AI Dev Environments Are Taking Over: Why Tools, State & Documentation Matter | Agents of Dev Ep. 2

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

Copy that. Dev. Stand by for, Go.

Standing by. So, Brett, what's your view? Why do you think we're seeing so many AI ides springing up?

Seems like every major tech vendor thinks they need, need a ind individual development environment. They do. They desperately do need them.

Um, and I, I think in a word, I, I would say jealousy. I I think that there is, um, you know, a lot tied up in, um, how do, how do you put this, you know, uh, the, the dominance that Microsoft has enjoyed for over a decade with GitHub and VS. Code and the ecosystem that lives around and, and literally lives on those tools.

You know, that drives so much in our industry. So I, I look around the, the marketplace and I think, okay, you know, everyone who wants to do ai, you know, was trying to make models. And turns out you can only have so many frontier model makers.

Um, and I think everyone now wants to do AI dev because it's cheaper than building a model. And it, and it perhaps drives just as much revenue because what it does is, is it draws developers to, you know, your platform, your software, your, you know, viewpoint of the marketplace. Definitely opportunistic, right?

You wanna capture the mind, share the time, the attention of the developer community. And GitHub's enjoyed their 150 million user base. Um, not that the all of them use vs code, but a lot of people do, certainly, Or some derivation thereof.

Well, That's what I was gonna say also, I mean, Microsoft made it easy to easy, relatively easy. You don't have to start from scratch, right? Because VS code is open sourced and not all of Microsoft's innovations, but you've got that to work with.

People already know it and understand it. Yeah. So it, it's, it's a weird situation.

I know you were comparing this to the browser where wars comparing, but not comparing. 'cause it's not the same thing. They're very different.

Yeah. I mean, it's this same idea. You know, you build an ecosystem of value around this point of interaction.

You know, if it's a consumer looking for where to buy the cheapest gas, or it's a developer looking to, um, build something a little bit faster, a little bit easier, you know, those decisions are, you know, entry points. You know, it's like knocking on the front door of a house and looking in and going, yeah, I like the living room. I think I'll stay.

Yeah, great. He goes there, kitchen. All right, we're good to go.

All right. About some bedroom somewhere. We'll figure that out.

Right? I dunno what's happening in the basement? I don't wanna know.

Uh, let's just stay on the first floor. Yeah. Hopefully down, like you said with the browsers, it's, um, it is, you know, a point of entry and the browsers, as it turns out, are very difficult and expensive to make.

Um, and a lot of that is not just the engineering, but creating that, um, uh, exposure, uh, acknowledgement, um, momentum in the marketplace. Remember the browser wars in the nineties? Oh, Oh yeah.

And, uh, yeah, that, and before that it was a search engine in the, in the browser. Yeah. Wars.

And you know what I, so well, let's talk about some of them. I know you were just working with Project Bob. I'm sure you're not ready to give a, a full, you know, evaluative read out just working with it for a day or so, but what was your take on project Bob?

Yeah, I, I, I didn't actually do any modernization of cobol. Um, but darn, I, I do actually intend Ball you got laying around, Right? I have so much.

Um, but I, but I do honestly really want to, to try that use case because it's one of the things that's unique about Bob, and I think it's one of the things that, you know, when we talk about this Cambrian explosion of ides, you kind of have to think about, because what IBM is doing and what some others are doing as well, is creating these sort of value propositions that are tied to a particular problem space, uh, or market itself or platform itself. And so for IBM, they, they actually, you know, develop some operationalized, you know, workflows and, um, fine tunes of models that are specific to the problem of modernizing cobol. Where else do you get that?

Yeah, they, they do have, probably have some hanging around as some of their customers do. They they might. And that, and that's really it.

It's like, uh, the same reason why the New York Times is suing everybody to, so it is, you know, in this day and age, the agent AI age, you know, that unique content is worth quite a bit. And that IP that IBM has been building up over a decade upon decade upon decade with COBAL expertise, that's your one stop shop were to get it. Very much so, you know, the, the other one I didn't necessarily anticipate coming out with an IDE was Kira with, with AWS and when you really step back and say, well, why would they do this?

They really do wanna capture the developer's attention, get the mind share. And I think once you, if you can establish that as your sort of portal or window into all the services that can hang off of an ID, um, yeah, totally makes sense. I mean, a WS has been a little stingy with accounts.

I dunno if you've played with it. I have no Chance. I used others.

I have. Yeah, I you have. Okay.

I uninstalled it. Uh, actually just the other day. I mean, how many idd can you install, right?

Is that, Well, right, that's just a, there there's only so many electron apps that a single laptop can run. That is what I have discovered. Yeah.

But, but, And seriously, I I, I did not install it because it, it did anything bad. It was just trying to keep my, my living room clean. Um, and, you know, I, I felt like in using it, and in using Bob in particular, they, they are very consistent.

And if I had never used either of them, but I had used VS code, the barrier of entry is almost nil. You know, either adopting either of them or moving from one to the other. It is that familiar.

Yeah. Well, especially that model that that's kind of, even if it's not BS code based, it's still the same idea of how the UI is designed. Very much so.

'cause it is familiar to people. I mean, I've used Quad code, um, you know, the, uh, I forgot what the name of the engine underneath it is that you, you use, uh, within, uh, via studio or vs. Code?

Um, yeah, I'm not sure Either. Yeah, there, there's, there's a name for it, I forget. Um, but you download that separately or you can do it in, in ins, do it there.

You know, it's also interesting that now, and once the IDE started taking off, then we're in parallel, CLI become cool again, right? Yeah. And guess who, guess why?

You know, I bet every developer in the world says, this is great, but I don't do everything in, in this interface. I still do a lot on camera. Well, I might have my favorite editor that is not vs code, you know.

No, my, my EMAX needs some love. Mitch, I thought you were gonna go back to vi you know, I, uh, yeah, well actually yes, I, I am a lazy vim because, you know, it, it's, you had the, the funny, um, Marine, you know, this gun is unique and it is my own, uh, et cetera, et cetera. I I I just want the same neo vim experience that everyone has.

There you go. There you go. I, I asked a really good friend of mine one time, why do you always type with your left hand kind of hanging off the top left of the keyboard?

He goes, because I use vi hit have to hit escape all the time. He's still trying to exit. Yeah.

I dunno if he still does that today, but, oh, okay. I get it. I get that.

But it's interesting. Uh, so much has gone even emulating a lot of what you can do or, or doing the same things you can do in the browser plugins, uh, generating code, um, code. The, the other thing I wanted to talk about too is some, uh, opinionated ides.

So what I mean by that is Project Bob spec driven, right? Yeah. Um, uh, or intent, sorry, intent driven.

Oh, it's both. And then we have both. Yeah.

Mm-hmm. I think is, they call it It's is what calls, but it's a combination. Yeah.

And there's even, you know, kind of going back to what's the playbook, right? Development, planned development playbook that you put together on a project. So, so let, let me run this theory by you here.

Here's my take on what's happening because things are happening so fast. We're seeing companies sort of walk up the value chain. Sometimes it's a walk up, back down, back up of the value chain of the kind of tools that developers use, right?

Starting with gave you a natural language interface. Then we give you some, some IDE capabilities. Then we add to that, um, command line interfaces.

Then we add more agent creation capabilities. Now we're adding more the things like AWS, um, agent HQ for managing, controlling. And it's not the only one, right?

ServiceNow has theirs. That's the next battlefield, I think. So we're seeing everybody kind of walk up this value chain.

Of course, along the way all the vendors say, and we'll be compatible with everybody else, especially the models use your own models ex I guess except for fewer anthropic or open ai, less so. We'll see. Um, but it, it seems like that's what's happening.

And I don't know where, where we'll kind of meet in the middle again, but I think it's that control plane of managing what we're creating in software when it comes to agents. So that, that's my hypothesis. You can, you can tell me that's who, or you have a different one, or that some of that seems valid to you.

No, it is, it is All it is cyclical. Um, or a wave form, whatever, whichever way you wanna look at it. Si Is this a sine wave?

Exactly. So I, I don't know if we're in the tr or not or at the peak, I don't know. Um, but uh, it, it does, it does seem, you know, if you think back in time and you think about if you're a Java developer, let's say, and you know, you're a net beans guide or you eclipse guide, you know, that's mm-hmm.

You're kind of make an investment in knowledge and that investment is tied to the software you use. It's just like the Adobe ecosystem and what still powers that today is control, uh, and enclosure. And I feel like the industry right now is, is indeed walking up that stack, as you say, but at at the same time, it's trying not to reenter the Adobe mindset.

Mm-hmm. So if I think about my own, you know, dev tool chain, I, I want as much flexibility as possible. I, I want to invest in, um, vim motions because that's, I can use that anywhere.

Like I can use that in the browser for crying out, you know? And, um, I want to invest in, uh, some open, uh, models, uh, that are available both hosted and local. I want invest in, uh, a tool chain.

The meaning, sorry, the, the interface that, that can work on this. I, I want to be able to use, you know, Claude code today and Gemini, CLI tomorrow and open code the day after and work on the same project across those. Because maybe if I'm doing something like, um, you building test cases, I wanna use open code for that because I like the way that it does planning.

Um, but, but if I'm just trying to like, debug something, maybe Gemini CLI, or probably Claude, you know, code or CLA CLI is the best for that. So I feel like we're, we're going up the stack, but at the same time, we have a, a lot more optionality for what we, you know, the investments that we make in knowledge to use those tools. Maybe it's a spiral stair ski staircase that we're going, we're circling.

Wow. We're It's a stairway to heaven. Stairway to heaven.

There you go. There's a, there's a guitar solo you're not supposed to play in the, anyway. Um, well, good.

Any other thoughts on this? Because we, we'll go to our last segment. We're gonna introduce a new segment coming up.

Anything else you wanted to say, Brad? Yeah, just that, um, you know, if, if you guys, uh, that are listening to this are, um, you know, interested in trying out, you know, these age agentic ides, I would encourage you to, to pick up some of the, you know, the full spectrum. So like, start with the CLI tools.

Um, add in the IDs, uh, the IDs, and go for the, um, completely code free. And I'm not gonna say n innate N because I just know, uh, but, but you know, there are a lot of interesting tools coming out. Like, um, Google's DeepMind Opal as an example, that you can, you build workflows with that will generate the lovely TypeScript, you know, that you can, you can then work on.

Um, so there, there's like a full spectrum of, you know, tooling available that, you know, use the same models and build the same artifacts, um, that can match, you know, both the, the problem you're trying to solve maybe, or the project you're working on, or just your mood for the day. Good. Great.

Um, so let's introduce our next segment. We have a little new intro video for this. So guess what time it is?

It's time for the drop. Okay. It's time for the drop.

I didn't bring a microphone. Wait, drop microphone, Mic drop. That would be another kind of drop.

Okay. All right. So we have a little fun.

We liked our intro, so we made another little video clip there to use. So the drop is our last segment that we're gonna do most, most podcast episodes. Kind of what's on your mind, um, since kind of you, you were wrapping things up there.

Uh, one of the things that is on my mind that, that I'm working on is kind of finishing up the work I've got for the end of the year, right? Because I've got just a few more full weeks of, of work. And one of the reasons why I wanted to talk with this is I actually wanted to pick your brain to steal it 'cause I'm your, your ideas and combine 'em with mine and put it into an, an analyst insight report.

So I have to credit you now, uh, which is no problem at all. So that, that's, this is definitely on my mind is where is this going? And I know that developers will choose the tool that both fits the environment they're in, but it's gonna also be the work, the work and the workflow and the experience that they feel is most productive.

You know, I think it's a appropriately, so a choosy bunch, because you're not writing just code, you're doing a lot of other things in your IDE. And so we do more agent things and control planes and guardrails and all of that. You know, that's where we're headed.

So we'll see what the shakeout is. How about you? What's on your mind, Brad?

Yeah, we, we talked about it just briefly a minute ago, and that, that whole literate programming thing, uh, because one, one thing I, I've really kind of learned the hard way, um, in managing a number of projects, uh, that are agentic, uh, in, in the nature is, um, remembering what the heck is going on in there. And, uh, you know, when I, when I think because I, I used to do a ton more work in Jupiter lab. Nope.

You know, notebooks. And, um, that is the definition of literate programming wherein you are coding and documenting as you go. And, um, I, I've tried, I've really tried to, to sort of, you know, document each session, uh, to sometimes like with open code, for instance, you can save a session, uh, or you can even share a session terrifyingly, so with other people.

Um, thank, fortunately you can like, you know, as a hosted server, so you can just give somebody a URL, they can see your session. Um, but you can, you can kill that fortunately. Uh, but anyway, um, I'm trying to, you know, once you get to a certain, as a single human being, you know, working on a project that isn't, you know, very small that grows over time or is multiple projects, you know, being able to, to sort of maintain state, if you will, is I think harder in an age agentic world than it was in the previous world that we lived in.

And I appreciated Jupyter Notebooks because it kind of, you know, was that was a part of what you were doing. So I've, I've found myself trying to like think about how I can force the age agentic experience into better documenting, doing more literate programming with what I'm doing, not just like writing better comments in the code, but actually, you know, documenting what the project is, how It works, and how it evolves over time. Really, really, you know, it's something you hadn't, I hadn't thought about this way before in what you just said.

That was, we're managing two memories, our own and the memory state of whatever sessions we have working with ai, sometimes getting those to line up, like, okay, what did we do? Do you, you know, recall that? Is that still in your, in your memory buffer?

And is your memory memory, is that large enough? Or how do you, uh, persist that, all that? So anyway, anyway, lots of good stuff.

We have plenty things to talk about. No doubt. com.

So send ideas, comments, what do you like, not like We'd love to hear from you. We're both available on LinkedIn. It's probably easiest way to find us.

Um, we'll go from there. Yeah, everybody, we'd love to hear what you would like Mitch and I to talk about. Um, you know, we, we have a lot of varied interests, um, that we're, you know, quite willing to, uh, to dig into.

So we, we, whether we're qualified or not, we'll dig into it, right? Right. And we'll have fun doing it.

Whatever it's, so no, we definitely love to hear ideas of what you want us to talk about and, uh, might even spur some kind of new research areas that we look at too. So thanks for listening everybody. Thanks Brad.

Thanks Corey on the back end our producer for helping us make this happen. We'll see you on the next episode of Agents of Dev. This is agent Dev.

I'm in position copy.