The Internet of Agents Is Being Wired Up Right Now — Are You Ready?

The chatbot era is winding down — and what's replacing it doesn't wait to be asked. In this episode of The Wired Garage with Pops, Pops and co-host Steele sit down with Matt Coatney, a technology leader operating at the intersection of enterprise AI and legal industry practice. Matt breaks down the real difference between a chatbot and an autonomous AI agent, shares what multi-agent systems actually look like in production today (not the sales pitch version), and offers a clear-eyed take on governance, accountability, and responsible adoption. From his own experiments building with Claude Code at home, to running AI workshops inside a major law firm, to advising on where to move fast and where to pump the brakes — this conversation is grounded, practical, and a little bit urgent. The Internet of Agents isn't a concept on a roadmap. It's being wired up right now, one workflow at a time.

KEY TAKEAWAYS
 - Agents act. Chatbots answer. A chatbot waits for your question. An agent has knowledge, skills, guardrails, and can be proactive — more like a coworker than an advisor.
 - Multi-agent systems are real, but still maturing. Most enterprise deployments today are the same capability wearing different hats. The leap to agents filling entire job roles — not just tasks — is where the real shift happens.
 - Move fast on stable infrastructure — not on everything. Target high-repetition, high-cost, low-risk tasks first. In regulated environments (law, health, finance), some things in the value stream should never be automated, regardless of capability.
 - When an agent makes a mistake, accountability still sits with you. If you didn't set up the right guardrails, that's on the human who deployed the system — the same way a manager owns the outcomes of the people they supervise.
 - Governance for agents isn't new — it's just scaling fast. Test harnesses, simulation, failure mode analysis, escalation paths. The questions are the same ones any good manager asks. The challenge is applying them at speed and at scale.
 - Skill atrophy and over-reliance are real risks. After ninety-nine good AI outputs, you stop checking the hundredth. That's fine for low-stakes work — dangerous for skills that still matter when the tool goes down.
 - "AI powered" is a marketing claim, not a fact. Get the technologists in the room. The gap between vendors who've embraced AI in a mature way and those who've just applied the label is already showing up in product quality and stability.


KEYWORDS
AI agents, autonomous agents, multi-agent systems, internet of agents, AI governance, AI accountability, enterprise AI, AI adoption, AI in legal, IT leadership, AI vs chatbot, agentic AI, AI guardrails, AI risk, AI washing, ServiceNow AI, Claude Code, MCP protocol, AI productivity, future of IT, IT service desk AI, AI skill atrophy, AI in enterprise, responsible AI, tech leadership, future of work, AI tools, wired garage


If this episode got your gears turning, share it with someone on your team who needs to hear it — especially that person who's still convinced AI is just a fancier search engine.

Subscribe and leave us a review wherever you listen. Every rating helps the garage reach more people who are wiring it up.

Take Matt's 15-minute challenge: pick one task you hate, hand it to Claude or ChatGPT, and let it show you what an agent can actually do. Then come back and tell us what happened.

Connect with Matt Coatney on LinkedIn and follow the conversation as agentic AI keeps evolving. He's one of the most grounded voices in this space.

👍 Subscribe to The Wired Garage on Substack so you never miss a recap, deep-dive, or behind-the-scenes drop from the garage.

Transcript

SPEAKER_02 0:00

There is this sense that once you start to get really enamored with these tools and you see their capabilities to really just seed everything over to them. And I've had people who are in senior roles in regulated industries say, well, can I just point it to my email and have it just read all my emails and tell me what I need to focus on for today? I just want the punch list. And I'm like, you're going to seed that decision to a glorified search engine. And what happens if it misses that really important thing that a client demanded that that was due today? And oops, the I missed it. Back to a point about accountability. Like, can you just say, well, my agent messed up? So so I do think even if it's not a privacy or intrusiveness concern, you do need to be mindful of how much you let it turn loose and and let people sort of just have it run with things. Like that, there needs to be again that personal accountability and responsibility.

SPEAKER_00 1:55

But now we're going to take talk about taking that next step. We'll break down what's coming after the chatbot era. So fully autonomous AI agents that don't just answer questions, they take action. We'll talk about systems that plan, execute, and coordinate with other AI systems. No human in a loop for every decision that's made. And it sounds like science fiction, but until you look at what's already happening now in enterprise environments, you'll, if you want to work in IT, run a business, relate a team, this one's for you. So the Internet of Agents isn't a future concept. It's being wired up as we speak.

SPEAKER_01 2:28

So, Matt, welcome back. Thank you for joining us again. It's always a pleasure to have you. Uh, and for the record, I do think you are interesting. Before we get into all of all of the agents and the Internet of Agents, all the things that are going on. Where are you with AI personally? If you think about your day today, what are you actually using it for?

SPEAKER_02 2:50

Yeah, great, great question. Uh, I'm in it deep lately. I joke that about six months ago I wandered off into the wilderness with this thing called Claude Code and Opus, maybe it's not even that long ago. And uh I came back sort of espousing the wonders of it. No, I've I've been on both a personal and professional journey with it. On the personal side, um, I have been experimenting and and prototyping and things like that with clawed code, a lot of the harnesses, a lot of the new design skills. And what I've just been uh just so blown away with, and I'm a kid in the candy store with this stuff, is as a back-end algorithm and database guy for most of my early career, I could build pieces of an application, but never the whole thing. And I could write some JavaScript and some HTML, but you wouldn't want me to. And uh, so just from a pure satisfaction of having an idea and moving it through to creation and being able to sort of do that whole stack, even though I didn't have all the skills, is just super exhilarating. Now, is the code maintainable? Is it scalable? Is it production? Probably not. But from an idea, a concept to reality, at least from a prototyping perspective, it's been amazing. So I've been doing a lot on the personal side. I also use it a lot in personal productivity, bouncing ideas off, brainstorming, and all sorts of just different sort of fun stuff, right? Professionally speaking, I was a little slower to the game, to be honest, to integrate it. I was am a manager, so most of my day is spent in email and in decks and spreadsheets. And I would use it occasionally, but not as often as I should have. And then things changed. Claude came out with some of their plugins to Excel and PowerPoint, uh product that we use at my firm, integrated that harness as well. And all of a sudden, so I led an AI workshop internally for our administrative staff a couple of weeks ago. And it created the agenda, the PowerPoint, including notes and speaking talking points. It provided a prep sheet and it provided an a follow-on sort of working guide and next steps, all to brand, completely end-to-end produced with modest guidess from me. So that's when I'm like, oh, okay, that gets pretty interesting too. About productivity. Yes. Yeah. That's for me personally. But we can talk more when we get into it about some of the things that I see happening more broadly within teams and organizations. Yeah.

SPEAKER_00 5:10

So when when you heard we were talking about this one, we kind of called it the uh Here Comes the Internet of Agents episode. What was your first reaction? Did it excite you, scare you, or did you roll your eyes a little bit at it?

SPEAKER_02 5:23

A little bit of well, a little bit of the first two. I didn't roll my eyes, but uh my my first joking reaction was it sounded like one of those 1950s sci-fi movies. Yeah. And I was trying to picture like what the like the preview, the B-roll would look like. But no, I, you know, I've been talking about this concept of like everyday AI for too long now, a couple decades at this point, as something that was sort of coming into the future, always seemed around the corner. And uh with the advent of some of the technology we've seen just in the last six to 12 months, I think we're starting to get there. Uh so yeah, I'm I'm excited for it. Equal parts excited and trepidatious. I think short-term, very excited, medium to long term, lots of question marks, lots of uncertainty. I don't think that's all necessarily bad. What I'm most excited about, though, and I wonder if you feel the same way, is like this is how people at the in the middle of like the industrial revolution or the automobile revolution or flight or the computer revolution, like this is this kind of moment on steroids. And it's just a fast, it's a fascinating slash terry, terrifying time to be alive.

SPEAKER_00 6:28

Yeah, we we kind of talked about that in our last episode where it's it we've had these revolutions. This one's just exponentially larger, but each one has gotten that way, right? Maybe we've gotten larger impact and change and things like this. It's just, but it it it is another phase of our lives, and we just got to figure out how to how to live with it and live in it in a sense.

SPEAKER_01 6:50

So and hopefully scrape enough value for ourselves so that we're reaping the return on the productivity and and all that, those integration points. Uh because I'm there with you. I was very, very hesitant to just start open the floodgates and feed everything to it. I still have precautions and some boundaries, but just like you mentioned with those integrations with Excel, it it's in PowerPoint, etc., it's amazing what it can come up with.

SPEAKER_02 7:16

Just meets you where you work for sure. Yeah.

SPEAKER_01 7:19

So go ahead. So if we think about all the terms, all the vernacular that's out there that's thrown around. Kind of give me your take. What actually separates an AI agent from a chat bot or a regular bot? What makes it deserve a different category of tooling?

SPEAKER_02 7:37

Yeah. Yeah. I'll give you my thoughts and then also just sort of the feel or the perception of the experience when you're working with one of those tools versus a chat bot. So if you think about a chat bot, it's very much like having an advisor that's hanging around waiting for you to ask a question and you ask a question. It gives you some great advice. Maybe it gives you a tool or two to work with, but it's generally just sort of their waiting. It is responsive to you, but it's not much more than that. And the agent is is much, much more like a coworker or an employee or an assistant, right? Like it is, it's has a quote unquote mind of its own. It has knowledge that it can work from, it has skills, it has guardrails and instructions of what to do, what not to do. And it can be proactive, right? Depending on how you work with it. It can also be waiting in the wings for you to go do something, tell it to go do something. But it can also be like, okay, every I had a one of the experiments I ran at home was I had every morning, I had a clawed agent of sorts called a Mad OS, because I'm terrible at marketing, running in the background on my home computer. Every morning would wake up, it would look through sort of my prioritized task list and sort of loose ends and things like that. And it would send me a morning digest, what I should be thinking about prioritizing. I would chat with it throughout the day and it would synthesize that and keep that memory. So it was like having someone who really knew me, what I was focused on at that time, my priorities. It could keep me honest. It actually got a little too aggressive with being really like hyper accountable. Like, why aren't you doing this? Why aren't you this? You said you would do this. And I'm like, okay, tone that down. I'm a busy guy. I don't need all that pressure. So yes. So that's as I think about it, it's sort of the whole, you know, scaffolding or framework around that that turns it more into a doer than a than an advisor.

SPEAKER_00 9:25

It's kind of funny. I I use Claude right now more than I used to, and it'll ask me because sometimes is this a Pops podcast question, or is this work, or is this you cooking and fishing? You know, it kind of knows my personas now. So it kind of, you know, is saying, okay, how should I answer this, Brian? You know, and it's kind of cool. So when we talk about multi-agent and multi-agent systems, so AI talking to other AI to get something done. Where's that actually showing up right now in the real world? Not the pitch deck that you get from sales, the real world. Yeah, that is a great question.

SPEAKER_02 10:00

So I'll I'll sort of extrapolate on what I'm seeing in the real world and where I think it could head. Most of the I'll call it multi-agent systems that I've seen are really the same capability, just wearing different hats and sort of exchanging messages with each other. And the good example is like a marketing, you know, quote unquote team of agents where you've got someone that's doing the research and someone that's pitching the ideas, someone that's drafting an editor, you know, chief of staff, a SEO person, things like that. So it's really all centered around some task and they're they're choreographing a broader project. And the same holds true in software development. We're seeing a lot of that with uh you've got a developer, a designer, a developer, a tester, a release manager, all of these sort of wearing these different hats. It's still, I'm sure that some people have this working out really brilliantly well. But in most experiences, they tend to come up with a lot of they get cluttered, they get confused, they they run off the track a bit. And you have to sort of bring them back in. You need to ground them, you need to QA it yourself. So there's still very much a human in the loop. But again, that's all sort of just all working around one particular objective, and it's sort of a team of agents marching to that. Where it gets more interesting is when you have, you know, take a finance department and you could have agents that are reviewing invoices for potential likely rejections. They're scrubbing the narratives, then they're actually submitting the batches to an e-billing provider, and then they're getting kicked back from that and they're scrubbing that as well. And they're almost doing not just a an individual task or set of tasks, but an actual job or role. That starts to get interesting because then how do they, you know, how are they interacting? It's more instead of just a scrum, it's more of a coordinated organization with roles and interactivity, dependencies, guardrails, things like that. So when it starts to reach at that organizational level, I think there's going to be, and again, it's not there yet, but I think that's where it's heading next.

SPEAKER_01 12:02

Kind of like personas all working together. And then you have that agent-to-agent interaction. One place where I've seen it is you have folks that like to live in their tools. They don't like to go to other systems or log in, depending on what their persona is. And so they stay in that tool, ask for the thing that they need. And then no matter where it is, there's some kind of integration. Now that's kind of more like the MCP use case. Agent to agent, you'll have the agents doing that with each other. And then I think, yeah, now you have your finance department, all your personas running together, interacting with the different systems, whether A2A or most likely A2A. So I think that's going to be really, really interesting. And a lot of the tech is at least there where it supports the protocol. We might not be doing anything with it yet, but get there. So you're coming at this from inside enterprise law firm, high stakes, very heavily regulated. What's the argument for moving fast on agentic adoption? And what's the legitimate argument for pumping the brakes? So basically, why should we evangelize this? And then how should we show some caution? Yeah.

SPEAKER_02 13:15

Yeah, no, that's that's a great question. And it is it's the conversation that I'm having, you know, internally as well as across the industry, no surprises, right? When you start getting into agents that do, and particularly if we're heading to a point where everybody has one or more agents working, you know, quote unquote working for them, that's what I think that, by the way, I think that's what the Internet of Agents is, is everybody has agents. Everybody has multiple agents that are coordinating and working with them. Those agents are talking to other people's agents, and you start getting this very spider webby kind of thing where the agent-to-agent interaction far outpaces any human involvement, right?

SPEAKER_01 13:51

But but so you're telling me that I can get my car warranty renewed automatically.

SPEAKER_02 13:56

And it will automatically subtract way more than you should pay for that out of your bank account as a substance to you. Yeah. So br bringing it back though to your to your question, right? Like that's where the guardrails do come in. So you can imagine this place where humans in the loop become a really inefficient bottleneck to things just happening. So we need to be cautious in thinking about where do we need to pump the brakes and where should where do we not need to be as concerned, right? So it comes down to questions of judgment and of law, you know, patent. You still can't be a computer, an AI patent author. So there needs to be human in the loop on there by the law. That's that's the current law, at least in the US. So I think there's all of these sort of really nuanced checks and balances to think about when you do want to put brakes on and controls. And law is one good example, health is another, insurance is another. There's things that can be automated in that value stream, but there's definitely things you do not want to automate in the value stream. So in any organization, it's looking at what are those tasks that are high repetition, high cost, low, low risk. And those are the first targets you tackle. And you should be moving fast on those. And before you actually go to automate all that, ask the question of whether it's really adding value at all. Maybe you could just cut it out. But if you can't cut it out of your value stream, okay, then look how to streamline and automate it. But, you know, I do there's that Mark Zuckerberg who was famous for saying move fast and break things early on. And then people realized that was just a terrible way to run a scalable business. So his new set newer saying was like move fast on stable infrastructure. I think that's the right thinking about AI overall. Because I'll share my experiments, yeah, with with coding. Like it's like I said, it's great for that Rev 1. And then you go to try to maintain it or update it or scale it or test it. And you're like, so you know, there are pieces outside of it that still need to be done and probably done by humans to really get to where we need to go from a quality perspective. That's one example.

SPEAKER_01 15:56

Feels uh very much like going from mail or facts to email. Kind of feels like that feeling again. Removing those bottlenecks. If you had to place a bet on the adoption versus not adoption, what side do you see winning in the next 18 months? I'm assuming you mean for agents specifically. Yeah. Yeah. Like, can you just put your head in the sand and ignore that these things exist?

SPEAKER_00 16:21

Or do you have to try it? Or at least a little bit at least, you know, let some of it do that goal.

SPEAKER_02 16:26

I I honestly think, you know, and this will take some time to flow through the systems, through the technologies that are out there, the vendors and so forth, service now being a great example of that. This will reach a point of soon, I think. I don't know if it's 18 months, but soon it'll reach a point of pervasiveness where it's like an Alexa in every house kind of thing, where it just becomes part of the norm of you're going into your ticketing system, you're going into your financial system, your ERP, your CRM, and you're just asking it to do things and it does these things for you. And it does some added value that you hadn't thought about. It's like, oh, thank you. That was so preactive. And it doesn't even feel like magic anymore. It just seems like that's part of the product. It's a product experience. That I think is what will tip. You know, if I were bullish on adoption, it would tip more in that way. If it's going to be a bunch of people custom defining what an agent does, not necessarily coding, but like writing the instructions and doing all these things for every single task that they're like, I think that the barrier to entry on that will get done in pockets, but it's not going to be mass adoption.

SPEAKER_00 17:30

So has anyone in your world actually had to answer this question or is it still theoretical? When an autonomous agent makes a mistake, sends the wrong thing, touches the wrong record, who's accountable?

SPEAKER_02 17:42

I will say, I won't give a specific example from my firm, but I'll say in the legal industry, there is an ethical obligation on the lawyers that they have the final check. And that could be whether it's an associate providing the partner a draft or an AI or whatever. Like it's at the end of the day, the person who is the client representative, the partner, the attorney, it's their job. They, if it's messed up, it doesn't matter how it got messed up. They can blame the associate behind the scenes all they want, or they can blame the AI. But at the end of the day, it's their neck on the ethical line. Uh and that's played out hundreds of times as lawyers use ChatGPT to make up cases where they didn't realize they were making up cases and then get sanctioned by courts and otherwise, and they still don't learn. So I think it is ultimately now that's on the, again, on like the chatbot kind of assistant. I think the same holds true for an agent. If you didn't set up the right guardrails and it made a mistake, that's on you. Like that was you as the supervisor, as the manager of these employee, you know, digital employees. It it sits with you. That's how I think it will persist for at least the next couple of years. Again, when agents dwarf our ability to manage and corral them, that becomes a much more sort of existentially difficult question as it gets more integrated into our systems and our environment. But I think in the short run, it's us.

SPEAKER_00 19:03

So and just talk about the legal vertical for a moment. And most people may not care, but you and I have lived this for almost a lifetime, our lifetime, right? And in this vertical, this is nothing new to attorneys, having to spot check and double check and make sure it's correct. I mean, they had to do this with precedent in case law, right? Make sure there wasn't an overruling precedent or something else, you know. They they've they've always had to. And and the way they searched in Lexus Nexus was almost like an AI prompt. If you're good at it, it was like an AI prompt, you know, thinking it through, coming up with the entire scenario, searched, because they had to pay per search and per but really it was per returns. They paid for the returns. So if they had horrible searching methods, they they really ran up the bill on Lexus. So this is something attorneys probably, to me, it shouldn't, it shouldn't even be a factor because I had to do this anyway. You know, I'm not supposed to talk about cases in the elevator. I'm not, you know, it's just one of those things that they just can't do and or they need to do. And for others, I think it's a test on your willingness to, you know, bet your horses on everything it says and does. But for attorneys, they lived in this doubtful, debateful world anyway.

SPEAKER_02 20:18

And I think, you know, as a society, don't always uh we like to joke about lawyers and and so forth, but there are some things that they've done that I think we can learn from. If you fast forward the clock a little bit, Brian, from from that those original days. So back in the days of looking up cases, whether it was print or digital, it was deterministic. Either it existed or it didn't. The text said what it said, and it was up to you if you missed, you know, if you didn't find the thing, it was on you. Fast forward a little bit, and then there's e-discovery, right? So that used to be that you would sit and read every single document and a litigation, you'd really say you'd read every single email that was sent. And it very quickly became just so much a massive volume of data that they couldn't possibly go through it all page by page. And so then they invented search engines and and predictive algorithms that became some deterministic, but increasingly more statistically provable that you found what you found or that you had turned over every stone. The challenge with AI is that it's neither of those things. It's not deterministic, it's not even statistical because you can get different answers every time you ask and it not be coherent. And that becomes a real problem when you get into agents taking action and making decisions on your behalf because it's it it doesn't. It does a really good job, but it can also make mistakes. And how do you know when it when it will and when it won't? So I do think like we we are so enamored with AI, we're getting increasingly enamored with agents, myself included, about the easy button. But there is a place in our workflows for more deterministic outcomes that absolutely should still belong in an organization, if that makes sense. So think finance, for instance. Yeah, you saw someone say, I set up an MCP to my ERP and it was making general ledger entries. And I'm like, are you kidding me? I think the AI tools can help us create those deterministic workflows and integrations and things like that much more quickly, which is great. But there is a place for if-then kind of logic still in the world of organizations.

SPEAKER_01 22:20

What does the governance model look like for an agentic process? Would you requ what would you require before you signed off on something running autonomously in your environment? So you you give them the keys, go do, don't bug me. I don't need to hit approve every time. I don't need to set the auto approve. It's designed to do it autonomously. So any kind of guidelines or thoughts on what you would require?

SPEAKER_02 22:48

Yes. I'm thinking through if my answer is any different than what it would be for any other manager managing a process. You know, if I'm I'm thinking sort of in the olden days I had a team that would build something I would ask, okay, you know, how'd you test it? Well I clicked here here and there. Well did you try this? Did you try these edge cases? Did you try to break it? Did you overload it? How do you know it's right? What's the approval you know to your point, what's the approval escalation process? What happens when it hits an error? Like I'd be sort of drilling into those kind of things to really stress test how it was designed. And I think in the current iteration of agents, you can do that by looking at definitions and instructions and things of that nature, as well as the test harness you should absolutely have a test harness around that agent to understand, okay, run it in a simulation a hundred, a thousand times and see what happens. So I think they I think the answer is pretty much the same, but using tools to help you because again it's going to be so much harder to validate these things at scale because there's so many of them and it's moving so quickly. So you need, you need test harnesses, you need test scripts, you need a good sense of what the outcome is going to look like and where it can break. But I would go back to then okay, what's the what's the sphere it's working in, what's the data it's working with and how damaging is the outcome. So I'd I'd have almost a uh thinking back to like uh in cybersecurity land or in both SWAT. Yeah, swat yeah the sort of the analysis sort of like what's the likelihood, what's the impact, and you get a sense of, you know, what's the worst case here? If the worst case is not that great, not that big and and it's also unlikely and it's recoverable, then I think you're like, okay, we can we can live with that. You know, if it's an operating room, if it's if it's a healthcare provider, it'd be a little more concerned.

SPEAKER_01 24:33

I uh actually an analogy of that I I know someone who works in healthcare and they were in the OR and it was an operation with the robot and the robot stopped working and they didn't know how to do the surgery manually without the robot. And luckily there was an attending there and so they were able to get through and do the surgery and they showed them how to do it the old school way. But yeah, that's I I think that having some kind of evaluation framework or some kind of dashboard that will provide that recommendation of yes, this is good. Here's the data set, here's what we tested against, here's probably some risks that you should know about. It was synthetic data that we tested against instead of actual data and just creating some kind of repeatable framework there. That again using ServiceNow as an example, skill kit comes to mind and data kit, which allows you to generate data sets, but also test both prompts or skills and the agentic agents and workflows and just how successful those agents are at running.

unknown 25:31

Yeah.

SPEAKER_02 25:32

It's worth it's worth drilling in by the way we may have talked about this in in prior times, but you know the two things that I see as bigger challenges or opportunities. So on the challenge side the how to phrase it, the laziness is real. Like it has gotten so good that after the hundredth time of it doing well, you're like, what what is it? Why do I have to look at it the hundred and first time like I just keep clicking okay. And so that works well again in low stakes things that you don't care about. But if it's a skill that you need to continue to hone and you need to be sort of on top of your game on that, like if it if you if you atrophy at that skill and you're asleep at the switch, that's a that's a bad outcome. And to your point, these systems are systems. They fail. There was a uh situation where one of the tools we use one of the AI tools was unavailable for a few hours on a Monday afternoon. And for most like search or productivity tools, you know, that happens you get a workaround like it's not the end of the world I had a few users that were like I have I have this thing due and this is down. This is unacceptable and you've had this for three months. What what did you do before? Do that thing. So it really but but I I but I I get it because you get so dependent on these tools so quickly which is a risk which is a risk. So and then there's a concentration risk too if you all your eggs are in the clawed basket and you're a defense contractor bad news you're scrambling to put in open AI or something else or uh you know or you are again in one of those enterprise uh platforms and they switch from all you can eat into usage based, token based pricing and your bill goes up four or five X overnight. Uh these are real considerations that we all in the tech space have been dealing with with other innovations, other technologies and and architectures and we're here again. Which is the good news. We a lot of the playbooks we've used elsewhere we should be applying here and not get so sort of sucked into the AI hype vortex that we forget you know some of the basic blocking tech. I like that analogy.

SPEAKER_00 27:35

So we've got uh three hot takes for you before we close out one of them is and you know service desk and like what kind of tickets are in service desk. So to me that's a big piece of this question. But in five years, do you feel most IT service desk tickets won't be touched by a human from open to close?

SPEAKER_02 27:55

I think yes. I also think the vault if Windows and Apple can sort of get their their stuff together and use some of these technologies for good in the operating system, half the things that are ticket creating in the first place shouldn't have happened. The system could have fixed itself here's hoping hope springs eternal on that one.

SPEAKER_01 28:16

Yeah yeah fingers fingers fingers crossed here's here's one thing that I keep thinking about there are certain folks that are very very loud about the AI risk. However those folks are typically not as in the weeds or not really users of the technology there there's still a lot of that hesitation we think about the bell curve of adoption these would be the folks kind of the far far right of that and then there's I mean kind of on the reverse side there's folks that were early adopters that are just building away not thinking about the risk and then their database and all their snapshots are deleted. Is this kind of a fair read and do we think this is going to follow that same kind of curve or what what are your thoughts on that?

SPEAKER_02 28:56

Yeah I I think it's a fair read for sure I see I see both ends of that spectrum. It's almost like a kumbaya opportunity of getting those two groups together because the the builders once you see if you see that happen to someone else or you internalize like oh my gosh that could happen to me you start to think about those differently you start to put in guardrails, things of that nature. But yeah the people the people that are the biggest opponents of it have never even opened a a chatbot window. So they're there's they're so far behind where the current state of the art is that they really can't speak with any experience or if they have used it, they used it four years ago with GPT 35 and they're like, well this is worthless because I tried one thing and it didn't work. And I'm like, yeah, you should pick it up and try it again. Yeah I do I do think there is like any other technology wave that's hit us uh there are the people that sit in the roles that are at the intersection of those two. I'm thinking of the risk officers and the cybersecurity managers and people that CIOs are another like that that have had to live in both worlds and see both sides and empathize with both sides and find common ground. I think those people will continue to be really important to help bring those two groups a little to bring them along and bring them a little closer to the to the center of responsible use for AI, not just AI at all costs.

SPEAKER_00 30:13

There you go. All right so last hot take most companies call themselves AI powered right now and are basically just running a fancy form with a submit button. So at what point does that become false advertising or what do they need to actually do to earn that label? Yeah and you said AI powered right AI powered and that that can be a lot of things. I mean I don't mean to answer the question all for you but it's like if it's AI powered you may want to know that that there's some risk involved in the answers you're going to be getting or the delivery of the product you're going to be getting because there is AI but in the background you don't know the levels of trust and things their systems and their guardrails or is it really powered AI powered you know so I don't know.

SPEAKER_02 30:59

Yeah there's a great there's a great Simpsons episode with the Phil Hartman who's a realtor and his ex you don't need to know Simpsons or the or the character but he goes there's the truth and the truth. So I think that a lot of this is marketing spin, it's AI washing it doesn't hold up past the sales pitch. And so you know we've within our firm and others that I advise the same is get technologists in the room to have conversation with the vendors technologists and really get to the root of it pretty quickly. The vendors technologists, the product folks and the people that are building they're pretty quick like once they see someone that is of their own tribe, they'll pretty quickly be like, oh yeah, no, that ignore all that. But what we're really doing is over here and it's pretty cool and let me let me show it to you. So I, you know, I do think to your point though, like throwing the marketing spin and hype out of the window I'm seeing with technology providers that it's starting to diverge between those that are AI powered AI native, however you've embraced it in a in a really mature, nuanced, structured way and you're seeing them innovate quicker, higher quality work product coming out, more stable systems, you know, you're seeing them leverage this and their business is getting better and better. And then there are those more legacy companies that aren't investing, don't have the skill set internally, haven't tried to build it out and they're still moving quickly to keep up, but their quality is suffering, the stability is suffering and the products are landing flat because they really don't understand the market and the use cases and things like that. So I am seeing this sort of divergence between different kinds of vendors and they can all call themselves AI powered all they want, but in the outcomes you see, you know the difference. That's been my perspective already and we're only you know a year or two into companies really being able to leverage this to good effect.

SPEAKER_00 32:54

So Steele what do you want to do? You wrap up you got many more questions?

SPEAKER_01 32:58

I have so many more questions but no I I I think if there was I I've got one last one I'd we can wrap it up. If somebody was at that intersection and they had not yet taken the leap and used an agent what would you what would be the easiest lowest effort can do this today thing to go touch an agent and see what all the fuss is about.

SPEAKER_02 33:25

I will I don't know if it's 15 minutes it's probably not much more than that. A lot of the non-technical entrepreneurs founders friends that I will work with most of them are on ChatGPT they're on the web browser. They maybe tried Claude on the web they're maybe using a free account and they're just dabbling in it. And I've advised a couple of them like go get pay the 20 bucks ChatGPT or Claude whichever you prefer git codex or git co-work for Anthropic and take one task that you hate doing. It could be taking that spreadsheet to produce a monthly report or it could be taking all your time entries and creating that bill and you do it manually on a weekend and it just drains you take that give co-pilot or sorry codex or cowork give it you know the last three months give it some instructions and say hey simulate doing this for my next month right something like that. Low stakes easy to replicate that starts to get them familiar with the concept of repeatability and with instructions and skills and refining and seeing outcomes and seeing where things break down. So that to me like and and you know I would get I got a text from one of my friends on like a day after I said hey you should go do this. And all it was was a picture uh Barney from how I met your mother going you know exploding brain that was the gift that he sent me so yeah yeah yeah I think I think just getting it it's getting started right like it's just taking that first step and that's a pretty low low friction first step.

SPEAKER_00 34:55

Thanks Matt for joining us always loving having you on especially by this AI stuff just uncovering all the different uh possibilities and capabilities that this new age is going to be bringing us so um the inter the Internet of Agents isn't waiting on us it's it's being built right now one workflow at a time so the question isn't come isn't if it's coming to your world it's whether you're going to be the one wiring it up or getting surprised by it. So uh it you're your head in the sand I guess we hope you're we uh we hope you're wiring it so if this episode got your gears turning share it with someone on your team who needs to hear it subscribe leave us a review and we'll see you on the next one. I'm Pops and this has been the Wired Girl Cheers