https://www.listennotes.com/e/d9029e5660f547c685ebdb2c1818f7df

Get ready to discover some of the most actionable DevOps techniques and tooling, including performance and reliability for some of the world's smartest engineers. Hey, I'm Joe Colantonio, host of the DevOps Toolchain podcast, and my goal is to help you create DevOps toolchain awesomeness. Hey, what's the AI factory? Well, that's what you're gonna find out today because we have an awesome expert joining us. We have Maurice, who knows all about the AI Factory. It's part of this company called AIA Systems. He's a senior AI engineer whose fingerprints can be found across two decades of startup success spanning from Silicon Valley to Los Angeles. He also has a rare blend of system thinking, product pragmatism and a deep software craftsmanship to the agentic AI revolution that we all find ourselves in. I went to his site, found a lot of interesting things. I have a lot of questions, don't know where we're going to go, but it's going to be DevOps testing focused. You don't want to miss it. Check it out. Hey, before we get into this episode, I want to quickly talk about the silent killer of most DevOps efforts, that is poor user experience. If your app is slow, it's worse than your typical bug. It's frustrating and in my experience and many others I talk to on this podcast, frustrated users don't last long. But since slow performance is as sudden, it's hard for standard error monitoring tools to catch. And that's why I really dig smart Be Insight Hub. It's an all in one observability solution that offers front end performance monitoring and distributed tracing. Your developers can easily detect, fix and prevent performance bottlenecks before it affects your users. Sounds cool, right? Don't rely anymore on frustrated user feedback. But I always say try it for yourself. Go to smartbeard.com or use our special link down below and try it for free. No credit card required. Amaris, welcome to the Guild.

Hey Joe, thanks for having me on. Great to see you.

Great to see you as well. So, as I mentioned, you're an AI expert. What got you into AI? It seems like you've been developer for many, many years. When did you start dipping your toe in AI? Has it always been part of your background?

Well, I've been peripheral and actually working on AI for quite some time. Back in the late tens, I was working with a startup where we ended up filing a patent on an advertising platform for AI using more on the machine learning side of things as opposed to the Gen AI. That was obviously a long time before the Gen AI phenomenon started. But yeah, I've been interacting with AI for quite some time.

So is that a common misunderstanding you find? Is AI the same as machine learning? Does it matter? Is AI just an umbrella term nowadays?

Well, it's become sort of a marketing buzz term at this point. It's inevitably when you get this level of hype, terminology tends to lose its meaning from a technical perspective, it becomes more of a marketing term and then it becomes beyond that, it goes into the zeitgeist as meaning, you know, just whatever that general concept is of what AI is, it might be just sort of like a idea of a robot or a Terminator or whatever it might be. It sort of starts losing its meaning at some point. But yeah, I think it's from my perspective. Yeah, Gen I and AI and those kinds of terms have kind of blurred in their meaning.

Absolutely. So I know in the pre show you talked about having like a product background as well. Back in the day you'd have to get requirements from your users and do the whole, whole thing. Now it seems like you can go to like a tool like Manus or something and say, hey, research this, give me some requirements and bam. What do people need to know if they're building products nowadays? Is that. Have they lost that competitive edge of knowing their customers because AI knows the context of everything? Or like where do people go nowadays when they start with the product that they have an idea for?

Well, I mean the idea of product development or product origination or ideation has grant has shifted tremendously because of the Gen AI phenomenon. You can now and it will continue to change over time. I think the ultimate endpoint is going to be a little bit like. I don't know if anyone remembers Scotty talking to the computer in Star Trek. You just say there was a scene where he landed in on a version of Earth sometime in the past. And he picks up a mouse and he talks into the mouse like a microphone and he says, computer, you know, do this for me. And nothing happens. And he says, he's just looking around like what? What's wrong with the computer at this point? You can talk to your computer and you can tell it what you want to build and it will actually build it for you and it will accomplish a lot from what you say, it will interpret a lot from what you say and it may end up not doing exactly what you want because it's very, you know, if you're not initiated in product specifications and product requirements gathering and that kind of highly disciplined area that you might be going through in your career for, you know, spending 10 or 15, 20 years, you know, going through this product requirements process. Without that level of accuracy, the AI doesn't quite know how to get it right. So you end up saying no, that's not quite what I wanted. This is what I wanted. And you end up in this conversational loop with your AI and it will correct itself and it will ask for clarification and sometimes you could one shot it if you're lucky, if it's a fairly generic product description and it's built a lot of them before and it's in a well understood and well known category, it will generate so something like a to do list or something like that or a, a well known game engine or other kinds of like, like a Tetris or a Pong kind of game. It will, it will, it can one shot those kinds of things because they're fairly relatively low complexity and will relatively highly well understood but other than that when it gets more complex, it'll have a hard time understanding is that just.

Now I know things change rapidly. Do you see that even going further where it's going to become less and less where you have to keep prompting it to get what you want.

It's improving. The dialogue with the AI is getting better all the time. And I mean even the cases that I'm describing when you, these are low complexity but it's progressively being able to handle more and more complexity, more and more features. It takes on more of the backend as well. So like I was just on replit earlier today and I was able to develop an app that it builds out the backend, builds out all of the tables, it does the data modeling for you, then it creates the tables, creates the, stands up the, the postgres database, wherever it might happen to be, instantiates the tables and then populates the tables with test data all while it's building the front end ui. And then it will do the, you can assign it a virtual machine. You can say what size I want my virtual machine to be, how I want it to scale up or down during peak usage. And you can do authentication and authorization like a login pattern and, and you can build these really cool systems very, very quickly and deploy them and then you can slap your own domain on top of it. So it's not even deployed to a replit domain. It will go to your own domain. And this, this all takes like a half an hour and you're talking to your, you're talking to the Replit chatbot for, you know, either texting. Well, I would be texting. Or you could chat to it if you have the user voice recognition enabled on your Mac or what have you. And so it becomes a very natural kind of process. But you're sitting there most of the time just waiting for it to respond because it's doing a lot of work in the background.

Yeah, yeah. I don't know if I should be excited, which I am, but also nervous, I guess. I only have 10 years of work left in me, but I don't care. But if someone's newer to the industry, do you think they're gonna have a harder time getting a gig or what do they need to know? Because it seems, like you said, it seems to be all the heavy lifting, like what do they know what it's doing? Do they need to know?

Yeah, well, it kind of works in a funny kind of way in the sense that capital goes to where it gets. Get the. Both the best response, get the best return. And so if the cost of, you know, you look at this from an economic kind of point of view, when the cost of developing a product starts to fall as rapidly as it is now, then the demand for it increases dramatically. Right? It's supply and demand. So whereas formerly some kinds of products were not viable to develop because of the cost and the benefit, now these are much more viable. So it opens up a whole new space of creativity where all of those pet projects you've been noodling around in your head and not able to get off the ground, your unique perspective of the world, your insights from, let's say, 20 years or 40 years of interacting in the enterprise space or in the consumer space, B2B or B2C, now all of these projects are potentially viable. So you can fire off. It's a great time to start a new business. It's a great time to build a new app. And then people who aren't even very well initiated in, in programming can, you know, in some senses some, some skill sets will become a little less valuable, you could argue, but it doesn't. So I think really what. It really plays well for the subject matter expert in their industry. They can now kind of bypass a lot of that, you know, process of standing up a tech department and building out all of your developers. And for certain kinds of projects, that's viable. Inevitably, when a project takes off and it starts getting traction, you can't just keep talking to your AI to sort of scale it up. You're going to need much more highly specialized skill sets that like in the DevOps space or the testing space, those skill sets are still going to be pretty viable. Good to know going forward for the projects that are scaling.

Right, right, right.

And the ones that are big are the ones that, where you need the DevOps and that level of testing.

Right, right. So how did DevOps handle this then? Is it just like any other application they've been dealing with? Is there anything special they need to be aware of before they deploy it? Or is it, can you just tell this, to have deploy this now to production and not even worry about a human being? How do you, how do you do that with the DevOps now?

Well, you know, the, the AI will understand some of the basics of DevOps and some of it will be kind of a little bit hard coded like, like if you take the PLIT as an example, they have their own methodology for, for doing, setting up their scaling. They've done their DevOps essentially in the background and they built the AI on top of their DevOps API, let's say. So they have the ability to stand up these VMs and do scalability, probably using Kubernetes, but you know, they're not revealing that on their ui. And whatever scalable platform they're using, they provide a UI to you so they can say, hey, you want this many CPUs, you want this much RAM and your database is this size. I think they're using NEON on the back end, NEON postgres databases on the back end. So that's just an example. So the DevOps had to be done, a lot of DevOps had to be done on the back end to make this front end work, to make this AI agent work, AI co developing agent work. And there's a tremendous amount of value to Ripplet to be able to say, we will help you develop your code, but we will also help you deploy it and get it up and running and we'll scale it. That's a big sort of advertising, big marketing plus for them. That generates a lot of new subscriptions for them. So they really do depend heavily on DevOps on the back end to make that work. So you know, you can imagine that couldn't have been, that was not really very feasible as a business model a couple of years ago. Now Replit is kind of leading that. Well, they're one of the leaders in that development space. There's other people coming out, like Claude is coming out with their Claude code. Obviously you mentioned Manus and there's a cursor and Free. What is that company called? Code Surfer or something like that?

Yeah, Windsurfer or something.

Oh, Windsurf. That's it, yeah, Windsurf.

Windsurf, yeah.

I actually use it. I keep calling it Free Wins for some reason.

It's called winsurf Start one called Freewinds. It's a good one.

Yeah. Maybe that'll be a new one. And Windsurf just got bought by open API. I mean, open. Yeah, open API. OpenAI, I should say. And for 3 billion. And cursor is on the IS market cap, I think is around 7 billion. So these guys are going bank gangbusters. And an interesting thing, the just to prognosticate again about where AI is going, which is kind of like the theme of what we're talking about right now, the Claude Code tool. One of the interesting things that I was. I was looking at their presentation and one of the questions in the Q and A at the end, one of the people asked, why didn't you make it an ide? Because it's a command line thing. And they said, well, we've seen in house the way our LLM foundational model has been progressing, because they're one of the leaders in the code development space for their LLM. And they're saying in the very near future, it's not worth their time trying to compete with Cursor or Windsurf, because they see that this stuff is advancing so fast that it is almost at the point where you just talk to it. And you don't need an ide, you don't need to even look at the code. You just tell it what's working, what's not working. And then it comes down really to testing. That's really where it comes down to where the real benefit, the real added value in the skill sets of the people who are developing this, who are moving into this new, brave new world where everything is automated, all the code development is automated, comes down to testing. So you have to ask it how you want it to test it, how do you validate the product? And you have to. Again, you can do that through prompting. And the AI will try to stand up a, you know, a test scenario and run it through its paces, either by using a smart browser or using an API or a combination of both. So the smart browsers, you know, Playwright is kind of like the top of the browsers and it runs, you know, it can run scripts, but on top of that, you can have an AI browser that sits on top of Playwright and. Or an AI agent That sits on top of Playwright, which also sits on top of the browser. And you can control Playwright to manipulate the browser to exercise the front end just through AI dialogue. And you give it a test scenario. And again it comes down to product requirements gathering and test scenario gathering. And you basically pass that document to the AI agent and it will run the API and the browser testing. So it'll do the front end and the back end testing for you.

How does that work? Is the agent like a wrapper almost, where it understands the commands of Playwright, so you don't need to know the guts of Playwright. You're just like, I want to test a browser log in. It goes, okay, I'm going to use Playwright. And it knows all the APIs to call playwright and initiate. How does that work?

Yeah, it's basically lines, what you described. If Playwright can manipulate the browser, then it's a matter of what do you build on top of it to manipulate Playwright. And traditionally, you would just have to write scripts. And now instead of writing those scripts manually, those scripts get written automatically by the agent. So the agent is smart enough, instead of doing code generation, it's doing test scenario script generation for Playwright. And so you tell it what you want to test. It looks at the source code, it looks at the way it was deployed, the APIs, and the UI. It understands conceptually, it has to go in and understand conceptually how the UI works. And so it might. It might be building what you call them directed to cyclical graphs or a graph network of how to navigate through the UI so that it can get to the correct screens to do the correct kind of testing. And it also needs to be smart enough to know how to log in and that kind of thing. So that's an idea of the AI that's under the general category of the AI smart browser. And you can use that for testing, but you can also use that for actual integrate interacting with existing SaaS platforms. So there's like a million SaaS platforms out there that they all run, you know, they all make available a UI and people interact with those UIs all day long. And now you can overlay a smart browser or point a smart browser at it. So you might have a smart browser just sitting out on the Internet. Just like google.com, you go to smart browser.com and you just talk to it and say, hey, I want you to log into my SaaS app. And it could be Salesforce or it could be any other CRM or any other kind of product out there and say, I Want to. I want to run a campaign or I want to do a data entry on a lead I just got from a phone call I just made or, you know, any kind of scenario you can imagine. You can say. You can just tell it to go run a whole campaign for you and get back to me when it's finished, or something like that. It could be a few days later and say, how did it go? Or you can say, hey, monitor it for me and let me know, give me updates on. So it's literally how you would hire a person to do that kind of work. Now a smart browser can go in there and start doing that for you.

So is that a real thing, smart browser, or is that just the name of an agent?

That's just a category.

A category.

There's a couple of brand names out there, but it's such a new category that it hasn't really been. There's. Nobody has kind of surfaced. Nobody has gotten that far down the path, I should say. So I'm kind of prognosticating a little bit about the future in that regard.

Right, right. Well, I like it because I use a lot of applications that don't have an API and I need to physically log in and look for stats. Like something that has nothing to do with testing or I have to log into my podcast, get a number of stats, report to the sponsor, get YouTube views, get LinkedIn, post likes, and everything. Hard to do. I can't imagine how AI would do it, but it sounds like at some point I'd have a smart browser. It's almost like rpa. We know how to log in and we know how to grab all that without having to have an API.

Yeah. So it's like a little different from the playwright scenario where it's writing scripts. It's writing a script, I should say, for a discrete test scenario that you described that it wrote the code for. In this other scenario, it's not writing code. It's now just pressing buttons based on. So it's getting a conceptual understanding of the app. Whether it's Salesforce or. Or any other app, it has to have a kind of a map of the entire. So there's a lot of setup involved. You can't just point it at everything. There's no AGI solution for this. You have to give it a lot of context before it knows how to operate. But if you give it the full context of the Salesforce UI and it knows all of the screens and knows all the buttons to get to all of the screens, and it Knows all of the data entry boxes and all the submit buttons and all of the, the workflows that can be overlaid on top of that. It gets progressively smarter and smarter. Sometimes at the beginning you will probably just have to train it on each of the workflows. Like the workflow you just described going into your, your UI and asking for some stats and asking to submit to the sponsor and that kind of thing. That might be a workflow that you might have to just sort of program in at the beginning before it gets this smart where it can just figure it out on its own. But it will get smart enough once it knows the ui, it will get smart enough to say, oh yeah, Joe just wants to do the usual wrap up or the usual, you know, thing that he does at the start of a podcast or what have you and, or some daily maintenance on something. I know what he means and I'll just run it through the UI for him and give him the result. It's getting to that point, but it's not there yet.

Very cool. So what's also interesting about you is I think you mentioned this earlier, earlier, but your company did a lot with sound.

Yeah, voice.

So yeah, how do you test that? I always thought, okay, I can understand testing code. How do you test a voice to make sure that it's, how do you test what you're creating, your genic AIs that use voice?

I guess, yeah, it's the same thing. Well, because voice gets converted to text. That's the first thing it does. It does voice to text and then we do text speech. So everything ends up being text. So we have to rely. That's the ML part of AI is the text to speech and speech to text portion. Those have been around for a long time. But it's the LLM that takes the text and converts it to other text and then it gets outputted, converted back into voice. So we don't do any testing on the voice component at all because those are, it's either good at doing it or not good at doing it. You know, we can, we can go to Elevenlabs or Cartesia and add these different personality types for the voice and we can program it into certain kind of, you know, program the LLM just to adopt various styles. It could be assertive, it could be friendly, it could be sales oriented and you know, making pitches or it could be something that's doing a lot of demonstrating a lot of empathy. If it's like a, a therapist type of a role and. But those are not really tested that as at the applet level would. They would not be tested. They would be tested at the, you know, whoever's generating the ML voice would be doing that level of testing. They take the actors in and they do the, they capture the voice and they clone it and they might have different speaking styles and that kind of thing. Those are, I would imagine those are fairly manual, manually oriented tests. But in the, in the app layer, it's just back to the text to the LLM and text back from the LLM. As far as how you test that, that's. You would need another LLM to test that LLM to see if it's hitting the right, the right points. So it's like a, it's kind of a fairly new area of testing where you know, you, but you can, you can do LLM testing for your app and say, well, is it giving me the right responses? But the real way you do that kind of testing is usually you have an app that runs many, many workflow, the same workflow many, many times. And you, you then do a post analysis on the each run of the workflow and you look for patterns and you say, is this hitting my KPIs or is it giving consistent results or is it hallucinating over time? And then you try to identify, you know, you can fine tune some of the parameters that go to the LLM because you can fine tune higher and lower hallucination levels on the LLM. So that's kind of the more reliable way. It's at the app level over time, with repeated workflows that you can find these patterns emerging and you can say, is it reaching its goals like its KPIs, or is it drifting off or. And how can it be fine tuned and adjusted to get it back on track? So that's part of what we do with our workflows that we automate is we. It's in the general category of evaluation. So if you should be designing your product with a goal to be able to conduct evaluation on it, if you're using LLMs because you want to make sure that you're monitoring it correctly and each individual workflow you can do an evaluation on and say, did it reach, did it do what it was supposed to do or was there a certain amount of hallucination or did it make a mistake, some of it can be because of the original design of the workflow that may just need to be improved, and some of it is because the nature of the work that you're trying to do is subject to hallucination, it's not very well defined. Then you sort of got to think in terms of agentic systems. So instead of just having one LLM agent, you have a, what's called a sort of a gateway keeper who redirects the call to the, to other sub agents. And each of these sub agents are much more focused and they're less subject to hallucination. So if it's like a sales pitch agent or if it's a sales rep agent, you might have a specialty agent that just works on building rapport and then another specialty agent that just works on building pre qualifying the person who you're talking to so that they are the decision maker or try to redirect to get to the decision maker. And then you have a pitch agent that just focuses on doing the pitch. And then you have the foreclosing agent who just focuses on what's the outcome of everything. Did we close the deal? Did we run the credit card? Do we set the appointment? And each of those agents are less subject to hallucination as opposed to one agent that does all of that. So a lot of it comes down to the design.

So it sounds almost like in the DevOps world it's more like monitoring and production because over time it could start to degrade almost and be able to monitor these KPIs for drift and hallucination to make sure that you have a human check in and say, how's it doing over time? Almost.

Yeah. Well, from a DevOps perspective exclusively, you'd probably be looking for more like response times and scalability. From the testing perspective, yes, you would be looking for that kind of thing. And if your product is scaling up and it's getting a lot of traffic, then these evaluation loops become more and more critical because you don't know if the lag time. Because one thing that a voice, AI, AI voice is really subject to is latency. And if the latency, like an acceptable latency, would have to be under 200 and some of these voices are these because the text to speech and the speech to text, they have their latency and then the LLM has its latency. And then if you're using these, this crew of agents, then they all have potential latency involved as well, though you would typically just use one on each call. And you have an orchestrator, that was the one I was looking for earlier, orchestrator that decides which sub agent to call. And they all have their inherent latencies. And if you're not doing your DevOps, if you're not Doing your infrastructure allocation correctly and you're not allowing for scalability, you're going to run into increasing latency. But, and latency is a big, is a kind of a deal breaker for voice. So you have to be very sensitive and monitor it. And you can gather statistics or be monitoring certain statistics on latency into your system monitors into your monitoring system. And you can set alarms when that latency goes down too low or too high, I should say. And then you can start triggering. Then your DevOps would kick in and it would start deploying new servers and then you'll hopefully be able to load balance across the new VMs that you stand up. And hopefully that would help in a DevOps perspective. But yeah, there's a lot of DevOps and testing involved in keeping these things up and running, and they do change. It's a dynamic environment.

So did your company do a lot of performance testing then to test for these latencies before it occurs or to make sure it's scaling gracefully?

Yeah, well, we try to. We designed like our text to speech and speech to text models that we use are optimized for CPUs, but as we scale, we would try to optimize them for GPUs and then we would throw more GPU hardware added because we're hosting our own hardware for those two components and we're also hosting our own hardware for the workers and the agents that we design. So the custom agents that we're building, we're hosting our own hardware on, but we're outsourcing. We're obviously not hosting our own foundational models. We have to call the API for those and that goes out to OpenAI or Gemini or whoever else anthropic or who have you. And they're doing all of their own load balancing and response time to take care of latency and all that. So we have to do that ourselves for our own agents. And we're building out a suite of our own custom agents that talk to the voice platform that we're building. And it works over the web, it works on. We can integrate with Gemini's multimodal LLMs and so we can do voice and text and video and images. So we're multimodal in that sense. If we're working depending on what LLM we're working with. So that'll work on a web app scenario. And then if we, if we're not doing multimodal, we can just do voice and text over a phone line, mostly voice. So you'd have like a, you know, you Just use your mobile device and you would just make phone calls and you can talk to the agent. You typically wouldn't text to the agent if you're talking to them already. And then you just, you know that that agent, we would be hosting that agent and we would be managing the voice text to speech and speech to text on our hardware. So we need to scale that and manage that from a DevOps and testing perspective.

Nice. So Maurice, a little background with the AIA systems, the AI factory. What does your company do exactly? If someone's listening and like, oh, this sounds great, but maybe I need a little assistance. Like do you help with that or is it do a completely separate business? What are you doing there?

Well, we, we're kind of set up to work with subject matter experts in the industry, in their industry. So if you're like a psychiatrist and you know your business really well and you want to expand across multiple clinics, we'll help you automate your onboarding of your patients. If you're running a drug trial or something like that, we'll help you with automate that drug trial. Anything that has to do with unstructured data, basically like a therapist's business is unstructured. They just, it's all about talking. And so it's a good fit for what we're doing. And what we do is we like to talk to these subject matter experts in their fields because we're not the experts in their field. We have to be. They have to tell us what it is that they are really expert in. And, but, and that involves unstructured data because they know that things that couldn't be automated a year or two ago can now be automated. But it has to fit into the right category, which is unstructured data, essentially. And it has to be reasonably easily automated using the Genai tech stack. So a lot of that is just voice or text. Then we integrate in with the CRMs of the world and the backend databases of the world and we create these new workflows and that's a huge opportunity space. Right now the deal flow that's coming from that space is pretty enormous. Most of the, you could say fairly safely that every enterprise of any size that wants to adapt to this new world allocated budgets for this. And now they're just looking for the right tool sets, the right expertise and the right people to talk to and they send their subject matter experts there, we engage with them and then we figure out what the best way is for them to move forward with their, with their new workflows. And we put it on our platform and we're highly incentivized to work with them for the very, very much into the long term because we're, you know, we're, we're kind of attached at the hip since we're managing the part of the platforms for them as well. But we could do in house versions of the platform as well if they're, if they're concerned about that.

Any lessons learned working with subject matter experts say someone's working in a company and they in house have to do this and they're like, I'm not a radiologist but my, I'm creating software for radiologists. How do I create AI, a workflow that will meet the real demands of the real user? How do I capture that correctly?

Well, let's say like a radiologist is one of the earlier sort of careers to be subject to this AI phenomenon and they were more in the ML space where it's machine learning. These are models that look for patterns in the actual X rays or the scans and so on and can spot things that radiologists can spot. So that automation's been around for a while. But if you then want to add a gen AI layer on top of that, you would, you know, you would automate the onboarding, you would automate the scheduling, you would automate. These are things that you would do that are fairly generic to a clinic kind of thing. So if he was, if he was running, if the radiologist had his own clinic or you know, they oftentimes they have their own multiple branch companies where you have these giant CAT scans and X ray machines and, and they're scattered all over a city. Companies like that, we can help them automate that entire process. But yeah, just because they're a radiologist happens that that's not a great fit because they've already been automated, they've been knocked down already in a sense, but their function is still, you know, there's plenty of radiologists around and the fears of, you know, TV kill the radio store or whatever that song is.

Radio star.

Yeah, yeah, yeah. That phenomenon never happened. Radio is still around.

Yeah, yeah. And everyone says, oh, everyone's going to be out of a job. But it sounds like I can't imagine a doctor or a chiropractor or whatever, therapists creating their own AI. Maybe I'm wrong creating their own AI to do all this. Am I wrong? Even if it gets really good, I don't know, it seems far fetched to me.

Yeah, I think there's definitely the copilot phenomenon where it can work in companionship with a therapist. And I think it's just to me, you know, I would say that it's just a matter of time before it becomes easier and easier to conduct the actual therapist therapy session itself because it's unstructured data. And a lot of the regulators, like the governing bodies of these kinds of skills, they have guidelines on how these therapy sessions should be conducted. And that's kind of like those guidelines can be automated because they're fairly strict and well structured guidelines. So how to conduct a therapy session in order to make an insurance claim is fairly structured. Well structured is probably an overused term here because we're looking for unstructured data. But the actual, it's a decision tree. It's kind of like a script, if you will, branching script. And things are being scored based on the answers that the patient would give. And that's, that's not something you could automate without having a genai component. But now that we have a genai component that, that can now be automated. So these prescriptions, these pre prescribed therapy formulas have been knocking around for a long time. But it's not up until now that they can be automated. And so yeah, it is. To answer your question, yeah, it is possible that a lot of therapists could be, could become super productive, you could say, until they become no longer necessary at some point. Right. You know, but that's probably too harsh. I'm reluctant to use that kind of terminology, but it's kind of a way of saying that's the logical progression that's happening, whether it will get there or not. I have sincere doubts that we're not going to get to that endpoint. But it's just that we're in that logical sort of progression to that endpoint. But it's the investors and the people who are building these kinds of platforms, they love to say this because it draws in more investment and it allows them to build out easier. You know, they're building the giant data centers and all that kind of thing. A lot of it is hype.

Oh right.

It just remains to be seen. Like, like we just don't. It's really hard to tell. My suspicion is, is that there will be fewer, there will be a need for fewer therapists, let's say, and other skills and other more creative endeavors will, will become viable. Whereas before they weren't viable. Just like when we moved out of the farming era into the industrialization era, we had no idea that all of these new kinds of jobs would become viable. Even before that, when we, when we were hunters and gatherers, there was no need for a horse farrier or a, you know, a guy making horseshoes or a guy just baking bread. But now all of these discrete jobs became very viable. The same thing's going to happen when we move into the AI space. So there's like, I was just at a meeting on Monday where we were talking about what is the future going to look like and how can we build a more empathetic world. I think the future is going to be one of universal wealth. So this idea of a universal basic income is probably not going to happen the way we think it will. Probably going to be universally wealthy and people won't need to have jobs anymore. So then it comes down to what do we want to do with all of our time? And you know, that's where we get into the Star Trek future again where, you know, we've got the, this egalitarian, peace loving, you know, space exploring race or civilization that just goes out and explores. Just exploring might be good enough and you know, making contact with other civilizations and what have you, all of that stuff. It just sounds crazy now, but we don't know where the future will take us. So we can have people that are really, they might be really good at building next generation spacecraft, which wasn't viable before. People that want to go out and explore new places. We have the ocean that hasn't been explored properly. We have to take care of our environment. There's all sorts of things that we weren't capable of doing as a civilization that now we will become eminently more capable of doing. Looking into the future.

Will we do it? Probably not, but we'll see, hopefully. Fingers crossed.

Yeah, yeah.

If she says anything. Utopias never end well, but we'll see.

Well, it's not a. Well, hopefully it's not a utopian ideology, but it's an optimistic view of the future.

Right, right. Awesome. All right, Maurice, before we go, is there one piece of actual advice you could give to someone to help with their AI efforts and what's the best way to find or contact you?

Yeah, the best thing is to do is to, you know, keep up to date on like I like to watch the all in podcast and they're really good at prognosticating about the future. They do delve into politics a little bit, but they're very futuristically oriented. I like to watch those guys and stay up to date in general on AI trends. I'm on X a lot and I just consumer a doom scroller on X sometimes and I find a lot of useful information on there as well. Tons of useful information. Yeah, you can find me Morris M A U R I C E a I a system.com and I'd love to hear back from anyone in the DevOps and testing and other areas that want to talk to me about their workflows and how we can help automate them and improve their the quality of their outcomes. Whether it's better DevOps, better testing, better enterprise workflows, what have you, I'm there for that kind of thing.

We'll have a link for all that down below. All right, before we wrap it up, remember frustrated users quit apps. Don't rely on bad App store reviews. Use SmartBears inside Hub to catch, fix and prevent performance bottlenecks and crashes from affecting your users. Go to smartbear.com or use the link down below and try for free for 14 days. No credit card required. And for links of everything of value we covered in this DevOps Toolchain show, head on over to test guild.com forward/p192 so that's it for this episode of the DevOps Toolchain Show. I'm Joe. My mission is to help you succeed creating end to end full stack DevOps toolchain awesomeness. As always, test everything and keep the good. Cheers. Hey, thank you for tuning in. It's incredible to connect with close to 400,000 followers across all our platforms and over 40,000 email subscribers who are at the forefront of automation, testing and DevOps. If you haven't yet, join our vibrant community at Test Guild Test, where you become part of our elite circle driving innovation in software testing and automation. And if you're a tool provider or have a service looking to empower our guild with solutions that elevate skills and tackle real world challenges, we're excited to collaborate. Visit test guild.info to explore how we can create transformative experiences. Together. Let's push the boundaries of what we can achieve O the TESCAL Automation Testing Podcast with lutes and lyres, the bards.

Began their song A tune of knowledge.

A melody of code through the air.

It spread like wildfire through the land.

Guiding testers, showing us secrets to behold.