Transcript

Episode 113 - Pat Barry

Erik Martinez: [00:00:00] You know, back in December, Pat Barry and I sat down and made our predictions for 2026. Agents, AI shopping, rising costs, governance, all of it. I've been thinking about that conversation a lot because we're six months into the year now, and the honest answer is we got some of it right, but not the way we expected.

AI shopping, it didn't turn out into the buy right inside ChatGPT thing that was getting announced last winter. But I've personally been doing these house projects, and I realized that I'm doing all of my research, all of my planning, all of my material sourcing inside an LLM. I just didn't make the final purchase inside ChatGPT.

So did the prediction come true? Kinda, but just a little sideways. I stopped doing most of my work inside my tried and true apps like Excel, Word, and PowerPoint, and now I do that work inside of Claude CoWork or ChatGPT Codex. Agents are real and powerful, but they're not the autonomous swarms that were going to do all of our work for us.

They're more like a coworker who gets you eighty-five percent of the way there, [00:01:00] and then you finish the project. So the question I keep coming back to is, if the predictions came true sideways, what does that actually tell us about where we are? Pat and I dig into that in this episode.

I'm Erik Martinez, and this is the Digital Velocity podcast.

Pat, welcome back to the show.

Pat Barry: Hey, thanks buddy. Good to see you. Good to be back. yeah, it's a little bit different than when we chatted, was that mid-December?

Erik Martinez: Mid-December.

Pat Barry: We chatted a few times in between, but, yeah, it was funny. I actually don't recall what we said would happen. pretty sure not Terminators, but, we may have to review that 'cause I know some things, I think some things I think we were probably right about.

And it's only six months in too, so I don't wanna say, you know, we got another six months to get some of these right. Let's not, let's not jump the gun. So it's like, like, like the halfway review point, like thing we'll hit kinda, measurement like type of

Erik Martinez: Let's talk about the thing we missed. AI shopping has not become a thing.

Pat Barry: Yeah.

Erik Martinez: I look at it and [00:02:00] go, "Well, it's kind of a thing," 'cause I use AI to find shit, I've done three fairly significant house projects. I use AI to help source all the materials and help me do the planning for these projects, which has been almost spot on.

Pat Barry: Yeah

Erik Martinez: It has been spot on crazy in doing these house projects, yet I didn't complete the purchase on ChatGPT or Claude, but I did do the vast majority of the research and shopping in an LLM.

Pat Barry: Well, pretty much. I think a lot of it was probably, if I recall at the time, right around when we made that podcast, there was big announcements from ChatGPT. ChatGPT was trying it, like integrating, it was Walmart, I believe. some sort of partnership program with several large retailers where you could just buy right in there, and I feel like it was around February, March-ish, they killed it.

Like, it just was not going well. I mean, I can think of a million different things, like looking back on it, that yes, would definitely [00:03:00] go wrong, but at the time it seemed like, all right, well, it looks like they're all gonna try and monetize this way. 'Cause I mean, there's still... It made sense.

Lord knows what the rip was for ChatGPT, Gemini, whoever was gonna do it. I think probably security, maybe trust, user trust. I don't know. I can't really remember, but I just remember it just ended. Like that was just kinda it. ChatGPT, "Well, we're gonna go in a different direction."

But from a research perspective, That's just where I started as opposed to the web or... I used to go to YouTube for my home fix-it projects and that I'd spend hours just trying to find a video, and I fix stuff a lot faster now 'cause AI just tells me how to do it.

Erik Martinez: Yeah, step by

Pat Barry: weeds.

Erik Martinez: Here's the parts you need, what size, the whole nine yards.

And obviously being an amateur we misuse stuff. So, I mean, that, I think that's one form of shopping that we're, lots of people are doing. I hear lots of people planning their vacations and travel plans using, AI. I've done a little bit of that. but it's not direct e-commerce. I still think it's coming.

Pat Barry: I mean

Erik Martinez: I think [00:04:00] Google will get there. Google launched that whole universal commerce protocol with a few other companies. It hasn't taken root yet, but Shopify jumped on a parallel tangent to kind of get some of that stuff going.

Pat Barry: I think some of it'll come back around. I mean, people shop. That's just how it is. I mean, I remember, I'm old enough to remember when I started my career in digital, like, that internet shopping was just kinda starting out. Like, nobody really did it yet. It was still not... UPS and FedEx weren't all, like, crazy looped into it.

But it change over time. It kinda makes me think just, I feel like we talked about tools and, consolidation, just how things are evolving as opposed to last year, how spread out they were. do you think, you know, have we seen, kinda changes just in how the tools are presented, compared to our prediction?

Erik Martinez: Hell yeah. Six months ago, I was still in the chat interface on a browser, and today I do ninety percent of my work inside of Claude Cowork or ChatGPT [00:05:00] Codex.

Pat Barry: Yeah. It's turning into... They're all turning into office suites.

Erik Martinez: If you had asked me six months ago, "Hey, man, apps would you uh, would you be working in?"

And I would've been, "Yeah, I spend a lot of time in Excel, and I spend a lot of time in PowerPoint, and spend a lot of time in Canva." You know, you go list, right? And, you know, the reality is I spend almost zero time.

Pat Barry: Yep

Erik Martinez: I do spend time in them, but it's more to review information. I'm not actually doing most of the work.

Yeah, okay, once in a while I'll type in a cell formula, but I can tell Claude or ChatGPT or Codex what to do, and it goes and does it.

Pat Barry: I showed some folks Copilot in Excel, just so to just demo fake synthetic data. The one thing I told them not to do is you shouldn't even be writing formulas anymore. Just type it in the AI. So what are you doing writing those formulas, man? Come on!

We'll see how things play but, forever it was Microsoft just was the Office Suite, then [00:06:00] Google, I don't know when that was, 10 years ago, 12 years ago, kinda came in with Workspace and started to gain traction. And I think it's gonna be Claude and ChatGPT ChatGPT 'cause that, it, that's literally what they're doing.

So I think it's gonna evolve. Last year just everything was just everywhere. They're all changing, and I think it's all gonna consolidate. Yesterday Microsoft released their version of Cowork, I feel like we called that right.

Erik Martinez: One of the things that's really interesting for me, because I was having this conversation yesterday with all these tools. So you've seen some of the little dashboards and things I've been creating using Codex. I was, I was having this conversation the other day, and I'm like realizing, if you look at what Salesforce is doing, they've gone from the whole agent force, now they're going to kind of a headless, technology. And I really think that's probably the way is gonna go, because there's so many features inside these native LLMs and the frontier models, and even in some of the open source [00:07:00] models that what you really need is the glue. And, I'll give you the example. So if you are an account manager, and you know this because you've automated a bunch of your stuff. But before you automated your stuff, think about all the applications you touched. I touch my email, I touch my,Slack threads. I have like Slack and, Teamsand, Google Chat and, WhatsApp. Those are the four that I'm in like constantly,

Pat Barry: Yeah

Erik Martinez: and I get messages across all that stuff. And then you're in your Excel and your PowerPoint, and you're in SharePoint, and if you're an agency, you just can't get away from some clients being Google-oriented, so you've gotta have workspace whether you wanna or not, or you're in Microsoft whether you wanna be or not, right?

We got Teams and we got Zoom and my point is all this stuff, we would have to click around and get in all this stuff. And, you know, I'm a big fan. You turned me on to him, Jordan Wilson at [00:08:00] AI, and I don't agree with everything he, says, but one of the things he he likes, the term he likes to use is human duct tape, and that's what it that's what it feels like.

You're touching all these things. I realized, oh my God, I can build a unified display system, funnel everything into one screen, and manage everything right there.

And almost never go into the apps anymore. I probably spend too much time in my emails still. I've got work on that. But the reality is, you have the ability to create work, I'm gonna call it surfaces, for lack of a better term, but the UX for the way you actually work, not that hard. Does require a little expertise and savvy, I couldn't have done that six months ago

Pat Barry: It's how quickly it's advancing and it will continue to. It just seems like how it is. What were some other predictions we made last year?

Erik Martinez: The big one was token costs. You were dead right. You called that

one, [00:09:00] man

Pat Barry: Not yet. We're still halfway through. 'Cause Microsoft, I know I mentioned they just released, CoWork in their platform. I have to ask a client if we should preview it next week actually. It's usage-based pricing, which, here we go.

So I don't know. I have to look into that for a client, give them a little preview of, like, cost and stuff like that. So I'll, I'll probably get to try it and not have to worry about paying for it.

Erik Martinez: Well I think these businesses can't just offer us a $20 or a $30 a month plan and expect... Even the $200 month plan isn't worth it. The challenge is, is the existing SaaS is such that what I'm gonna call legacy companies. Let's... You know what? Let's not use the word legacy. Let's say non-AI companies, to change their whole systems overnight, and one of the things I believe is where we're headed, and we see this with the LLMs. We're not seeing it with the SaaS models yet, but back to Salesforce it's all [00:10:00] gonna be API-based .

Pat Barry: Oh yeah.

Erik Martinez: The challenge with that is you just don't know what meter is. Paul Roetzer and Mike Kaput on the Artificial Intelligence Show, Jordan on Everyday AI, the guys at This Old Marketing, Joe Pulizzi and Robert Rose, all those guys. They're all saying it from slightly different perspectives, but it is maddening to figure out what your token costs are. Absolutely maddening. You have no idea. Like I watch my my Claude usage like a hawk. Which model am I using? How fast am I burning through tokens? Is that an appropriate level of model for the task that I'm doing? I'm starting to play some of those games, and then you hear the stories about, I don't even remember the name of a corporation that burned through half a billion dollars in a month.

When when Claude launched Fable, they doubled the price.

Pat Barry: Oh, yeah.

Erik Martinez: Well, they had to [00:11:00] they had to pull it, and that's a whole different discussion that we're not gonna get into. But is, you were dead on right about this token cost.

I was just talking to, this firm in the Philippines that does digital marketing support and and, and a wide variety of things. And I'm like sitting there going, "Okay, well, I could build something in Claude to do this, and it'll take me this amount of time, or I could just have guys do it, and it'd probably cost me about the same." And maybe you have a completely different opinion, but I think we're getting to that point where the costs are going up, the time to do the more complicated things requires more time

Pat Barry: Yeah

Erik Martinez: To do them, you can do them you can do them fairly quickly, but it still takes time and expertise to do. And at some point, there comes a tipping point, you say, "You know what? I can have the human do it." And it's gonna-- it might cost me about the same or a little bit more, but it saves me the the [00:12:00] probability of having to develop it. What do you think?

Pat Barry: Yeah. It's more, build versus buy. What's the opportunity cost? I think it just depends on, like, what you wanna build? Cause, like, I've been able to keep my costs down, but it's also 'cause I, you know, I have to work in the client's environments, and they have big, you know, enterprise licensing, so stuff they don't care.

Like, "Here, just use ours and work here." At the same time too, thinkthat, that's, that'll continue. mean, probably Google and Microsoft will have subscriptions baked in. All that, like, infrastructure is already set. I think Claude's probably the most, like transparent with quote-unquote token usage.

Erik Martinez: Assuming you actually know how they're burning the tokens.

Pat Barry: I mean, if it's, if you're just using the desktop stuff, it's not you look at the bar in Claude and that's it. Nothing else really tells you what's going on. I don't know about ChatGPT, I haven't been in there in a bit.

Erik Martinez: Codex will tell you your usage.

Pat Barry: I mean, when you're using the APIs, like you can, there's definitely like they tell you how much they're using them, but it doesn't necessarily always translate it to X amount of tokens. I mean these these companies are not [00:13:00] making money. If I remember a stat from I think a couple of days ago, that ChatGPT had, I can't remember if this was past quarter reporting earnings, but like, I don't know, negative like forty billion dollars in losses.

Erik Martinez: $40 billion in losses.

Pat Barry: That's a ton. Well, they all look at Amazon and go, 'cause that's that's the Amazon story. Like they weren't profitable for I think a seven to ten years. But 40 billion, that, maybe that was a yearly figure. I don't know. It's a lot of money. And not, they're not all, making money. So at some point the VCs just stop because there's no return.Well, these companies just have to make more money. They, they've given it away cheap so people get hooked and get used to it. It's like digital drugs.

Erik Martinez: Software, right? Streaming TV.

Pat Barry: Yeah, and, companies have now implemented workflows with these tools and systems. You can definitely start to track, like, all right, are we being more efficient? Like, I start to get asked that more and more. Last year I feel like I'd get asked like not very much. It was more like, "I'm frightened. What do I do?" And now it's more like, [00:14:00] "This stuff's kind of cool, but I'm still I don't know." So it's more of that. But, um, still think they'll, they'll all move to some sort of usage-based pricing. Buy in bulk and you'll get a token discount. Picture tokens as a kid at the arcade. Like, I don't know. I went to one, then you bought twenty, you got like five for free or something.

They'll pull the same shit. It's, you know, get a deal, buy in bulk. So I think that's what it's gonna come down to. But I think some of it, you'll start to bucket into certain tasks. Like, uh, I started using like deep research again, like whole bunch, like for clients, but also for like my own business.

And some of the workflows I put together just on, with that alone, I'm like, damn, like you can fully improve like internal comms for things,research into, a, a deck a movie, literally a movie in Notebook LM or like an infographic. So I, I think the way to communicate internally that it's just people get hooked on that because it makes sense.

Like I do too. Like I get it. It's hard to go back 'cause even now I'm [00:15:00] doing things that again, I could kinda always do, but I would need others to help execute. Now I don't need that. I can just go do it with an agent who's like, I talk to my agents like a coworker or like somebody that works for me.

I think if people start to kind of work like that and start to understand more like how efficient it can make just their workday better. 'Cause I still like, I can handle, you know, I've been lucky this year. I've grown quite a bit, and I'm able to handle all the workload. Now it's starting to get to the point where I'm like, man, maybe I should grow or I'm gonna have to tell somebody like, I just, you need to pay more money 'cause truth ain't worth the squeeze.

But I this type of workload without AI. Just straight up. So those types of businesses like mine, like yours, like I think they'll continue to thrive, but even containing our own costs like that's-- I'm working with clients now on more sophisticated stuff and that actually comes up like model selection.

What are we gonna build this for? 'Cause we build, uh, RAG-based system where it knows which sub agent to go to and that's the first thing that comes up is we need to [00:16:00] know the cost. Cheap agent for easy work, more expensive agent for harder work. You know, and the more expensive typically those are longer projects that would have taken maybe two weeks. So yeah, maybe instead of, you know, you building it, farm it out to the dudes in overseas like you were gonna do, but instead you give that fifty bucks to ChatGPT and just said, "You know what?

I'd have to go buy a research paper anyway in the old, in, in the olden days pre-2023, uh, from Gartner. So why don't I just pay this agent that'll take you know, anywhere from ten minutes to an hour to put together this long doc with citationsand all that?" So as business becomes more and cost becomes more of a thing, I think that's where the evolution of jobs will occur too because you're gonna have to bring in specialists that know how to measure that shit. How much are we using the API? How much-- what tokens are we using? And I think there'll be some sort of laws or business regulations that will force the companies to tell you how much you're using because they will have [00:17:00] to. It's like gas. Like you just kind of have to have a meter and know how much you're gonna have to pay.

Erik Martinez: You know, I heard Paul Roetzer propose the idea of tell me what the annual value is of this agent or system or whatever it is that you have. If you can have a series of agents that do the job and let's say those agents can do 50% of the work, and you value that job at $70,000. Okay. Well, then charge me 35, and i'll pay it.

I listen to Jordan Wilson and, you know, he rants on Claude all the time for their uptime. And I'm like, I have never encountered it, but, you know, like today I was working on a strategy deck for a client. I mean, I've spent eight hours back and forth 'cause it's involved a bunch of analysis and historical data forecasts and, you know, all the things, right? And, I started building a strategy deck, then all of a sudden I'm like the cloud [00:18:00] API is having issues." And sure enough, you go to the status board and it's all red. It was only read for five minutes, but you know, it's the reason Claude went and spent what? How many billions of dollars with xAI for their Colossus Center out of Memphis.

Everybody, needs compute, and I think part of our cost issue, and part of what I think we're seeing in these agents is you could see where the metering is happening.

To a certain extent if you use them enough, you can kind of see where the metering's happening. 'Cause you're gonna sit there and go, "Well, this is a complex process. It's got 45 steps to get through this task, and some of those steps are actually fairly complicated and require a lot of back and forth." Yeah, okay. We need to meter because if everybody did this all the time, we don't have the compute to do itSo it's kind of fascinating.

We kinda set up this year as the year of [00:19:00] agents. And in my head, I wasn't in the agent swarm, You know, the the millions of agents, like, doing their pieces. Do you think that we've hit that reality? Like, are we really in the age of agents yet? Are we like a little bit ahead of the curve and it's not quite there? I'm curious what you think about this. I definitely have thoughts .

Pat Barry: Me and you are ahead of the curve for sure. 'Cause I know, I mean, I have agents that just do stuff for me, decks and proposals. I wrote a bunch to start, I think eight to 10, and they're solid examples 'cause the client signed them and I use those as like, you know, my knowledge base.

Yeah, I think, um, for me it is, ' like I've built a few like real ones. As for adoption, like I know I'm onboarding a new client, they have purchased Gemini Enterprise, and I haven't even done discovery yet. I don't know. But they're very much like been prepping for the past [00:20:00] like year pretty much to like start to do that, be able to sit down and be a you know, future office worker, if you will, where, Pat, come in and show us how to build these agents with workflows and what's our strategy for down the road."

And it's, it's healthcare, so it's kinda complicated. But I would say from everybody that I talk to,it's still very individual. 'Cause I know... I have, I have clients that have... There's individual people that work at the client that I'll talk to And they're like, "Look, I built this thing." And I'm like, "Jesus, man, you should tell other people here."

Like, "Susie can use this. Jim can use this. I was just talking to him." and then they'll say something like, "Yeah, they don't... They're not really into it," or,"Well, yeah, but they're, you know, just kinda got around to it," or sl- And I'm like, "All right. like superstar here, you know, that's picked it up." Um, and I think there's people like that, like individually, and I'm starting to see more of them pop up in organizations where it's like that's kind of like, how do I build one of those? I know you can do it without coding." Um, that's the big thing. I would say more is it's starting to hit people that you can [00:21:00] build more complicated to accomplish more sophisticated tasks for you.

Like for example, like, building decks like, you yes, I have my training decks. I start there. I end up editing, myself, but, gets me 85% of the way there. I don't know, not even like, like quarter of the time. Yeah. It's just like, well, why, why the F wouldn't I do this? And then it-it's mostly I would say with like, Claude Design.

When Claude Design came out- I feel like the hype has kind of died down a little bit, but I had a bunch of people get into that right away, and they "Oh, gee, if I just give it like a style guide, it'll like make my deck just like that every time." And I'm like, "Yeah." I'm like, "You'd just put a prompt a year ago and it would do it," but a lot of people didn't understand that. And so that's probably the biggest thing I see is people realizing if I need something in a specific template or format, I just need to have that in a markdown file in like bullets and plain language, and you just put that in the memory and more sophisticated agent, you know, you file it in the knowledge file and whatnot. But I, [00:22:00] I'd say that's where-- I'd say probably by the end of the year we'd be closer to that kind of agents or people, a higher percentage of people, and actual teams like using agents as a part of the team and part of a workflow. I think that's definitely coming, and I think that'll up more and more. the other thing I could see killing it is the cost, honestly.

Erik Martinez: Well, and that's one of the big drivers behind token usage, right? When you look at, when you look at the data, the advent of agents and more complicated tasks, multimodal, it just uses more stuff. Like video uses more stuff. Audio uses more stuff, right? Video uses more than audio. Audio uses more than text, so on and so forth. Like, if you really think about it, you're like, yeah, we're doing like complicated spreadsheets where you're doing a bunch of calculations almost uses nothing.

All right. Let's move on . The other thing we talked about was shadow AI and governance risk. [00:23:00] You're working with institutions and businesses that this is really important to them. So tell me what, from your perspective, what are you seeing on the security, governance, shadow AI side, and how do we make sure that people are doing the right things from day one without restricting innovation?

Pat Barry: Yeah, great point. I'll reference two clients. I have one that's a higher ed Every listener's heard of then I have, I'm about to onboard a, a new healthcare client that's quite sizable in my neck of the woods in the Chicagoland area. The healthcare client is very well-positioned.

They have use case policy, they've had trainings. Like, this is kind of the stuff they've just sent me in the past couple of days, like, "Hey, get a..." I'm going to meet them tomorrow just to kind of talk potential things we can do, research and stuff like that. But I would say, frankly, that's one of the questions I kind of have for them and to gauge because they have a use case [00:24:00] policy. they've committed to Gemini Enterprise. They've been using the stuff, experimenting, surveying. they have a prompt library that they share with themselves. So it's very-- I'm like, "Wow, guys are pretty advanced, like, for what I'm used to." I'm like, "Good." So that's kind of one of the things that just even onboarding myself with the company was, right, what are you using AI with?"

Well, nothing, 'cause I'm gonna go give you access to your systems, and like I'm an employee, 'cause that's how I like to work. But two, even reviewing have in place. Use case policy, it's clear-cut what can be used and what can't. But I think from what I understand, and again, I'm still reviewing stuff, they had like a trial period where they just said, "Okay, try these things out with these things.

Do these tasks." It's very very specific to, like, healthcare work. I don't wanna get into it. I don't think people understand. It it felt like that, that this team that I'm about to kind of start working long-term strategy and them how to build agents, build agents myself, like do complicated shit with data I think [00:25:00] they'll be very well-positioned, and I look at them, at least right now, today, that's probably, like, the team of the future right there. Properly trained know what they can and can't do. But I think it's 'cause they-- it seems like they were given the freedom for a decent chunk of last year to kind of figure it out like everybody else. I'd say the other client, academic, definitely use it in the day-to-day. There's like fifty thousand people that work there. So they're slightly larger than the healthcare company, but more restrictive, and as they should be. My my project with them is very hardcore data engineering-type stuff, where we're using Snowflake. I justhave a basic level of access and, like, tool usage in it, and that's fine. Like, I'm not complaining at all. But we did chat with my client and some of their internal data engineering team about releasing, " Hey, can I use AI some of this work in Snowflake? Because we'll be done in, a week."

And they were like, "No." And I said- Is it because you'd have to turn it on for everybody, and people are just gonna get in here and [00:26:00] start asking crap about their data and not understand? And they were like, "Yeah, pretty much. How did you know?" And I'm like, "'Cause that's-- that would've been my biggest concern, is people logging in."

So they're, you know, they give people access to everything. It's all secure. I mean, their security unbelievable. They're still experimenting but nobody's really working in workflows yet. And a lot of my work there is really data engineering, so it's not, you know, sitting there showing how to use AI.

It's building attribution models and, helping them understand measurement strategies for marketing. So, that's kind of part of the rub too, is like I'm there to like go implement AI for them. They're already doing it. it's a good to see between kinda two, big companies and how they're functioning. Very restricted, especially with, you know, with university life, it's all PII. I mean, there's student data everywhere, so we have to be very conscious of that so as with healthcare, they have PHI and that's all kinda some of what I'm gonna try to figure out with this healthcare company. You know, how do we use AI in a safe way to help bring more patients to the hospital, but in a [00:27:00] safe and like non-creepy way. Like that's, we've got a way a while to figure it out.

So what about you? Any, any things you see with the folks you talk to, especially in agency life?

Erik Martinez: I think it's a little bit different for a lot of the folks we're working with. They tend to be smaller, organizations, a little more nimble, but they do have a lot of access to client data. And, you know, so the question is how much client data can actually go into a project?

So I was doing this strategy analysis, but I went and pulled data from Google Analytics, Shopify, Meta, Google Ads and Bing Ads and so on and so forth, and I compile it. And, um, you know, Claude was I could just go get this information using MCP out of Shopify."

And I'm "Yeah, I can't do that. I can't authorize you to do that because it's not my system, and I don't have permission. I have, permission from another client where I can, But I've I also very, [00:28:00] very carefully chose which permissions. Like you can't access this, that, or the other thing.

Like, you can get some reports and a couple of things, but you're not getting any customer data or anywhere near any of that, right? So I think it's a little bit case by case on the risk tolerance of the customer and what we're trying to solve, we're doing some of the similar things, maybe in a little bit of a smaller scale. Like we're doing some forecasting projects, We're trying to really understandthe relationship between, all the different marketing channels and the related budgets and how they actually play out in terms of, a holiday strategy for our client. So the problem we're trying to solve is can we actually hit the revenue numbers with the budget you've given us?

Pat Barry: Yeah

Erik Martinez: That's the question, so you got to pull a bunch of data, assemble it, do some prediction, understand that your data is not complete, understand that you're dealing with [00:29:00] first-party pixel-based attribution from the platforms, and then and Google Analytics and- Shopify data, you know, and understand that there's gonna be differences and all of that And I think, you just need to be careful about, mixing the, the data to do that kind of analysis is cool and fun, making sure that you're not giving too much away, and using personal information, because once it's in the LLM, who knows? And, you know, you got Claude who came out with Fable and said, "Yeah, we're gonna a 30-day data retention policy on everything."

Pat Barry: That's right

Erik Martinez: Well, that's that's kind of scary if you're a healthcare company or you're a public institution that has lots of taxpayer data or student data. No, we're not doing that. [00:30:00] So I think governance is evolving, but we're still on the frontiers of it.

When you get access to some of these MCP servers, Some are set up better than others. Some give you way more access than you need, and some have really fine controls. We can kind of set them up so that you just get what you need. So it's gonna be a continuously evolving scenario.

Well, look, we're getting close to our time. I'm curious, one prediction for the balance of the year, and then one thing you'd say, if you're not doing this, you should be.

Pat Barry: For the year I'll say, ChatGPT's IPO does not go well. They're losing subscribers like crazy. So that, that's mine.

One thing you're not doing... I've been ending my prompts now, and again, for more complicated tasks, with, "Ask me questions for clarity." I've always done that, but I've added, "Do not hallucinate [00:31:00] and verify all claims." and it actually cites things when I need it to. So give it a try in your prompts. What about you?

Erik Martinez: I'm curious, why aren't you setting that up as a system instruction?

Pat Barry: Oh, I have it in system instructions. That's what I add to the project at the end to double-check.

Erik Martinez: Yeah, I have a I have a two-source citation policy and graded by quality. Because you don't have the time to check every source. So if you don't have time to check every source and you're gonna publish something, you better be damn sure that it's good.

Prediction, this isn't my idea, but I think this is true. We were talking about agents and you know, teams using agents, and what I'm seeing happen is, more individual agent use. And you're starting to see the LMs start to promote "Hey, here's a marketing agent, here's a da, da, da, da." They need to be heavily customized, they're gonna be able to do some [00:32:00] generic stuff pretty quickly, but I think what you're gonna see is as those things start to come out, like, okay, it's too genericized, and I actually do my job this way, which is a good thing. I think that's gonna challenge more people to think about their job and how they can make it unique. So, that's my prediction.

One thing to do is you, if you're not working in CoWork or Codex, you really need to start doing it. It is an amazing game changer, and I've watched a couple of people this year, just within the last 60 days, like, do a complete turnaround in terms of how they're working, how they're using AI, they've started to adopt some of these tools. So I think, if you're gonna do one thing, if you're not using Cowork or codex, I think you'll you'll find that to be incredible.

Anyways, Pat, as always, man, so much fun to [00:33:00] chat. We didn't tell as many jokes this time.

Pat Barry: We gotta make sure we're little Nostradamuses and get our predictions right.

Erik Martinez: We gotta get our predictions right.

Pat Barry: No worries, man. It was a good time as usual. Great to see you.

Erik Martinez: Appreciate you coming on and spending a bit of time with me to take a 30,000-foot view and throw some ideas out there.

Pat Barry: Hell yeah, buddy. Great to see you.

Erik Martinez: What I'm thinking about after that conversation is that we got our predictions right, just sideways. And what that actually means is the predictions were about AI doing things, and the reality is about how us humans do things differently with AI. Pat uses AI to complete eighty-five percent of our proposal or training deck, and then he edits it himself.

I build a unified work surface so I can stop being the human duct tape between six different apps. A healthcare team writes a use case policy before they deploy a single agent. None of that is the tool doing the work for us. That is us figuring out how to work differently because the AI tool gives us that capability.

So here's the one thing you can do this week. Pick one task you do [00:34:00] regularly, something that touches three or four apps and takes you too long, and instead of asking, "Can AI do this for me?" Ask, " How would I work differently if I could build a unified dashboard or automate those steps?" That's a different question, and I think it's the one that actually matters right now.

Thanks for listening to this episode of the Digital Velocity podcast. Have a fantastic day.

Narrator: Thanks for listening. If this one hit home, pass it along to someone who's working through the same stuff. Leave us a review if you get a chance, and check out digitalvelocitypodcast.com for more. See you next time.