AI Bots, Security Gaps, and the Cost of Automation | Agents Of Dev Ep. 8

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

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

Hey everybody. Welcome to Agents of Dev, the podcast where we'll talk about anything related to development and have fun doing it. My name is Mitch Ashley, and I lead the analyst practice around software development, software lifecycle engineering with Futurum Group.

And my co-host Brad Shiman, also an analyst with the company. Hey, Brad. Coincidence?

I think not, uh, Hey, Mitch. Good to see you. Yeah, good to see you.

And you're, you lead data intelligence, infrastructure and, and analysis and some combination of those words, right? It's, yeah. It, it can be any combination.

I I, I'm not picky on that. It's not a dag it's a word jumble. That's kind of, it's a word salad.

That, that's the analyst world. World. Anyway.

Well, you know, it's, it's great doing this podcast because there is a, a plethora of topics that we can talk about every week. I mean, we have to like pare it down to like, can we fit all that in? Can we talk about all that?

Yeah. And then what do we say we're gonna do? Because there's a lot to choose from.

And it's definitely true. You, you brought up the, uh, well, it a lot of interesting things about Malt Bot, which was Quad Bot with a wd um, yes. Over the, that really kind of blew up over the weekend for several reasons.

One of, just because what it does, talk about how it was created, the, it has some kind of biblical reference or something. I'm not quite sure I'm making that up, but, well, I, I mean, any software developer worth their salt will look for any number of puns, obscure references as they can, and build that into the, the working title, which eventually becomes the actual title of their software. That is a given.

Everyone knows that the, the, the Living in infamy, the title of, yeah. Right. And is that a bad thing?

I don't know. Because sometimes it, it really gets the point across. And when I, when I looked at, um, Claude Bot, which is the, the current craze right now, um, as of 9 0 7 on Tuesday the 27th, um, it is, uh, you know, built by a, a fellow named Peter, who is what I would call a, a serial vibe coating enthusiast, uh, who's made a tremendous number of, of tools and is just, you know, a font of creativity.

Uh, and he has been working on for a little while, but has created this, uh, basically what I would call it is, is, is like an extension, or not even a, not a maturation, but an extension of what we saw with Claude Cowork. Um, but the difference is that it, it, you know, well, it works the same way in terms of it's designed around an autonomous agent that can, uh, do take action on your behalf, on your machine, uh, whether that machine is your local laptop or A-A-V-P-S, um, uh, yeah. In the cloud.

And, uh, the difference though is that this, this has a bunch of, uh, built in sort of connectors and a running listener, uh, which, which I think they call it Gateway, that will, um, basically just run 24 7 looking for things to do, waiting for instructions or, or, I don't know, checking Aron job, uh, or, or responding, I should say, to Aron job to kick off some sort of action on your behalf, whether that is, uh, go get me the news for today. So I have a nice in message in my inbox, or I get that message with news sent to my Slack channel. Um, or, uh, it's to send out a, a, an email in, in mass to your LinkedIn, uh, family of, of, uh, friends.

It can do all of it because it's just built to be a general purpose agentic, uh, interface to your personal life. Whether that's calendar, uh, whether that's files, whether that's your line of business apps, whether it's your code base or your desktop. Doesn't matter.

Anything local to your computer. Um, browser included, right? Yeah.

Because it has chromium support. So there you go. That, you know, you can, uh, give a, give a man a browser tab or you can teach him how to the full chromium API, sorry.

It's, it's a weird, it's too early to think too many times. Here's, um, but, but, but that's, it's okay to be early, and that's what this is, right? Because I, I guarantee that you and I will be on this podcast, you know, um, tech god's willing in less than a month going, okay, we're gonna talk about Google's, you know, Gemini bot and Anthropics Claude Bots, uh, which will be an extension to cowork.

Uh, and they're all gonna do this. Everyone is gonna do this. Yep.

It's, it's kind of what, uh, copilot should be in, in Microsoft land, right? If you'll, it's instead of the cut and paste mentality. And you, you and I have talked about this too, that, that, that, uh, the way that people inter interface with software is, is changing very quickly right now.

And, uh, we, we went through the phase of a separate chat window, uh, that you took, you know, cut and paste from, and then we went through the sidebar phase, which was a little more contextual, but still cut and paste. And now we're in the M-C-P-A-P-I, uh, era of just letting, letting the, the machine take action directly. Yeah.

So, and we'll talk about CCP another day 'cause there were some security vulnerabilities that came out about it. Oh, that, that won't be the last, uh, it is, it won't incredibly dangerous. Can we, can we just give a psa?

Is that okay? Yeah, sure. Go right ahead.

If you do this, everybody, um, please use it on a dedicated machine. Put it in a container that's, that's like cordoned off from everything. So meaning if you're gonna run it and docker, lock down all the ports, uh, that you don't wanna use or others to use, um, and, you know, don't, don't just put it on your, your existing work laptop.

That, that it is, is a very bad idea. And the, the main reason for that is the same reason that generative AI in particular is quite dangerous and will always be so, um, is that it's, it's subject to injection prompt injection attacks. Mm-hmm.

It's very susceptible to that because it's a probabilistic system that responds to what it reads and what it's told. So, and when you're, when you coding, if you're a citizen developer using AI tools, right? Mm-hmm.

And, and even a dev developer, not every developer is a security expert, you know, by far. Oh. But, um, it's really easy to generate yourself into a, well, we not only took the, opened the, the locks, we took them off the doors, and then we just took the doors off the, the, the house.

It just come right in. The whole thing, you know, the, the attack surface just went, you know, global. So suddenly, like everything in your life is now open up because it's all on your computer.

Yeah. So you have to be careful about what you're, what you're doing when you're going over the network and accessing services or using open source code, any of that kind of stuff. And that's what I think is particularly dangerous about, uh, Claude Bot here is, is that it, it, um, has an attack surface that is, you know, would, would really demand somebody who's, you know, steeped in system security to, to really set it up and manage it correctly, um, and safely.

Yeah. And interestingly, what you're doing, did you know it actually comes with its own auditing, um, workflow? Um, so I did not, you can, you know what, you set it up, you can actually just run like a, you know, an extensions command forward slash audit, and it will tell you, you know, what the, what sort of vulnerabilities it might have and suggest to fix them and obviously fix them because hey, it's autonomous.

Yeah. Yeah. Interesting.

Okay. So self-correcting, self-healing, if you'll, well, you know, there was a whole big controversy about a tune Now cloud bot was renamed to Malt Bot MLT because some, uh, trademark. Oh, yes.

Yeah. Well, you would be mad too, wouldn't you? Yeah.

Yeah. I'd be a little, little upset. I can't imagine why somebody had an issue with that.

It's the pro pronunciation. It's not the spelling. It's not the spelling.

Yeah. I, I dunno who could have objected to that. But during the renaming, uh, during kind of that, that window, uh, scammers hijacked the old GitHub repo and X accounts, so they, they onto dollar, right.

With an ad. Yep. That token.

And then they briefly hit about, I think it was $16 million market cap before crashing. Yeah. Yeah.

They pushed a fake dollar, Claude. Um, so it was, it, it hit a big, it suddenly, like, I mean, innocent thing to do, I need to rename my project. Right?

Boom. Okay, well, what do you do with the old thing? Oh, since that was, had its own following, let's take that over, right?

As a, as, uh, you know, the, the bad guys are wanting to do. So it, it, um, it sure brought to light a lot of, uh, attention to Claude Bot and Malt Bot, whichever, whichever one you want to call it. I I still call it Claw bot.

Yeah. Just for, just for, uh, nostalgia. That's right.

Thinking back to last week. Yes. Yes.

Fonder times. Yeah. That's all it takes.

Um, well, anything else to say about this? I, you know, I haven't checked it out yet. I'm gonna give it a try and, you know, see if I can open up my computer to the world.

Yeah. I'm gonna, I'm gonna do it too. Uh, I've got a Mac Mini that, that's not doing as much as it should.

So I'm, I'm gonna wipe it and give it a, give it a go see, see how it likes that. Um, but yeah, I do actually wanted to, to mention briefly that, um, one of the things that I, that I found very interesting about Mt. Claude Bot is that, um, it, the way it is, the way that it manages memory and, uh, you know, you, you and I are fans of, of e old markdown, uh, and plain text files as, as the lingua franca of all file formats that we should everything, eventually we should use it for spreadsheets, you know?

Um, but, uh, it's, uh, it's interesting the way that it, it's trying to, you know, Claude Mtba is trying to smartly manage memory using, uh, key value pairs and, and time to live as well as, um, you know, automatic trash compaction and or trash removal and compaction as well as manual. Um, you know, remember this kind of things that we we're all used to mm-hmm. Uh, in agentic tooling.

Um, so it's this kind of multi level of memory management package that they've built into the, the tool. And I think that's not how it's gonna be. Um, but I do think that it points toward what we should start to see, which is, uh, something that is less what you, the developer or you the user has to manage and, and instead exists as, as sort of a, um, you know, uh, I hate to say, say the words, but, you know, like a, a a message bus for, for, for memory and context for, uh, for models that just manages itself and make sure everything is running.

And it makes sure that the models have, uh, the right context at the right time. It's a, it's a challenge. You and I have run into this, and I'm sure every, you know, everyone, everyone has in their own way, but different AI tools, LLM systems manage memory differently, right?

It's all contained within a project. It's contained within, across sessions of the or not, or, um, you put it into artifacts, whatever the, whatever the method is, and they're all a little different. Or you use skills and use other ways of mm-hmm.

Persisting some capabilities. But then you switch tools and now you're all that, all that, I dunno if I call it training, but all that curating of memories and how you want things done and how you work and what you do, and, you know, the stuff all changes already done are kind of trapped in that. And, you know, you mentioned Mark Markdown.

I'd recently switched from doing a whole bunch of things in, in, uh, chat PT and moving over to Claude to do some other work. And I'm like, well, I've got this other stuff set up. So essentially in, in chat, PT created this, it's a 17, uh, file markdown, uh, liturgy of all the stuff from my writing style to how I use AI, to how I play games, to all these different things into this, you know, here's the index and how it works, and then put different files into different projects.

And Claude, because they're relevant, they don't need to load the whole thing every time into the context window, um, as a transition. Now, it's not perfect, but it sure like, saved me about three months worth of work to kind of get back to where I was. How did you make that, Mitch?

How did you create that, you know, corpus of knowledge? Well, it was interesting. I was, um, I was in Claude and I said, all right, here's my problem.

How do I do this? How can I, how can I port this over? And so it said, well, here's a series of questions to ask, um, to ask Chachi, bt.

And I tried that out. And after I got to that, I said, well, I need a structured way of doing this. 'cause I'm not gonna play copy and paste and, you know, that kind of thing.

And said, well, here's, here's a set of topics and here's a set of, uh, draft markdown files structurally, um, ask GBT to fill those things out. So it filled those things out. And then I went in and modify it where it wasn't quite exactly whatever.

And then I, I, I remember, oh, yeah, I have a system of work the way I kind of classify what work I'm doing and what its priorities. And so let's add that to it. Chat, pt, describe, I already had those things, documents.

So I took those things, put it into this format, and I moved that over. So it went from about 11 files to 17 over a process of, eh, probably a weekend plus. And now I've, then that's, and then the, uh, canonical references, obsidian, that's where it lives.

Nice. And so when I make changes to it, it always goes back there. And then I pull it from there into, now I use it in both environments.

I have it in, in gt and I have it in cloud, and I use it in Gemini. So I ha it's transportable i'll, albeit I'm the transport, but at least it's transportable. Well, you could, you could sync it with GitHub or actions or, or just use their built in sync, right?

And then Yep. Every machine you're on would would have access to that obsidian vaults and, and be able to Exactly. iMedia, by the way, EDIA Yes.

If, if you ever wanted such a thing here, it's at least itself, it's useful to me anyway. Yeah. But it was, it was an, a very interesting exercise in how much memory I had in invested Inchi pt and how important it was to have that.

Um, because it just saved me a lot of times. And along the way I kind of proved some other things. Like, here's some things I wanna templatize that I do frequently, and here those will be here and let's, let's do things this way.

So, you know, and of course, like every project it turned into seven other projects. Let's reorganize my vault while I'm at it, and, you know. Right.

Well, I, I, I think you, you probably got a ways to go before you, you beat Peter Steinberg. Uh, is it Steinberger? Yeah.

Quad guy. So get Mitch on, that's interesting point that you bring up to me with, with the way memory works with, with agentic tooling. And that is, um, locality.

Uh, and you know, I do appreciate how it has evolved over the last six or so months wherein if you're in cloud code or you're in Gemini, CLI, for instance, and you, um, you, you would have, as you have the Edia, that is everything you like about development in general. You know, I like Python, I wanna stay at three 13, you know, you know, on this more, you know, specific side. More generally, I, you know, hate Pep eight.

Uh, I made my own, I use five and a half spaces for, for this. And, um, you know, whatever other, each and three quarters line preparation, uh, you know, all of those, I don't wanna say ex and specificities, but, but the, the things that are specific to you, and the way that these tools have evolved over the last six months is, um, you know, such that I can have that knowledge generalized for Gemini CLI, for instance, but I can also have knowledge that takes precedence, uh, within a specific project. And I can have that for every project.

And it's not just, you know, two conflicting competing, um, bits of memory that the agent is working with, but instead a hierarchy, um, to, to say, you know, if I see an instruction for, you know, do use Python with four spaces for this project and go for this other project, you know, strange thing, you, um, you don't have to try to reconcile across projects so long as you, you know, create that edia for the project, you know, it can live with the project and it can have the broader context that lives with you, the developer that can, can move from project to project and from tool to tool. I like that. Well, you, you have described what I'm working on, which I call it the memory portability system.

NPSI have a board already, by the way, Theia called the context operating system that in co Yeah, yeah, yeah. Yeah. That's nice.

I like yours better though. Well, EDIA is a good, you know, maybe infamy, living in infamy, um, but it, it solves a problem. I don't know if anybody's working on solving yet is, and that's the transportability because it's in the interest of, maybe it's a good open source project, but it's in the interest of all the vendors to like, I wanna create the environment that you wanna be in.

Right. And so are you trapped there? No.

But you have so much invested in it, you have preferences to stay there. 'cause there's advantages, right? Well, that is their goal, is it not?

Absolutely. That is the argument for proprietary software, uh, or the argument for open source software with proprietary extensions or proprietary services on top as the monetizing force behind it. But, um, you know, it, it is very much the case.

I think that across the industry, especially for software development, that, you know, developers are lazy because, you know, developers are always looking, developers are the lazy, I mean, seriously, that is, it's, it's optimize everything. What can I do to optimize so I don't have to do this stupid task again and again, that's the laziness I'm talking about. Yeah.

And, and so it, it very quickly is a crucible of, of openness to, to it, you know, pushes you toward openness such that we, you know, with, um, MCP, even if philanthropic hadn't have donated it, uh, to, to an open source organization mm-hmm. Would have become a standard simply because everyone just like, you know, Claude slash Malt bot was like, wow, that's great. Let's do this.

Let's, let's make use of it. And it's the same with skills. Claude Code, you know, came up with skills relatively first.

Mm-hmm. Um, and that was a very simple markdown or yaml, uh, sorry, is it Tom or yaml? Which one did they use?

I can't remember now. I don't recall. One of the two, which I always have a problem with Tom, I'm pretty sure it's yaml.

Everything is YAML to Tom is GML with, with some like constraints. Um, but at any rate, um, you know, it's just text file with basic header information and instructions. Mm-hmm.

And that's very portable and as we have seen it already is portable. So if I go to Claude, um, you know, um, hub, what do they, what do they call it? The skills hub, I believe.

Yeah. I dunno. The name of it is, it is a good question.

Home skills Home, well, we'll, we'll, we'll, we'll put it in the link. Uh, but, but at any rate you can download those and run them in Gemini, CLI for instance, without doing anything to them. I can just install them the same way and run them just like prompts or code, text code.

Right? Yep. Same way.

Hey, if I can just kind of staying on the theme of, now that I have, uh, AI driven tools, I can create anything I want theme. There was a gentleman, uh, I think it was over, it was I think late last week or so. It came out that he had ported, and I don't know if his name is Peter too, but he had ported, chances are, chances are, um, he had taken Cuda code, uh, and ported it to run an AMD rock stack.

Uh, so it wasn't Nvidia that ported it, it was someone that talked about it on Reddit that what he'd done. Um, and he used cloud code to do this, to translate, translate the entire coda back end, um, into an AI engine for Ad Rock and did it in 30 minutes to do it. Right.

Oh my gosh. That is crazy. No manual edits.

Now I'll take that to mean the first time it worked was it took 30 minutes. It probably didn't take 30 minutes to, and I'm done and walk away, you know, drop the keyboard, drop the mouse. Um, but it really showed the, like, you know, what these tools can do in the hands of someone who is, I have, I, I wanna be able to do this.

How can I get this done? There's not a product, I'm not gonna wait for somebody to create it. I'm just gonna do myself, do it myself.

Now this created the whole thing about, oh, is this the thing that's gonna bring down the Nvidia Cuda moat? No. Somehow I think not the ecosystem is, is the key.

Yes. You know, he was not, you know, David and the Goliath. Right.

He had knocked the giant out quite yet. I don't think that's gonna happen, but it, it does bring up, for me, it brings up a lot of things. Now he, he put it under GP four for the license.

So it has some, you know, it has some strings attached. Right. So I'm not a big fan of that license.

I'm not sure why he did it that way. Yeah. Either had reasons or that's the one he knew.

Well, those gpls typically if you don't want your competitors making money with your software, but everyone else is okay. Yep. Gotta contribute it back if you, if you're gonna distribute it and you have to contribute your changes back.

Yep. Um, and g and four is even more restrictive. So yeah, it, there wasn't any mention of any kind of profit motive or where, where he might take this, but, you know, maybe he wants to be, be the, the loan shark that you hire to go port whatever to whatever.

And that's his, you know, mantra. I dunno. But it, it, it, it certainly brings up the licensing.

Like are you gonna see Nvidia and everybody else tweak their licensing to say, well, you know, they had already thing, I checked this out, they already had some restrictions in the Nvidia license, but talking about specific things like the SDK that you can't do with it. And since he ported the back end, it didn't really apply. Right.

Um, so is translating code, you know, is is kind of the term that of art that folks are using about this is translating code from A to BA violation of copyright of, of, or a violation of a, of a, uh, user agreement. I gotta believe there's gonna be a lot of reactions to like, we're gonna lock that down. We don't want our stuff to be reverse engineered.

'cause it really is, how do you lock it down? Is that is the question? I I get them wanting to do that.

But if you can reverse engineer and create something without just transing transcoding, you're refactoring, um, is that a copyright violation if you're basically figuring out how something works and what it does and creating it or recreating it? Well, so here, here's, here's my thought about this is, you know, the way to get around those things is your clean room develops some or reverse engineering something, right? Mm-hmm.

You take the team that's gonna go take what we want, we know what it is 'cause it's a product or an, uh, something that's been protected copyrighted and say, go do that. We're not gonna let you be influenced by anything else other than you developing this. Isn't that what AI is doing?

Now you can say it's being influenced by other things, but I don't have control over what it's influencing. It, it's development. I don't know what cloud resources, cloud codes gonna use, um, you know, from sonnet or whatever model that it's using to develop from.

But, so I'm kind of outsourcing that transfer. So me, you know, it's not me. I didn't do anything.

It's somebody else's model that did all of that work. And if go sue the LLM, you know, that kind of thing is that is the responsibility because you instructed it to do so. Well, yes.

I I was the, you know, that's a conspiracy. Was I conspiracy to commit fraud? I guess.

So I, I we're gonna see a lot of this happen. I think where this is gonna explode into everything's being rebuilt. I even saw, um, a gentleman named Patrick Dub Dubo, I dunno if you, Patrick, he is one of the originators of the DevOps movement.

I think he actually coined the term and he's very much into AI native development. And part of this group that's focused on that, he posted about this, that I'm kind of scaring myself. 'cause I'm running into situations where instead of going, getting an open source library and finding the right one and using it or whatever, I just, I just write it again.

I just have the ai, write it again, don't worry about it. So suddenly reuse is not a big deal. Right.

Why do I need to reuse code if I can just create what I want? And if it does the same thing, why do I care? 'cause it's gonna maintain it.

Well, right. I'm being flippant about it. But that's essentially where we're headed.

I, I could basically vibe code Pandas in a weekend, I'm sure. Yeah, yeah. And have the basic functionality to, to handle data frames.

Uh, but should I do that? Um, you know, I I think that, you know, and this is something you see with open source in general, is that, um, somebody gives of their time and expertise and their energy to create something, whether that is a re-envisioning of something or something completely new. And if people appreciate it and like it, or they want to support it, or him, they get involved in one of many ways and, and therefore contribute to it, and it grows as a project and you end up with Pandas, um, which is, you know, a beloved, you know, tool that, that widely used force everywhere, everyone's using it.

Um mm-hmm. But it, it is, it is very much the case that, um, it, it know with that without, without somebody who first said, wow, you know, I need a way to manage tabular data in Python. Um, does it exist?

It didn't at one point. Right. It didn't.

And so they made it. Not everything is like that though. Um, you know, if you think about like Linux and the core utilities, the, the new core utilities and how, you know, people will swap out LS for easy a for instance to, you know, because it's better.

You know, somebody came along and said, you know what, I, I think I can, I can do one up on that. Let's rewrite it in go, because go is a great CLI language and, uh, right, right. Or, you know, little larger projects, let's like z you know, I hate Electron vs code, and it's not just me talking, although, you know, how much hate the universe.

You're speaking for the universe. I am, but, but it's me right now. Uh, and, and said, let's, let's do it in rust and, and make something super fast.

Um, you know, is we can manage on our own. And I see that more and more everywhere I go on, on the interwebs. You know, you go down, uh, any, you know, forum that's technical, uh, on Reddit, and you will see post after post of people saying, well, I, I got tired of not being able to do X, so I built something, or I wrote this.

And they'll toss out their GitHub, URL that has their, their code base with, you know, an MIT or whatever license that they, they put on, or most do creative, believe it or not. Uh, but at any rate mm-hmm. Um, you know, it's, it's a very different world.

And like, well, you know, and this, this is, we should do it an episode just on this. Um, you know, what, what hap you know, I, I talk about systems change when constraints change. Mm.

Okay. Now, the constraint of who can build code and what it takes to build and how much you can build, obviously is shifting. Um, you Yeah.

Especially when you lift a constraint out of a system. So the, the, the question is, where are we headed with this? Because I think we're going from a world where developers are valued because there are the few who can create code, but when code can be created easily, and, and in large volumes of it, it's less about write the writing of it and it's more about the architecting and designing and instructing how it gets built.

Right. Whether you, it's sort of like, um, Edison, Edison didn't do all that work himself. Right?

He had a team of engineers and whatever folks that are doing all this, and he claimed all the patents for it. It's kind of what we're headed to. What a developer is, is, hey, yeah, I don't like that go library.

And I've never liked it. Just, just do what you need. Create create what I want.

Right. My version of it. Yep.

And so the amount of code we're gonna have is just gonna explode. And what the value it is, well, maybe I don't want a licensure. I'm just gonna clean room it and do my own thing.

And, uh, yeah. You, if you wanna start a lawsuit, go ahead. But, you know, lawyer lawyers will be happy.

But it, it does create a world where the code isn't what's valuable. It's the people who construct, which opens it up. Oh, I very much agree.

More people. Yep. Yeah.

It's, it's, it's an interesting proposition. Yeah. When DHH makes something, you know, people are like, oh, I like that guy.

You know, if, if the fellow behind HashiCorp makes something, people are like, yeah, I like that guy. It's the energy and, and the, the thoughts and the ideas that, that, you know, revolve around these open source projects that are the real, the real value to me, I think it's the renaissance of software. You know, it's the entrepreneurial renaissance of software where entrepreneurs really thrive in this environment.

Right. Because now the world, your oysters, they say of what you can do. And you know, I'm, I, they're not, that's a big company and that company's not gonna license to me 'cause I'm whoever, I'm puny whatever right now, but I will be great someday.

And here's how I'm gonna be great by doing it this way. So it, it, I think, you know, everybody talks about developers, well, they did talk about developers not having jobs anymore, and suddenly we build, we're building tools for developers. Um, but I think, you know, my, my theme for 2026 is that developers become the engineers of agents creating software.

That's what, that's the world we're, we're creating right now. And we're, we can argue whether if we moved into that, certainly not fully, but that's, that's what we're heading towards and that that shifts the dynamic and the value. So I guess we just had the podcast episode right there.

We don't even need the podcast. No, we do. We need the podcast and, and you know, we need the creativity and, and, uh, I, I like that analogy.

Mine is, um, the architect, you know, that builds a skyscraper. It isn't just one guy on a table drawing everything out all the way down to every nut and bolt. Mm-hmm.

It's a sort of idea to fruition, you know? And, you know, to create something requires many different skills at many different levels, uh, of coordination and responsibility. And so, like you say, if, if you can have that funnel, you know, shortened where it's, it's like almost zero, but still have, you know, the idea person at the head of it, starting with the, I've got an idea.

I want something that does structured data and python. You know, what if pandas had been made in a weekend, it could be, it can now, right. Probably already has actually it can, you can feed, feed it to and do an LLM say, create the requirements for this for me to create my own version now tweak it how I want it, and now go build it.

Right. Essentially reverse engineer that way. Um, well, well one things I'd love to, um, I think you have, uh, the mantra of the, is it the data Shepherd the Shepherd?

Yes. The analogy you're using the engineer to Data Shepherd. Yes.

The AI Shepherd, I should say. Yeah. Shepherd.

Say, say a little bit about that and then maybe we'll do an episode on both of those themes. Yeah. Yeah.

So, you know, we, we've seen over the last year that, uh, you know, through Gen AI in particular, that, um, the tasks that the data professionals do on a typical basis, you know, on a daily basis, you know, modeling data, importing data, managing, you know, Knolls and, uh, imputing, you know, missing bits and all of those TA tedious tasks mm-hmm. Are increasingly getting automated. Um mm-hmm.

And not just to automate them in terms of, you know, here's something we already know how to do. Now just do it, do that. That's, that's just automation programmatically, like you could just write a script to do it.

What I'm talking about is understanding the data professional's intent and then executing on it, you know, ala bot, you know, to to, to just say, I, I want to incorporate a new data source and to create a data product on the back of that that serves this customer with this use case and need, help me spec out how to do that and construct it. And so it is the exact same trajectory that you're talking about with software engineering in general that we've just been talking about. Mm-hmm.

It's just playing out in a, in a specific domain where mm-hmm. Your data engineers are no longer doing con you know, do doing syntax instead they're doing intent, you know, trying to whoever can, uh, articulate and describe and then shepherd, uh, that description from idea to fruition is going to succeed, uh, in 2026 and going forward, you know, and, and so it's, that's a cool metaphor. Cool model y.

Yeah. I, I I I, I feel like it's, you know, I want to have sheep dogs that, that keep the sheep, you know, going where I want them. And I, I don't want to run after every sheep.

Exactly. I don't want, all my sub agents need to have somebody watching over them, in other words. Yeah.

The, the, the colleague Border Collie that's rounding all those up doesn't know, well, it might know what you want 'cause it's learned what from what you instructed it to do, and you have to tell it every, you may not even know why you're doing it. That's kind of the analogy of we're gonna, the shears, we're gonna go get sheared. He doesn't know that.

Yep. Exactly. Well, we've, we've had, uh, a pretty wide ranging, but a thematical Right.

Podcast episode here about, you know, you can do anything, put your mind to it now with these tools, and that's just gonna increase the level of complexity and breadth of what you can do. And the is like the attack surface. It's like the attack surface.

Yes. Yeah. Be careful what you do to yourself along the way.

You might engineer a bunch of wolves to go after your sheep at the same time, hired wolves accidentally open season, to use the analogy. Um, well, like we, we've got a long time. So maybe just a quick version of the, of our next last segment, which is the drop.

Okay. It's time for the drop. It's, it's alright.

So, so real quick for mine for the drop is I just dropped the new, um, edition of the dataset for Software Lifecycle Engineering, which encompasses observability, encompass software testing, software engineering, uh, DevOps. Platform engineering is a wide, wide ranging, um, data set with some really good information. And the number of people that are investing in using AI tools and development has exploded.

It's over 90%, uh, between, I'm already using it, I'm using it for everything too. I'm evaluating it. And, uh, that's how rapidly we're changing.

So that's just one of the data points that's, that's in it through the future of intelligence platform. So how about you Brent? Yeah.

Similarly, uh, just dropped. Um, it's like, you know, the, uh, mass drop, if you remember, I think they're still going actually to, to have projects that are self-funded. I'm just, I'm getting ready to drop the one H 2026 data intelligence, analytics and infrastructure market sizing and five year forecast.

Boy, here's a mouthful. So we just, and, um, it, it's, you know, uh, something you can get a lot of that data on our public site, but of course we, we keep the majority of it for our, for our paying customers because we do, uh, need to make money to, to keep doing what we do. And, uh, just, you know, give give you guys one, one stat for the overall market, which I look at pretty much, you know, the entire spectrum of data acquisition to, uh, AI outcome.

9%, which is up from where we were last time we did this. 2 trillion by 2031 for this, this little market that I, I like to call home. Wow.

Yeah. C'S over 10, you're talking, that gets people's attention, right? I got 16 and above.

Wow. That's phenomenal. So it's, it's amazing.

You can claim a, there's an AI bubble, but I think the AI bubble is creating new bubbles, new not, and not the bubbles that pop, but new growth areas. But even if they do pop, they, they create value. I mean, you know, we, we would be kidding ourselves if we didn't say that, you know, um, the, the idea of capitalism is, is predicated upon disruption.

Um mm-hmm. And Oh, absolutely. You know, a bubble is nothing, if not disruptive.

Yeah. My, my, my entrepreneurial job is to pop your bubble while I create mine when somebody gets after mine for sure. Right.

Without getting any stuff. So exciting. com slash mitch Ashley, Brad, uh, Shiman.

It's all Brad Shiman. I'm a, I'm a one Worder. Oh, is it?

Yeah. Oh, wow. You, you've progressed.

I haven't evolved that far, so I'm still a dash. You're hyphenated. Yes.

I'm, and there's all kinds of great stuff we're producing, you know, some really, I think awesome content across the board with all the analysts and your stuff's phenomenal. So thank you Brad, for the work that you do, and thank you. I'm not hyphenated actually, I'm sorry I'm hyphenated.

I just really, you're hyphenated. Okay. Well, okay.

We both have some work to evolve to the next stage of evolution, I guess. Future. Alright everybody, thanks for tuning in.

Thanks for liking, thanks for following. Please share the podcast with your friends. You have ideas for topics I, uh, guests that we might have.

And we're kind of getting some, getting some things laid down here about what the podcast is about. And then we'll be adding some guests, uh, to our, and it can be practitioners, can be, uh, thought leaders in the industry, vendors, et cetera. We'll try to bring a mix of really interesting, uh, conversations and topics.

So thanks for tuning in on, on behalf of myself, Brad, and our illustrious podcast engineer, Corey, thanks for joining us today. 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.