AI Agents Are Changing Software Development | Agents Of Dev Ep. 4

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

Copy That. Dev. Stand by for, go Standing by.

Hey everybody. Welcome to the Agents of Dev podcast. I'm Mitch Ashley.

I lead the software Lifecycle Engineering Practice Analyst practice at the Futureum Group. Brad, welcome. Always good.

Be good to be doing this with you, my co-host. Yeah, it's, it's great to be doing the show with you today, Mitch. And, uh, I think, I think we have a, a fun topic, uh, on hand.

Some something that I, I'm sure a lot of people that are listening in are, are exploring and or dealing with right now, which is the idea of Audi use age agent to tools to, to actually build code on day two, not day one. Mm-hmm. Mm-hmm.

Listen, when you go beyond the travel agent book, my book, my Flight, find me a Flight example on every developer blog, how do you use this idea to actually do, do your own work? So, right. We are talking about, but we're gonna probably touch on three of them.

I know we, we've had a lot of the most recent activity with Google Anti-Gravity with AWS Kiro, some announcements at, uh, at, at uh, AWS Reinvent, and of course Bob. Bob. Uh, what about Bob?

Some BM And what about Bob? Sorry, I'll never tire of that. What about Bob?

Bob? I'm a child of the sixties, seventies, so I, I definitely know about Bob and we have of course, you know, copilot and all, you know, cursor and windsurf, and there's all, you know, everybody has an ID these days, which I can't remember if I said this on a previous podcast, but, you know, I, I thought all those developer jobs were going away. We're sure creating a lot of tools for these people that don't, aren't gonna have jobs.

So I don't believe those jobs are going away. They're changing for sure. That's for sure.

Um, yeah, actually, you know what? I, I think you're right. I, I, I like the notion that, um, all of this tooling is going into, um, helping people who, uh, do the jobs that they lost.

Hmm. There you go. That's basically to, to do the work that you used to do, but do it in a new way.

So you're saying this is a retooling to how to get your next job. Next job. Is that what it is?

Yeah. Just, just to get a raise. 'cause otherwise you have to quit and come back.

That's right. Well, I think that's, that, that's an axiom of the universe. I think that seems to be universally true in about every That's so true.

It's funny when you give people that advice, how do you get a raise? Well, you should probably go, go somewhere else and then come back, go get a lot more money tends to happen. It does, tends to be true.

Um, of, of, of the, of the ides, have you spent more time with one than the other? I know you do a lot of development, Gemini, um, but of course anti-Gravity is pretty new stuff. Is there anyone you have more familiar with than the other?

Yeah, well, I, I, as you say, I use Gemini See a I a lot 'cause I, I got into computers because of Ask Art. I will stay in computers because of Ask Art. So you have that mono still on, still, you know, on Green Bar in your It was, it was, it was Jerry, Jerry Garcia, uh, believe it or not, was, was my first like, mind blowing ask art experience.

Mm. Um, and so anyway, yeah. So I I, and as you know, I, I have a profound hatred of electron, so I, uh, don't, don't use many of the electron based tools, which are quite a few of them right now, because so many tools are based on vs code.

That's great because it has a rich ecosystem and a lot of plugins, et cetera, et cetera. But I've been, I've been using, if I'm not using z uh, which has agen tooling built in, uh, I for, you know, non Gemini, CLII, I've really been gravitating toward Open Code, which has a degree of what we're gonna talk about today, baked into it. And I've noticed this more and more with these tools that, um, you know, they have two modes and you can just toggle them, you know, from the command line because that's why we use a command line so we don't have to reach and click, um, anyway.

So you can basically just toggle to, to thinking mode or planning mode and then toggle back to coding mode or fixing mode or whatever you want call it. And it can be anything you want actually. 'cause you can customize those to be whatever you want.

A, uh, uh, was it Claude's, um, uh, skills kind of, kind of idea. Mm-hmm. Yeah.

Yes. The skills. Yeah.

And, and part of what this, this toggling does to planning mode is to help you to, to set up, you know, a very rigorous or supposedly rigorous, um, methodology and framework for doing something. Be that creating your new find me a flight app or on day two, uh, fixing your flight app because it's throwing a silent error that you can't trace back to, to the source. Mm-hmm.

So mm-hmm. If, if, if you go into planning mode and say, wow, I can't find this error, it will say, okay, he doesn't want me to just jump in and fix it. He wants me to think about what the problem is to find the problem, to find the steps to resolve the problem, and then we can switch back over to execute mode and, and fix it.

And that's spec driven development. Mm-hmm. What we're seeing right now, uh, is sort of a maturation of that.

We're in a lot of these tools are sort of bringing in, uh, purpose-built, spectrum driven tools. Sorry, these IDs are bringing in spectrum tools. So you have, uh, as you mentioned, Bob has it baked in, uh, KIRO from AWS has it baked in?

Um, you can download and use, uh, OpenAI, or is it, sorry, it's GitHub's spec Kit. Mm-hmm. Which you can actually run in pretty much anything you want.

And these tools basically do a couple of things. They'll, they'll let you sort of initialize your projects if you've never run them in there before. And when they do that, they'll, they'll do, actually, I wrote it down 'cause I did this last night with Conductor, which is plugin, they call it, it's the Google, Google calls it an extension.

An extension. Yes. We need a different term.

We do. Well, because, because it's really different than just using an MCP server. Mm-hmm.

An extension is the entire package built into the surrounding, um, tool itself. And so it can itself include MCP servers. So it's like a mod in, you know, Minecraft or something.

Yeah, totally. Right? Yes.

Yes. Uh, for those who know, Mitch enjoys a bit of the game. I have been to partake a few or this last one, and it's, it's, yes.

We must talk about that, by the way, because, uh, I, I, I found that very fascinating because what, what Mitch did this weekend in sort of, uh, asking agent AI to talk to him or to create for him a, a sort of means of saying, what does Mitch like to play? I found that really, really cool. I asked it, this is with chat GPT, just as an aside, I asked it.

I said, so I do a lot of work with you. You know, a lot of my preferences have explicitly created artifacts with you about a number of them, none of them about gaming, but have asked your questions. So analyze for me what kind of game, what, what kind of games do you think I like?

What kind of games don't you think I like? And why? And it went through and it said, and it was very accurate.

And it's, it's because I, I have such short time windows. I can't do games that are, you know, sit down for, for a weekend and really play the whole thing. Like my No, no.

Skyrim for you Skyrim. Well, you know, I could, so, so it's possible, if it's possible, and you'll have to step back in and like, there's a whole, you know, storyline that you're following that you have to remember that you talked to this person way back when. Right.

All that kind of stuff. So if you can kind of step in and out of it, it's a much more accept. So, so even like, um, a Civilization Civ six, I'm not a fan of seven yet, um, but it is com gets complex towards the end of the game, the end game.

But I, I can pause, I can save and come back, start over, do whatever. That's kind of the games that I like. Yeah.

This, so when it, I have to ask, do you have memory turned on in chat GPT? Have you, have you had it turned on for a while? I have.

I've had it turned on for quite a while. Probably two, three months at least. And I think that that actually speaks to the topic of, uh, hand today.

Does it not? Because spectrum development is, is really nothing more than treating context like a managed artifact. Mm-hmm.

That sits next to your coat. I was thinking exactly the same thing in, in getting ready for this podcast is podcast. You can see the parallels between the, the, the, let's call 'em retail versions of these products.

The, uh, Chacha BTS of the world, Claude, Claude, et cetera, is very much the same thing as they, they're creating what they call artifacts. And, uh, like you open a canvas that's now an artifact that within some context window within a particular tool, that it's gonna remember those things. And so if you want it to remember, one of the things I did was create a little memory system at one point where I wrote out, uh, remember these things out to chase.

Oh, you didn't, you didn't make a graph database to No, I I didn't have that much time that weekend, but it was experimental. Kind of see how would this work and recall. And so, um, but that, that's what, that's what these kind of spec driven, or the intent driven, I guess is what Bob calls.

It's interesting to me that we're, weren't we doing spec driven development all along kind of. Yeah. We kind of were, but now we have a way of actually integrating it into the workflow, you know?

Yes. It was, yeah. I was actually stories and all that before a spectrum and driven human in a past life.

Uh, I was a business analyst. Um, and so I would work with the developers to define the project, to, to set all of the goals for the project, to define the tech stack we were gonna use, et cetera. And that's, that's really what these tools do.

Mm-hmm. So, to, to jump back to, you know, this conductor extension for Gemini, CLI, I, you know, had a project that was only, you know, 350 lines long. So very tiny.

And it did something. It was, it was basically a research agent. 'cause you know, we're analysts, so always looking, it's looking for someone to do some work for us.

Right. Not lazy at all. And, and, um, so when I, when I initialized it, it, it did a number of things and it was, it was kind of interesting because it actually opened up and ran within the CLI tool itself, another CLI tool that had a user interface where it, it was basically a, A, B, C, you just would say it would ask you a question and you'd answer a, you know, and then hit return, or you would enter, uh, just a sentence to say, no, this is what I mean.

And it would guide you through the steps of trying to initialize this project for spectrum and development. And, and what it got to at the end of that process was it defined the project goals, it defined the tech stack, you know, by looking at what I had, which is something we should come back to actually, because it's an ongoing pet peeve. And, and spectrum and development doesn't solve the problems that we already have in these tools.

Just everyone note that. Um, anyway, it, it would define the code style, uh, for it, the style guide, uh, PEP eight for me. Uh, and it, it, the overall project guidelines and, uh, importantly, and I've found this really fascinating, is that it, it identified and documented known blockers, uh, to the projects, to, to getting things done.

So a blocker is a constraint or missing information or whatever, you know, would keep the agent from completing whatever task you give it. So, so you finally found a way to constrain the agent from writing code. That's been the problem with the whole vibe coding, is it loves to write more code for you, more code than you wanted.

It does, yeah. In all the wrong places. Say some more.

In, in, in the kero, uh, IDE it very much has phases, right? Has, um, planning, design, and development, you're explicit. Yeah.

Just like you were talking about, you go into planning mode, and it sounds like that constraint is, before I move forward, I need these things, because that's part of the spec process. But if you defined that to be yours defined in the tool, which is a good thing, sounds like, I mean, you can prototype, you can take it so far, black box some things while you're working on the spec to try out ideas, but still contribute to bat. And, okay, put this in the spec.

This goes back in the spec, you know, to, uh, memorialize make that part of the artifact that you're creating memorializing memory is, is like such a difficult part. And goes back to context as artifact, which is what these do. And prior to this kind of of tooling, um, for me anyway, I, I would demand, you know, I would type forward slash I think it's memory.

Um, my, my memory's that bad. I can't remember. Um, but at any rate, you would, you would use an internal command to say, remember to, um, that remember that I'm using this version of this API.

Mm-hmm. And 'cause you know, that's kind of critical and is a big problem with Yeah. Large language models.

Um, so it would write that to a Gemini MD file in the root of the folder I was working in, or for the whole system, you know, for the whole my laptop. And, you know, if you didn't curate that, if you didn't go back in and carefully manage that memory file to, to get rid of things that have been, you know, ob obviated by, you know, discoveries or changes down the road, you really could get wrapped around an axle because it would be like, wait a minute, Brad said that I wanna use version X, but in the same file, he's telling me to use version Y. I'll, it's interesting what memory, what it, what things it will save.

Sometimes it's like, that's not one I would've wanted you to save. Right. That's actually not right.

Right. Which, which goes, you, you were talking about the tech stack earlier, and I find this for just using the, you know, using uh, uh, Google Gemini or, or chat GT when I'm not doing it inside of an IDE or a development tool is, you know, even when you tell it, the tech stack, first of all, you have, you really need to have a running list of what your development environment is, what tech stack you're using for this project is explicitly what versions, um, because it will either assume things. I mean, I've had it like say, okay, go download this.

Well, that, that doesn't exist anymore. They canceled that. They deprecate deprecated that actually, right?

Or, or worse, they use a diff totally different package manager than what you're using uv, PEX and pip, for example. It's just like Exactly. Toss a coin and package manager.

So you gotta be very explicit. I mean, I list everything from mm-hmm. You know, the full development environment that I have set up through, you know, through home brew or in, in that environment to, you know, what are the, what are the keyboard, um, memory, um, macros that I use on the Mac, because I might use some things in automations that way.

Like, and, and I have to explicitly tell it. Do not suggest that I use things that have been deprecated, do not use, suggest things. You are not in the product yet, but may have been mentioned in a product release.

So Yeah. It's, um, yeah, like, you know, if you're not using robots txt, you're already in trouble. It should already be there.

Yes. Yeah. Im sorry, more, more specific to what we're talking about though.

Sorry, I was just, I was just thinking about, you know, protection, protecting yourself, but, but, um, requirements, you know, if you don't have a requirements file, you're, you're asking for, you know, a misunderstandings, like, like you're talking about. So you gotta start from a good foundation, especially if this is a brownfield, you know, endeavor that you wanna use spec driven development for. Mm-hmm.

And, uh, actually, you know, I would add to that, and, and I know we wanna talk about the, um, tech stack a little bit more, but I, I, I felt from my short experience with Kiro and Bob and now Gemini, CLI conductor, that, you know, you need to buy into this. This is not like a, I think I might use a bit of it here or a bit of it there, or, uh, just try it today and not use it tomorrow. You, you know, are basically saying, I want this to be spec driven, and as such, I'm going to be using an internal tech.

Uh, what do we, what does, uh, conductor call them? Um, where, where you have a task, they, they call it, uh, hang on, I'll just look it up. What you're describing is opinionated software.

It has an opinion about how you have to develop software and you have to buy into that. It makes it very opinionated, right, man. Mm-hmm.

So they, they call them, um, tracks. So you have a track, and I, I should note that, um, it's a very agent doing the initialization for, for tools at least like this one in that mm-hmm. It read my code base, it asked me what my objectives were, et cetera, to do all those things we were talking about and setting up the spectrum and development, you know, basically a, a file structure, a directory structure filled with markdown files.

And I was shocked to see it come up with the first track. For me. It's like I, I just finished initializing and it said, oh, I really think that you, you want to do semantic search, um, uh, enhancement, because I, I was trying, um, Gemini's Gemini, uh, has a deep research, uh, which is basically you're not calling a model directly.

You're calling a, uh, a, a sort of implementation of a model that that is very opinionated. And like we were talking about Claude, uh, like skills that that's basically built on skills. Mm-hmm.

And this thing, which is if anyone's interested, is called deep research. Sorry, these are all hyphens. When I pause deep research pro preview 12, 20 20, it's a mouthful.

And it, it, uh, it said, oh my gosh, why don't you add semantic search to that? Because I had a different, I have like a, a different sort of, um, mechanism for doing this outline research, either through semantic search or through the deep research. And it said, why aren't you combining them, Brad, you idiot.

Mm-hmm. Mm-hmm. And, uh, so it specked out that that track with all the requirements, all of those planning steps that you mentioned to build this, it's really fascinating.

Well, I, I'm not an expert in, in any of those tools. We just don't use them, uh, you know, living inside of them doing, doing work on 'em every day. But it seems to me the kiro is the one that's bought into the process as much or more than any of 'em.

'cause yes, it has those planning phases and Bob has architecting, et cetera. They all have kind of some ver version of that. But kiro really, I mean, it includes specs for, uh, acceptance criteria, traceability testing, um, yeah.

Yeah. User guides. Um, it's really, you know, AWS made a big deal about where we're using this, this is our development environment across the company.

And you, you could see how that might be because they put so much into not just doing agent development or using agents as part of development, but making it, I dare I say methodology. I don't wanna start any methodology wars, but, um, yeah, some opinion made way stream programming. Programming.

Yes. Yeah. Well, I lived through a few of those.

The age agentic wars are upon us. Mm-hmm. But where whereas, um, like with anti-gravity, you could definitely see that they've leaned heavily into the agent first.

Uh, ID, you know, using multi-agent orchestration to do work for you. Right? Yep.

And I don't remember if they explicitly call it spec driven, but a, a, a version of that you, you can, it's kind of doing that points and screenshots doing that's like open code and uh, you know, they just have it built in as different modes of working. Mm-hmm. You know, it's one of the things I'm curious too is uh, 'cause I think antigravity is supposed to be good at working a much larger code basis, and that's sort of the, one of the, there's lots of challenges and different vendors have taken on parts of it.

And when you work with much bigger or multiple code bases Yeah. Keeping that, that in the context window and the MA working with that know where you're, you know, God forbid you say, you know, Hey, put in that search and whoops, that's not where I wanted it. That was a different project.

Right. Um, but that's part of the complexity too. Them being able to take on more complex work, not just coding tasks or types of code to write, but the work of writing code and the environment that you're doing it in.

Yeah. That's a big deal, isn't it? And that, and you can see that reflected similarly to Kiro in IBM Bob in that they have built in using their, what would you say, 50, 60 some odd years of experience with Cobalt.

Yeah. Don't come back to some something called bunch cards, I think. Wait decks are switches on a computer.

I believe those exist in their osi. They, they have fine trained, fine tuned, sorry. Wow.

They have fine tuned, um, their mo you know, the models that they use for these specific tasks, like you're talking about in managing a large code base mm-hmm. To, you know, they were early to, to come out with this sort of, let's use LLMs to refactor from one language to another or to trans transcode and, um, with, I think it was COBAL to Java that they did. And that has grown into Bob and it plays a role in Bob for managing large code bases, not based on what they got off TE stack, sorry.

Um, stack overflow or God help us Reddit. Uh, it, it's actual like, you know, IBM you know, professional services, you know, use case number two nine x 94, uh, has a very, you know, distinct solution and pat design pattern that's tested in, you know, some World Bank and now you can enjoy that and use that to solve problems across a complex code base. Love that.

It's, it is interesting. Yeah. The Bob leaned into the modernization path, right?

Helping people modern modernize. And a lot of that is, of course, the LLM that they're using are LLMs, uh, by default, uh, which pushes that button. You know, the, the other we haven't really talked about, I dunno if you considered an IDE cloud code, you know, is something you plug into your IDE or you can use it very much from the, the command line, but essentially you're doing the same thing instead of them issuing their own vs code version.

Um, you do it just like you do with, uh, I guess it's Klein is what it's called. I've used it for a fair amount on a project I was doing a while back. And then, um, same thing for, for, uh, for open, OpenAI models that you, you interface to it either through a window, uh, in the command line on a, in your IDE as well as directly on the command line.

So there's sort of these different phases of, seems like so far, anti-gravity is kind of the most opinionated in the user experience of we wanted you to work this way. AWS kiro is the kind of development process. Opinionated.

We want you to work through these design steps, these, these requirements, spec driven steps. Bob, on the re-engineering side, we'll see, but Bob was, is pretty early when we saw it, so we'll see where it goes. But the, the tracks that pe different people are going, and you can see part of it is the customers that they serve, right.

Lean, heavy, heavy into agent development for Google. That makes sense. You could see why IBM would do more of a refactoring and, and helping people modernize code.

Yeah. Because you know how we used to talk about data gravity, we still do, um mm-hmm. Data has gravity because it's all designed around locking your data into a platform that you'll pay for, you know, in, in perpetuity.

Um, but I think it's called cement cement boots. It's cement. Yes.

Yes. It's lucrative cement. Um, but it, you know, developer gravity, it means a lot.

It, it always has and it always will. And what I see evolving right now, um, is this sort of, you know, my walled garden, my semi walled garden is better than your semi walled garden either because it has better models in the back end or better tooling on the front end, uh, or better services in the middle of the two. Mm-hmm.

And so, you know, these are all I've said for a long time that the, you know, the frontier models were always meant to be platforms, not just models. And they are absolutely evolving into that. But the tooling, you know, the things like you and I have been talking about today that we decide we like something and, and so we invest time in it.

And if you invest time in it, this thing, you know, that, that we all know and love, uh, called not entropy, but, uh, uh, what, what is it where you build momentum and can't stop the momentum from going, um, can't think of it. It's that brownie in motion inertia. Inert.

It's inertia. Inertia in, yeah. Need, need, need to think about the second of thermodynamics that inertia carries forward.

And you know, the more in an organization, the, they get a nerd in these tools, the more likely you are to buy this backend services models that, that are associated with them. And that it, so you see, like, you see like all the, the, all the investment right now in, in, I think making these enterprise grade is critical and really a, a good step. Um, 'cause this is, we are, we have changed how we think about code.

Have we not Mitch? Mm-hmm. Absolutely.

It is, this is first time that I can think of that we really are buying into, at the very front end of the development tool, the IDE of how we're gonna develop software, right? It's been more of a window into our repository and tools and plug area. It's been the Swiss Army knife, right.

For development. Um, I call it the, uh, smoking shop. You know, I I I, we, they're all brands of cigars, but we'd like you to stay here and smoke them here.

That's right. We have a lovely room with leather chairs. Just sit here.

Yes, exactly. This is where you should stay. Not that everybody out there smokes cigars, but, and I don't that much either.

But anyway, it, it, it, it is an opinion way, opinionated way that organizations, when they sign up, they're gonna use this tool as their IDE, then that's the path they've kind of chosen to go down. So yeah, it's, it's, um, you know, how, whether it's cement boots or walled garden or a lighter touch, it's to keep you in that environment. Of course, they work really well with their technology and make it easy to do that as well as, uh, third party stuff.

That's the key with the platform, you know, oriented tools, the hyperscaler oriented tools you see from like Kiro and Bob, like we've been talking about. Kiro, you know, is steadily surfacing functionality that's been, you know, sitting inside of Bedrock and, um, their broader AI platform, for instance, SageMaker for some time. And that's only going to broaden out, you know, I, I can see all of their analytics tooling, for example, starting to, to be baked in with, you know, headless business intelligence being a part of your application development process.

Your spec will have a spec on how you want to visualize data and work with data, for example. Very good. Well, we've, we've overstayed our welcome here a little long time.

Yeah. Sorry. It's, it's the holidays.

It's the holidays, Mitch, you know, I think we're being watched and, you know, we're being trailed here, so we should, uh, get to our last segment. Last segment is the drop. Okay.

It's time for the drop. Well, you know, it's, it's, uh, end of end of the year for, uh, us going into the, the next year. We're, we're obviously deep in our planning cycle, kind of towards the end of our planning cycle of what we're working on.

There's kind of two areas of focus that I'm looking at right now. One is how are we moving observability, uh, security behavior governance into the development process? We're using agents and, you know, companies like New Relic and Dynatrace, and I could go down the list of companies who have been part of announcements part with some of the major vendors that they are now part of AWS's security agent or Dev DevOps agent, or pick your vendor.

Um, so other parts of the software development lifecycle are not waiting, um, not every vendor's doing this, but I think the folks that are wanna make sure the carving out a path for themselves to be part of this, uh, kind of AI SDL or ai DL c um, very much part of more letters Please. More. I'm, I'm warming up for New Year's Eve, I guess I don't know what I'm doing.

But anyway, that's, that's definitely on my mind going into next year, uh, as well as many other things. How about you, Brad? Uh, yeah, and I'm sorry, the, the longer the acronym, the better, you know, three letter acronyms are aren't even trying, I can't memory, I can't even approach the one you gave then.

Well, that was the name of something, but you know, it was a, right. Yeah. Yeah.

Uh, so I, I think for me, and, and uh, it's interesting, given what you just said, it, it jives with, um, one of the predictions I have for the coming year, which is that data engineers are, are going to morph a little bit into integration engineers and that companies won't be buying their data platform based on, you know, the specs of the platform as much as they will the meet the way that that platform integrates with their investments that might be in oth with other vendors on other platforms. So it's definitely, you know, changing how, how these age old in these stable jobs that we've had in the industry for so long. Um, for me, what's on my mind right now that I'm trying to like wrap my head around a little bit is I, I saw a note from one of my colleagues yesterday talking about how, um, agentic development was going to break the database.

And I'm thinking, well, I, I don't know, because we, we built databases when we had very little resource to play with, and we built them in order to scale massively and be concurrent as possible across that scale. And, um, so I I, I think the problem isn't so much in doing reads and writes, 'cause we, we know how to do that and mm-hmm. Pick your database structure.

It doesn't matter if it's, you know, a blob storage or column or NoSQL or, um, you know, KV cache or wide, you know, field wide table, sorry, um, like Cassandra, it doesn't matter. They, they all are pretty darn performant. Uh, the problem is agents, you know, need instant context.

They need, and I, you know, I think you and I chatted about this a while, a while back mm-hmm. With, um, if you're outta sync even a little bit, um, big, big problems can happen. And so timely access to accurate information is critical.

And how do you do that? Well, you certainly aren't doing a query from a, you know, data warehouse to do that. You certainly are bringing the data closer to the code, to the agents, and, um, you can do that through streaming, uh, which is why we saw IBM pick up confluence, uh mm-hmm.

Because that's kind of a big deal. And another way is the developer's favorite pastime, which is caching every you and managing cash almost every, yeah. Ev almost every problem can be solved with cashier.

Um, and uh, and it already plays a big role with the agentic tools we're talking about, like, every time I hang up from a session with Gemini, CLI, it tells me that I saved, you know, 80% because it had that many hits from cash, that percentage of hits from cash. Wow. Like, wow, I really am forging new territory, aren't I?

You're supposed to your manager. Yeah. They you 72 this, this trip to, because Brad is so boring in what he does, everyone's doing the same thing.

Um, post, post it on Reddi. Yeah. So I'm trying, trying to think about, you know, how companies can speed up, you know, access to data for, for agents latency, agent latency, waiting for those, for the data, the context that it needs.

Very interesting. Yeah. Well, my friend, it's been fun.

Um, a lot. I have so many ideas for, uh, things for us to talk to. I know you, you do too.

So we'll keep 'em rolling. com if you'd like to make a suggestion. Uh, if you're interested in making an appearance, maybe being part of this, we've got some folks that are interested in doing that.

And, uh, we're, we're kind of getting the, getting things rolling with Brad and I and then we'll start to have some guests as we do that. So send us the feedback. Thanks for following us on your favorite podcast platform.

And thanks for coming back to listen to, uh, to US Jawbone for a little while about building software in this new AI agent, agent era. Uh, behalf of Brad, myself has been fun. We'll see you on the next episode.

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

Copy that. Dev. Stand by for, go Standing by.