whether you hate JSON or not, uh, Apache Iceberg is the stratum that, that, uh, is now the norm within the decoupled, you know, data lakehouse in the enterprise mm-hmm.

For good reason. Uh, and an ecosystem builds up around it. So will there be something like that, uh, for, for orchestrating multiple agents and swarms of agents and sub-agents, and even development tooling itself then sort of sits in the background and emerges only when we need it in the context we need it in and to look and work the way we would want it to for that.

Like, maybe it looks like N eight N today, because I'm just wire framing stuff. Mm-hmm. Maybe it looks like, uh, you know, one of the cursor, you know, vs code spinoffs, you know, whatever.

'cause you'd like that paradigm, like the one you just described, Mitch, or maybe it's, it's something that's more like opal from Google, which we've talked about, which is like N eight N, but with no code anywhere, you know? Mm-hmm. Maybe it's all of those.

Maybe it's a freaking markdown file. Um, which I would, I would, I would love that a markdown Jason on a markdown. We can, we can tackle the world.

That's right. That's all you need. And, and EMAX and BS code.

Sorry, Neo, sorry. Not BS code. Oh, goodness.

Well, I, you know, I I, I tend to think that the likelihood is a little bit stronger that it's gonna be some open source type of solution, because I appreciate your feedback on this, Brad. And that is by being open source, it solves the multi-vendor dominance problem and, and vendor compatibility. You know, it's, it's one of the things about Kubernetes is yes, it's, it's, it really is vendor independent and, uh, people build ecosystems around that instead of, you know, Microsoft or Google or whoever is the, the dominant player in it.

So there's a lot of, I think there's a lot of benefits to that. I gotta imagine there's, you know, several teams out there, several Brads and Mitch's out there writing the next Kubernetes for agent orchestration platform control plane. There's a long title for You Gotta Think.

Yeah. Gotta think that's happening. Yeah.

And, and it maybe it is from one of the big players because we, we know that they benefit, you know, tremendously from open sourcing software. Mm-hmm. You know, just look at, you know, what, what Meta did back before it was meta with, you know, uh, AI software.

And so that itself, so yes. Code, that'd be an open source look at, uh, we wouldn't have all these IDs, I don't think FBS code wasn't open. We would not, no, we would still have Electron, which I, I really still am upset about, but Yeah.

Going back to emax. That's right. That's right.

But, but yeah, I, I've gotta think it's gonna come. And it's, it's interesting what we, the study we did this summer, past summer, uh, asking, um, data professionals, you know, why they used open source. And the top reason wasn't that it was, you know, free, it was that it was easy to integrate, you know, into their environment, into their stack.

And I think that will always be the case. And that's why we've seen the rise of so many, you know, popular open source projects, even though, you know, sometimes that can be a, a bit of a, a, I don't wanna say a trap, but maybe it is a trap when you think about like Redis as a, as a good example, uh, of that. Um, and you think about some of the, you know, complexities that when large companies buy open source, the vendors who are behind open source projects, you know, like HashiCorp from IBM, there's a great example.

Changing their license to be bought is what my opinion is what happened, you know, that is, that is what happened. Because it's because it is, that is, you know, the way the companies work is, is by making shareholders happy, not, not, uh, customers all the time. Mm-hmm.

Mm-hmm. So, you know, there, there are tensions in the industry and, and I think that what makes me feel good about ours in particular right now, is that the level of experimentation is, uh, I would say much more accepted and, and much more, you know, uh, supported, uh, within the broader ecosystem of those who would support it and, and drive it. So, you know, 10 years ago, you wouldn't see like the same level of, of like acceleration for, for building out new frameworks, ideas, and such that we do today.

It's incredible. Yeah. It is happening at breakneck speed.

Well, uh, for fear of overstaying, welcome again, I, we should probably go to our ending segment for this episode. So Lau let's turn our attention to the next, the closing segment, the drop. Okay.

It's time for the drop. Alright. I, uh, I think I kicked off last time with the drop.

You wanna start sort of what's, what's going on with you? What, what, where's your head space pointed right now? Yeah, yeah.

For, for me, the, the drop is, um, coming into a new year. We, we, you know, as an analyst, we, we try to put together a set of, um, predictions for we think is going to happen. And we usually do it annually.

And, um, we've chatted internally about this, uh, and, and I think all agreed that probably six, six months is the maximum window that we can really Two months. Oh my goodness. Yeah, right man.

No, twice a year. Twice a year. Oh, okay.

I I I was like quarterly. At least we should be doing this quarterly. Um, and, uh, you know, I, I think that that's okay.

And it goes to what we've been talking about and what Andre, you know, I think really pointed at so well in that, that post is that, you know, we, we are living with, uh, instability and those that can, um, swim in that current meaning. They, they don't, uh, you know, drown will, will, uh, i, I think, you know, benefit the most. And, and I, you know, you've, you've heard of like two paths and there's the middle way between them.

And this is a, a, a very old Buddhist idea, uh, that's been across many cultures. And that is that, uh, you know, the idea of WWE or, or no, the no way of, of not taking action is sometimes the best action. And I think that, you know, for those building software, maybe the middle way is the way right now, you know, maybe you should just not commit, just accept and use and enjoy and splash around in the, in the eddies the, and the currents and enjoy yourself.

Because I don't know if we're gonna see a similar sort of time come again like we are right now in our lifetimes anyway. Yeah. We never see anything like it.

It's, it's the, uh, jump in the water's fine, right? Sort of that, that strategy. Um, well the drop for me is I'm just getting ready literally to drop in the next few weeks, the next data set for the software lifecycle engineering.

And we've expanded that into a couple of areas. Um, more questions about, uh, the use of AI in development and operations, AI ops, um, in testing the importance of that and what the spending is looking like. And you know, kind of hint, hint, you can imagine ai, well, I always ask, so over the next 12 to 18 months, what are you increasing spending on?

And certainly AI across multiple dimensions and not just development code gen, um, is a big part of it. So, we'll, we'll be able to, uh, share that with folks and, uh, have that come out. And there's some signals coming up after that.

So I'll be doing one on observability too, so that'll, that'll be fun and taking it from there. But then, you know, two weeks from now I might say something a little different. Scrap it.

Is there any prediction for prediction window? Two weeks indeed. Well, thanks.

It is always great fun. I think one of the things I'm looking forward, I know I'm looking forward to, is more collaboration, doing things with you and with Nick and Keith and all of our analysts and, uh, Alex and Tiffany, the whole team here, as well as the, the, the companies across Futurum, uh, which is a lot of fun. So we had a really good last, good 2025 setting us up going into 2026, and we're ready to rock.

Um, it's a lot of fun. Disrupt. We are ready to disrupt, disrupt, disrupt the disruptors.

Yeah. That's very star treky. Yeah.

Uh, oh, wow. That takes me back. Yeah, there's fantastic film.

Fantastic film. Yeah. Well, uh, thanks for following.

Thanks for listening. Please subscribe. Um, we're on all kinds of podcast platforms, probably your favorite also on YouTube.

And, uh, we appreciate your feedback. com. There's a topic question, um, something you want to talk to us about work or about the podcast itself.

We'd love to hear from you. Um, I just didn't, if you'd like to join us yesterday, let us know, folks. Yeah.

If you wanna join us, you would certainly, we'd love to have some guests on and we do have people reaching out about doing that. So, uh, we look forward to kind of, once we get our feet under and this, I think we're getting there pretty quick, we'll have some guests come on and subject, subject them to this, which will be fun. We'll enjoy it.

Well, thanks Brad. Uh, we'll, we'll talk again soon on our next episode for sure. If not before.

Thanks Mitch. Bye. Everyone.

Control. This is agent dev. I'm in position copy That.

Go Standing by.