Cowork Builds Apps Now, and ‘Acquired Skills Will Appear Here’ w/ Garett Medlin

Announcer (00:04): Welcome to Raw Data with Rob Collie. Real talk about AI and data for business impact. And now CEO and founder of P3 Adaptive, your host, Rob Collie.

Rob Collie (00:20): Welcome, Garett and Justin, to, I think it's day 49 of the Rob is sick podcast.

Justin Mannhardt (00:26): Stop it.

Rob Collie (00:27): It's true.

Justin Mannhardt (00:27): No.

Rob Collie (00:27): But other people have this.

Justin Mannhardt (00:29): Like, in your community or like, broadly?

Rob Collie (00:32): Yeah. There are other people who are down and out in a very similar way, like sort of having the same energy, fatigue, brain fog that comes and goes, that's lasting long periods of time. I think it's one of those things that might eventually get a name. Maybe it's just long COVID. I don't know. I do know that it sucks with a capital S. Well, anyway, we should introduce you, Garett. So I wanted to introduce Garett as someone who actually has a really hard opinion about how to pronounce G-I-F. It is pronounced GIF. That is the correct pronunciation. I will not hear it otherwise, but Garett is convinced that it's pronounced G-I-F, because you're imagining Garett's name starts with a J, don't you?

Justin Mannhardt (01:15): It doesn't.

Rob Collie (01:15): It does not start with a J. Garett starts with a G. Garett is our AI practice lead here at P3.

Justin Mannhardt (01:23): Congratulations.

Garett Medlin (01:25): Thank you.

Rob Collie (01:26): He's been in this role for a while, really, but we made it official with that symbolic title.

Garett Medlin (01:31): I'm glad to be official.

Rob Collie (01:33): We have something very specific to discuss, at least as a starting point, which is I have been hyping the hell out of Copilot Cowork, which is a product I've never tried. Why not just hype a product you've never tried?

Justin Mannhardt (01:44): It's just your mode, man.

Rob Collie (01:45): Why don't you just say, "Hey, this is going to change everything." And then wait for them to release it and then find out that maybe it's not.

Garett Medlin (01:56): We're still waiting.

Rob Collie (01:56): We're still waiting. Okay. So just to recap, we've done multiple episodes now talking about cowork as kind of like a really, really, really strong gateway drug into the rest of custom AI. And at the beginning, it wasn't running well on Windows. Apparently it's running much better on Windows now.

Justin Mannhardt (02:14): Claude Cowork.

Rob Collie (02:15): Claude Cowork. Yeah. So Claude Cowork is running much better on Windows now and I might not have ever needed to-

Justin Mannhardt (02:20): Cool.

Rob Collie (02:20): ... do the Macintosh thing. Might not have been necessary.

Garett Medlin (02:23): I got it working on my dad's computer.

Rob Collie (02:25): I just feel shame. Shame. I feel shame. But anyway, so I was thinking that, "Hey, if it's not working on Windows, Claude Cowork, but it's about to become available via Copilot subscription and it runs in the cloud, this is going to be amazing." But I think two key legs of support under this hypothesis have both eroded. One of them is that it does work on Windows now. Okay. Kind of funny. And secondly, maybe the one, the Copilot version isn't the equivalent that I was hoping it was going to be. And Garett is here to report from the front lines.

Justin Mannhardt (03:06): From the field. Field report.

Rob Collie (03:08): He's like live via satellite.

Justin Mannhardt (03:11): I think technically that might be accurate.

Rob Collie (03:12): Might be accurate. Yeah, you never know.

Garett Medlin (03:16): I was in the Missouri wilderness, my first exposure to it.

Rob Collie (03:19): I'm pretty sure I know Garett's ISP. Are you on Google Fiber?

Garett Medlin (03:23): I am. I'm very happy to be on Google Fiber. I waited a long time.

Rob Collie (03:26): Ask me how I know that you're on Google Fiber.

Garett Medlin (03:29): How do you know I'm on Google Fiber?

Rob Collie (03:31): Because someone has been visiting my Sith name website from your city, right. That is why I confirmed where you live the other day. I was like, "I think who that is." I've not announced this URL. I've shown great restraint in not messing around with this site. I have not been doing much with this site. There's still a handful of organic people finding it anyway. Even when you filter out all the bots and all the scanning attacks that are trying to find ways to compromise the site, there's still some people here, like, who are these people? I haven't told anyone about this.

Garett Medlin (04:12): Yeah. If you've done no Google SEO work, how are they finding it? I don't think I've told anyone the name of it.

Rob Collie (04:19): There's a handful of people who know. Yes. So he's not on satellite. He's Google Fiber.

Garett Medlin (04:24): Yeah. Hard line.

Rob Collie (04:26): Tell us over that sweet, sweet, sweet fiber optic connection. Tell us the truth about Copilot Cowork.

Garett Medlin (04:34): I think they're starting small and they're dreaming big. It has a cowork-like chat experience. Definitely is paying a little more attention into planning out the tasks and collecting the context, kind of like a souped up version of the typical chat experience. But it can create PowerPoints and Word docs and Excel files. And it is hooked into your OneDrive for business, but it cannot reach into your computer at all, which could be a good thing for governance and enterprise people.

Rob Collie (05:14): That's okay. And for my purposes, everything is in OneDrive anyway. The fact that Claude Cowork won't reach into my email is really annoying. I was really hoping that the Copilot version would sort of be better in every way, but-

Garett Medlin (05:28): But some of the key things we really need, like, custom instructions and memory and projects, not in there, not from what I can see.

Rob Collie (05:42): Really?

Garett Medlin (05:43): So kind of a no-go right now for me.

Rob Collie (05:46): So I couldn't do Eddie, the editor, for my book in the Copilot Cowork version.

Garett Medlin (05:52): Unless you can cram all of that into a single chat experience, I don't think you could.

Rob Collie (05:58): Boo. That's what all these MD files are for.

Justin Mannhardt (06:01): So I'm naively living for the past several weeks in a very non-Windows, non-Microsoft world. Let's just call it a break, fellas. I don't even have Power BI desktop at the moment.

Garett Medlin (06:16): I'm about to jump into that myself for my wife.

Justin Mannhardt (06:19): I've struggled personally to kind of understand why I would jump from Claude Code into Claude Cowork, purely because I've made so much progress in Claude Code on certain things. And there's sort of like a tech debt inertia of moving over there. So hearing your description of Copilot Cowork, God, these nouns, right?

Garett Medlin (06:41): And it's got Frontier in parentheses next to it.

Justin Mannhardt (06:44): Oh, wow. Oh, right. It's like their Frontier program is the release channel or something.

Garett Medlin (06:50): Very simple.

Rob Collie (06:51): Shouldn't all of us get like Davey Crockett coonskin caps for this? Right.

Justin Mannhardt (06:54): I was just like ...

Rob Collie (06:56): It's like Frontier. It's like Frontierland in Disney is not Tomorrowland. Tomorrowland in Disney is the futuristic place. Frontierland is a place that's like .... It's like set in the mid-1800s.

Justin Mannhardt (07:09): Rob, when you're done with the current book, we'll do like a short graphic novel about understanding all the nouns in the AI landscape. So Copilot Cowork, what you just described, Copilot definitely left a lot to be desired in all of its implementations, either as a standalone or in different products like Word, Excel, PowerPoint, Power BI. So you just described it essentially can create those things, which is a big step up from like, "Hey, let me tell you what to go do."

Garett Medlin (07:37): Yeah.

Justin Mannhardt (07:39): It doesn't have skills, doesn't have memory. And then didn't Microsoft just rolled out like the default Copilot and all these apps is agent mode now? What's the value prop between what's happening natively in the apps versus Copilot Cowork if it's not fulfilling all these other promises yet?

Garett Medlin (07:58): They're definitely pushing agents, because technically Cowork Frontier is itself an agent and it does have a section for skills, but it says acquired skills will appear here. I don't know how to give it skills.

Rob Collie (08:12): The ultimate passive voice statement. Acquired skills will show up here. Like, okay.

Justin Mannhardt (08:16): Rob, is there a product manager part of your brain that just is like, "What? Why? Why? We release ... Coming soon."

Rob Collie (08:23): The part that really tickles me is, again, all of the nouns, right. We couldn't just call them skills. We have to call them acquired skills. We have to give them a name that sounds really official and I'm sure it means something, but does it explain what that means?

Garett Medlin (08:38): That is just the tool tip. So they better not start renaming core AI things 'cause then they're really going to be out of control if they do that.

Rob Collie (08:46): Yeah. I've told story before, but it won't stop me from telling it again. It's one of the things that I told all of my new hires that would come to work for me as product managers at Microsoft, the first thing I would tell them is like, "If you want to add a new noun to the product, you have to come to me. You're coming to noun court." If you think we should add some noun to the product.

Justin Mannhardt (09:05): They've loosened up on that over the years, Rob.

Rob Collie (09:10): There used to be at Microsoft, believe it or not, for a while there, a cloud court where ... Did you know this?

Justin Mannhardt (09:17): I didn't.

Rob Collie (09:18): Where for so long they were all under such pressure to ... Every organization under such pressure to deliver cloud revenue that they were all trying to find ways to reclassify regular revenue as cloud revenue, a phrase called cloud washing. And so if you wanted to get your revenue classified as cloud revenue on the official scoreboard, you had to go to cloud court.

Justin Mannhardt (09:43): That's hilarious.

Rob Collie (09:44): I mean, noun court is just like sitting there. Anyway, so acquired skills, whatever those are, and however you acquire them, unknown.

Justin Mannhardt (09:52): It's coming soon. It will appear here soon. That's disappointing.

Garett Medlin (09:58): My wife was the experimental subject and I was experiencing cowork through her. And then side by side was my dad while I was trying to teach him Claude Cowork at the same time to get a real side-by-side real time comparison. Yeah. My dad is continuing on with Claude Cowork. My wife, I don't think she's tried it again.

Rob Collie (10:24): Well, sometimes it's like throwing matches at a campfire, right. Like, the first one goes out, second one goes out, the third one hits the gas tank.

Garett Medlin (10:34): Takes a few tries sometimes. That's where the MacBook I'm picking up from UPS today comes in.

Rob Collie (10:40): Because it runs so well on Windows, right. That's why you needed to go get a MacBook. Have you heard the Scouting Report on the MacBook that doesn't have a delete key? Have you heard that?

Garett Medlin (10:48): I have. Yes, I have. And I listened to all the Scouting Reports, so that's why I've already had Claude print off several cheat sheets and default settings to change. And she already has a separate keyboard and mouse.

Rob Collie (11:02): Does it 3D print a delete key for you? Does it wire it into the motherboard of the machine? Because if it does that, now we're talking.

Garett Medlin (11:14): That's the docking station. She doesn't need to think about that thing's touch pattern keyboard. You plug it in and then you got your stuff you're already used to.

Rob Collie (11:23): This isn't for you.

Garett Medlin (11:24): It's for her. And it's for me to learn how to teach her as well.

Rob Collie (11:28): Classic case of, "Here, honey. Here's a great present." And she's like, "And why is it a Corvette?"

Garett Medlin (11:38): I did wait for the right moment though. I was patient, because the first laptop I bought her was 15 years ago and Windows 10 has been about to kill its support. So she needed a new computer and her work ones will not do the cowork.

Rob Collie (11:56): There you have it. So you're becoming a cowork family. So I really like that about the way you conduct yourself. You're very much on the frontier, Garett. You live on the frontier, like living on the absolute frontier.

Garett Medlin (12:11): Cowboy.

Rob Collie (12:12): Daniel Boone. Did you hear that OpenAI just released one of their new Alamo models? Their last stand.

Justin Mannhardt (12:24): Unreal.

Rob Collie (12:24): You live on the frontier, very much an aggressive adopter of all this stuff. But you also, you take time for the little people. You stop and you help people that haven't been getting into AI. I love that. I really do.

Garett Medlin (12:36): It's about the people. Can't have the AI if you don't have the people.

Rob Collie (12:40): And also, you can't re-experience what it's like to be new to it, except through other people's eyes. I think that's really, really, really impactful and grounding. And so please do continue with your wife and your father. By the way, Justin, the only reason I ever tried cowork is because Garett encouraged me to.

Justin Mannhardt (13:03): Aw.

Rob Collie (13:03): And I had the same statement, like, why the hell would I want cowork if I've already got Claude Code? Why do I want the training wheels version of Claude Code?

Justin Mannhardt (13:15): 'Cause they are different.

Rob Collie (13:16): They are different.

Garett Medlin (13:18): I haven't stated an official opinion on this because I have a lot of social training to not cause conflict.

Justin Mannhardt (13:25): Bring it. Let's go. The gloves are off.

Rob Collie (13:28): Before we even get started, let me lob another grenade into this story. Before we even get ugly, let's really mess things up. All right. So I'm working on this book. Kristy on our team has been coordinating all of the graphics for it. I've been making some of the graphics for the book, some of the block diagrams. And guess what? Using cowork. Just go make me the PNG file. Just go make me the PowerPoint file, whatever. But a lot of times what I need to do is I just need to write down a description in the doc file of the image I need and then move on. 'Cause if I get hung up on an image, I can spend a whole day on it. There's this huge backlog of images piling up that needs to be built and Kristy's been coordinating that. And I just happened to just throw away a thought.

(14:15): I'm like, "Hey, we probably need something to track my feedback on this stuff 'cause I feel this pipeline of images coming towards me. I hear the freight train. I know there's a bunch of stuff coming my way." I'm like, "I can see what's going to happen. It's going to be like Slack message after Slack message after Slack message. What do you think of this one? I'm going to be answering it. And they're not going to be able to track. They're going to lose track of my feedback. It's just going to be a mess." And I go, "We probably need some sort of app for this." I don't know, like, maybe 48 hours later, there's a Slack bot messaging me saying, "Hey, you've got stuff to review." There's a little button to click to review and I click it and it takes me to this webpage and it shows me the image and I go, "Oh my God, it's got an approved checkbox and it's got a send feedback button." I'm like, "Okay, okay. Now we're talking."

(15:01): And I turned around her and I go, "Okay." And we immediately start iterating. Right. I'll go, "All right, well, I need to see the paragraph before and after this so I know where it is." It's just like a disembodied image. I don't remember what it is. I don't know. 20 minutes later I've got that in the app. I'm not building this. Okay. I also need to see my director's notes, like, what was the image I actually asked for? All that kind of stuff. So this thing just keeps getting better. Kristy's doing all of this. Even the Slack bot integration. She's in the game and here's the grenade. She's doing all of this in cowork.

Garett Medlin (15:33): Yeah. I believe it.

Rob Collie (15:35): I didn't. If I was going to say anything, it would be that cowork is not the thing that you use to build apps. I wasn't going to draw any dividing line at all between the two.

Justin Mannhardt (15:48): Is she using the ... The name's going to escape me.

Garett Medlin (15:51): Like, the ability to host your app.

Justin Mannhardt (15:53): Yeah. Just build an app fast and host it. I wonder if, 'cause Anthropic's been doing a ton of that in Cowork new types of artifacts.

Garett Medlin (16:02): And I'm not sure it's even called something or it's just part of their artifact hosting stuff that's part of it now. So I will say cowork is my favorite when it comes to having people adopt and use AI who are not developers. Like, it's just, of course, right. You see people sticking to that once they get onto it. Claude Code is still my primary interface that I prefer, but it calls out to gooey apps and tools too.

Rob Collie (16:40): By the way, when I click on the app to review things, it takes me to a subdomain of a website called onrender.com. Never heard of this. This is not some built-in Anthropic thing.

Garett Medlin (16:53): A cloud for builders. Yeah. So it must have been something Claude suggested.

Rob Collie (16:57): Yeah. It's like another railway type of ... I'd never heard of railway until Claude started pushing me that direction, right. Yeah. It turns out Cowork builds apps. So why do I need Claude Code?

Garett Medlin (17:09): You don't need Claude Code. Nobody needs it.

Rob Collie (17:13): Every now and then I almost manage to get all of the slot machines spinning at the same time. I got the three tabs in Claude Desktop, right. I've switched to using Claude Code and Claude Desktop. So I'll kick something off in Claude Code. I'll flip over to Cowork and I'll have one or two things churning there. And then I go, "Oh, I really should just flip over to the chat tab and ask it something so that I have all of them going at the same time in parallel."

Garett Medlin (17:38): And they've really kicked it up a notch on the doing things in parallel for the Claude Code recently, you know, their little UI design overhaul that probably irritated tons of people.

Rob Collie (17:50): Well, it did for a moment 'cause I couldn't find my tabs anymore. Like, where the hell, guys? Not again. I mean, at least they didn't rename them. If they'd renamed them, that'd been really great.

Garett Medlin (17:58): So my version of that is Windows terminal.

Rob Collie (18:03): Yuck.

Garett Medlin (18:05): Three tabs, each tab with three panes all going at once if I need them to. Just use the keyboard to switch between them. Now I've got to click on nothing. Could take it too far. Can't do too many things in parallel.

Rob Collie (18:18): I haven't done much analysis on this yet. When I got the Mac, the MacBook, my impression of it in the first couple of days is, oh my God, this thing's battery life is infinite. Who even used a charger? When it runs out of charge, just throw it away and get a new one. It's sort of like that kind of feeling when you first get it.

Garett Medlin (18:35): You're not supposed to do that.

Justin Mannhardt (18:37): What are you doing with this Mac? It died and the new version of Claude Desktop came out today and it works better on the latest-

Rob Collie (18:43): I ran the battery down, man. Like, it lasts that long. But honestly, recently, I don't think my usage has changed that much. I'm charging this thing twice a day. And it's hot now. This thing is really hot now. And I haven't gotten to the point with a Mac yet where I have the equivalent of Task Manager where I'm analyzing what's chewing all the juice. But something on this laptop is like trying to heat the room now. And that's where my battery's going. The battery heat is being transferred into my legs as it's sitting on my lap.

Garett Medlin (19:20): Yeah. It's got to have a battery analyzer just like the iPhone does, I'm sure.

Justin Mannhardt (19:25): This would be a great sketch, by the way, to film you going to the Apple Store and asking if you can recycle your MacBook-

Rob Collie (19:34): This brand new one.

Justin Mannhardt (19:35): ... 'cause you care about the planet, you're not trying to complain or return it.

Rob Collie (19:38): That's right.

Justin Mannhardt (19:38): You're very satisfied with the use.

Rob Collie (19:40): It's very good. I used it for a week, got every bit of value out of it. Now it doesn't work anymore.

Justin Mannhardt (19:47): But almost like you expect it. The battery's gone.

Rob Collie (19:50): Yeah.

Justin Mannhardt (19:51): This was awesome. And I'm thinking about getting another one.

Rob Collie (19:54): And I go, "Is it just the battery?" I'm like, "Are those replaceable? Can I swap them out? I thought those were just built in." I have those Oral B toothbrushes that come with, they're electric and they come with a AA battery installed.

Garett Medlin (20:07): Oh, I got the rechargeable version.

Rob Collie (20:08): Okay. Yeah. Well, you can unscrew the base and there's the AA battery there and you're just like, "I could just replace this. " But it turns out that they rigged it so that by the time you pull this arm back enough to get the battery out, you bend the arm and now your toothbrush won't work anymore. So they kind of fake the AA battery's replaceable and it's literally like a Duracell AA battery, but designed to not be replaceable. So there you go. There's precedent for this, Justin. The battery runs out and you throw it away.

Justin Mannhardt (20:41): Maybe for your Darth.

Rob Collie (20:42): Darth, whatever. Yes.

Justin Mannhardt (20:44): Darth whatever, website-

Rob Collie (20:46): Darth, whatever. Which is probably now something else someone's going to run out there and-

Justin Mannhardt (20:47): We're just spitballing.

Rob Collie (20:49): That's right.

Garett Medlin (20:49): Article ideas. Yeah. I mean, Apple is the king of not letting people repair things, which there've been some improvements on, but it is still not easy. But for someone like me, I always just take that as a challenge. Oh, you think I can't repair it? Well, we'll see about that. It's a motivator.

Rob Collie (21:08): I don't know, man. This MacBook, I can't even find ... There aren't any seams in it. There aren't any screws in it. It's like the whole thing was extruded as one complete device. I just ... Good luck.

Justin Mannhardt (21:18): I have a question, which I was really curious about when Microsoft announced their new multi-model features in Copilot. I think they call ... I had to look it up. They call one council and one critique. I don't know. Have you tried either of those by chance?

Garett Medlin (21:37): I've not tried them, but I'm aware of the concept 'cause Anthropic and others have been putting out articles around all of these different architectural arrangements of agents and how they can talk to each other in parent child or sibling or counsel of elders and all the different configurations. I imagine, based on what you said, what it would look like, but I haven't tried it.

Justin Mannhardt (22:04): Here's maybe the philosophical question, and we could come back to this later after maybe we all go try it or something. One of them is you would ask, like, you'd give a prompt and that prompt would simultaneously go to ChatGPT and Claude.

Garett Medlin (22:20): Yeah.

Justin Mannhardt (22:21): And then you'd get responses from both and you could compare them.

Garett Medlin (22:26): I like that.

Justin Mannhardt (22:27): Or one was like, it kind of does the same thing or it'll go to one model and then the second model will critique the first model and then homogenize the final answer. Is simulated collaboration actually good? Or is this just sort of like hand waving magic? Is there something actually special that's going on here? It feels different than dispatching subagents. This felt like context pollution almost as a risk, but I haven't tried it out yet.

Garett Medlin (23:02): It is very true that there's a measured benefit to the result of having an agent with its context and goal and mission playing something out and then rubber duck essentially that off of another agent. And it doesn't even have to be a different LLM. It could be the same one. Just get a second take on it. And then once they have their individual opinions on it, then either the human in the loop or they themselves also decide, "Okay, what parts of this do I think were better that you brought up?" Just two people collaborating.

Rob Collie (23:38): Even if they're almost the same person. Justin, you know about this. The great, the musical act, really a genre and era defining musical like that. Hanson, these guys were twins, right. I mean, one of them was like, "We should make a song called mm, mm." And the other one's listening to it. He's like, "No, I think it should be mm, bop." And-

Justin Mannhardt (23:57): I love how you started this story with, "Justin, you know about this."

Rob Collie (24:03): You come from the recording industry. You know about the greatness.

Justin Mannhardt (24:07): The greatness of Hanson.

Rob Collie (24:08): Yeah. I mean, it could have been either way. One of them could have been pitching the idea and the other one critiquing, they would have landed probably in the same place. So Dave Gainer does exactly this manually. He federates ... When I say federates, I'm pretty sure he's copy pasting the same question out to multiple LLMs and then cross pollinates their answers and then eventually lands in some place that's a lot better. And he believes in this. If Dave does something on a regular recurring basis, it's because it works. So we can just sort of mathematically assert that this is a good thing, that this helps. In fact, I even, this is a couple of months ago now, I started building a new agent called Concilium, whose job is to automate that process for Dave. And I told Dave, "Look, I'm only going to continue this thing if you're invested in providing feedback, et cetera." And he's really, really, really good about avoiding any sort of commitment.

(25:08): So he didn't commit to being a live beta tester or anything like that, couldn't commit the time to that. I hope he's listening 'cause he's too busy spending his time copying and pasting between all these LLMs. If he wasn't doing that, he'd have time to help me with the app that makes his life easier. So it's just been sitting there, this project. If he and I had put, I don't know, a few hours into this thing, it would be life changing for him.

Garett Medlin (25:36): It's a popular vibe coded idea, right, 'cause there are lots and lots of different models out there and some of them are better than others at different things. Even Claude Opus 4.7 is not better than GPT 4.7 on every possible metric. I mean, I've actually been using Codex almost as long as Claude Code.

Rob Collie (26:02): I've heard from the cool kids recently that Codex is better than Claude Code in recent iterations. This is what the cool kids have been saying.

Garett Medlin (26:11): Better's a strong word.

Justin Mannhardt (26:12): Yeah. OpenAI, they've come a long way with it in terms of a formidable competing product stacks.

Garett Medlin (26:18): Especially compared to cowork, their new app version of Codex is pretty strong.

Rob Collie (26:23): I haven't seen it yet. I know Dave's been texting me about it. Why would you say it's pretty strong? What's its oomph? What's it called?

Garett Medlin (26:30): It's literally called Codex. So they do kind of have a naming problem there.

Rob Collie (26:34): Everyone's got a naming problem.

Garett Medlin (26:36): Yeah, 'cause they have the app called Codex and they have a model called Codex and they have their Claude Code version called Codex 2, I believe. So three different products with the same name.

Justin Mannhardt (26:48): And you need to be on ChatGPT Pro to use any of it.

Garett Medlin (26:53): Yeah. So it's a little harder to get access too, but I don't use the Codex, right, we've discussed. I'm a terminal guy at this point, so it would take a bit of motivation for me to use the Codex app.

Justin Mannhardt (27:06): What's a bit? Can we quantify a bit?

Garett Medlin (27:09): What it would take to get me to ...

Justin Mannhardt (27:10): Yeah. What's a bit of Garett motivation?

Garett Medlin (27:13): Is there someone I care about who would benefit from it more than the thing I'm currently using it? I need someone to benefit from me knowing about it and using it. What's the value of my time in the relationship.

Rob Collie (27:33): The cool kids would smile. That's some good purring.

Garett Medlin (27:36): I'm going to have to get the cat out of this.

Justin Mannhardt (27:40): No, we need that. We need that for the soundbites banks.

Rob Collie (27:45): Well, if you heard what sounded like someone growling, it was not. No one was angry. There was just a very happy cat, very close to a microphone.

Garett Medlin (27:52): Yeah, it is very loud.

Rob Collie (27:53): It's the closest we've ever gotten a purring cat to a microphone.

Justin Mannhardt (27:56): Yeah.

Garett Medlin (27:57): Oh, nice. Setting records.

Rob Collie (28:00): We're back to truce, Marty and I. He's sitting here in the carrier next to me, happy as he can be.

Justin Mannhardt (28:05): That's nice.

Rob Collie (28:05): Yeah.

Justin Mannhardt (28:06): And we've been talking about this a lot. Claude Code, Claud Cowork, several of the things that have happened on the OpenAI side with Codex now. This all to me has felt like closing what I'll call the Open Claude gap. Open Claude came out and was like, autonomous tool use does things, runs on your computer. And the promise of Copilot Cowork even fit that narrative. It's like people realizing I don't want something to talk to. I want something that can do stuff. And so all of those kind of table stakes features are very quickly coming out in all these other stacks.

Garett Medlin (28:48): It's kind of hard to catch up at a certain point. You don't really need to know about MCPs, 'cause it kind of has built into your Microsoft 365 stuff. So you don't have to care about that technical details. I can ask it to organize my inbox, but I don't think my account even has an inbox for it to organize, so probably wouldn't go far.

Rob Collie (29:09): Is the question essentially like, "Hey, Microsoft, what are you going to do about all of this?"

Justin Mannhardt (29:14): I was really excited about Copilot Cowork as well, Rob, just because so many people, especially in their work environment, the AI they have access to is Copilot. I think we're pretty unique in the broad spectrum of like, we can go use whatever the heck we want to, but so many people, like, that's the thing they have. And so it just seems like they're getting these little doses of step change. And I just wonder why not make a more aggressive move of what's behind that?

Rob Collie (29:48): I have a little bit of information that I really can't share. Beyond that, I have nothing but informed speculation.

Justin Mannhardt (29:56): We love good informed speculation.

Rob Collie (29:58): So I literally have no inside knowledge at all about the deal or the arrangement or the strategy behind the cowork partnership. I wish I did.

Justin Mannhardt (30:09): 'Cause it'd be cool.

Rob Collie (30:10): And the thing is, I know people who know. And it just so happens that the people that I know who know things about that happen to be some of the most tight-lipped people that I know. When it comes to cowork, the whole range of options, like, why it's not as good as the regular Claude Cowork, why it's sort of hamstrung, why you can't customize it, why all that. I think all options are on the table in terms of explaining why it's not as good as we hoped it would be. There's the conspiratorial ones, which are like, they don't want it to be too good. As a former employee there, I got to see what the outside conspiracy mill looked like about us over and over and over again, like, attributing to us some wild ass conspiracy.

(30:53): And I'm just sitting there just laughing so hard. Like, no, no, no, no. Y'all have no idea how mundane and stupid this decision is. It has nothing to do with anything that's because of this other thing that you wouldn't even believe. Even decisions that I personally made occasionally would make it out to the press and be interpreted a certain way. And I'd be like, "Oh, no, no, no, no. That is not at all what happened." But it's in play, right. That they don't want to seed the sense of and the center of gravity of durable value. You don't want that accruing to another software company, another tech company like Anthropic, over the top of your domain, like Office and O365. This is one of the ways that empires can fall, right.

Garett Medlin (31:43): Like, competing with yourself.

Rob Collie (31:45): But again, I haven't war gamed that out. I'm sure if I sat down for a day and just thought through it all, like, really, really deliberately, I could probably figure out where their lines should be in terms of what they're willing to do, but I haven't done that. I've been plenty busy with other things rather than trying to sit on their side of the table and figure out what they're doing. But it's possible that it's for those reasons. It's also possible, Justin, what you and I were talking about before you even came out was just like, if they turn this thing on in any real sense, it's going to cost so much on tokens on the back end. The people who adopt it, they're never going to fit-

Justin Mannhardt (32:20): Engineering choice.

Rob Collie (32:21): 100% engineering choice. It could be some sort of fundamental engineering challenge like, "Oh, if we start making people," blah, blah, blah. There's something about if it has to go and read the Claude MD file every time something else gets more complicated. Wide range of explanations, I think they're all on the table. I think it's sort of like this Schrodinger's probability cloud of where the real answer is. We just don't know and I don't want to be-

Garett Medlin (32:45): What's the common thread, right. It's all about their idea of how to avoid risk. That's what it feels like to me.

Rob Collie (32:52): Well, there's different kinds of risk. Like, there's financial risk, there's technical risk, and then there's product erosion risk. And the juiciest theories that people are going to have, the most conspiratorial theories are going to be the ones that are around product erosion, like, value erosion.

Garett Medlin (33:09): Yeah. If we push cowork too hard, then all of Office goes away because nobody wants to use the Copilot in Office, but if we push Cowork too much, then they don't need Office.

Rob Collie (33:20): And again, I'm not saying that I even necessarily believe that's the case, but it's certainly on the table as a possible explanation.

Garett Medlin (33:27): But I know what's not working, which based on at least two or three friends, I know their bosses have told them, "You must use Copilot at least once a month, and we are tracking that that happens." And I'm like, "Talk about getting people to not adopt."

Justin Mannhardt (33:46): I have a message for these bosses. Get bent.

Rob Collie (33:54): They might be acting totally irrationally. They're like, "Look, my boss ..." So that boss is saying, "My boss is telling me ..." Okay, here's the funny one. I went to Vancouver this weekend. I shouldn't have. I originally was going to run in this 10K.

Justin Mannhardt (34:08): Oh, 'cause your illness.

Rob Collie (34:10): I'm not running any 10K. I'm not running a 1K. I'm not running a one.

Justin Mannhardt (34:16): And I'm not running.

Rob Collie (34:17): No.

Justin Mannhardt (34:18): Rob will be walking everywhere.

Rob Collie (34:20): That is correct. Even though I was not feeling great, I did take the opportunity each night I had two Negronis. It was glorious. So one of those nights, I'm sitting next to a friend of ours. He's saying to me that at his organization, the edict has come down that IT needs to spend $10 million a year on AI. I immediately say, "Oh, we could help." No, no, no. Actually 20 million a year that they're being told that they need to spend and none of them know what to spend it on. Of course they don't.

Justin Mannhardt (34:58): Tokens, obviously.

Rob Collie (35:00): It's this top down edict from people who know AI is important, but don't understand it, don't understand what to do with it. So they just make it someone else's problem. It's the same thing again, like asking this other person, "What are we going to do about AI?" That's the way that you seize the moral high ground. If you're the one asking the question, the other person now feels like they're compelled to have an answer. Nevermind the fact that the person asking the question has no idea. So this edict's coming down from [inaudible 00:35:26] and I'm saying, "Look, none of them know what to spend it on. IT certainly doesn't know what to spend it on-

Justin Mannhardt (35:31): Of course not.

Rob Collie (35:32): ... and it shouldn't be IT." IT should be supporting it, but this is like the BI thing all over again.

Justin Mannhardt (35:38): But weirder.

Rob Collie (35:39): It's so weird. It's weirder. It's more intense. It's like we've seen this movie before, they just, like, crank the volume.

Garett Medlin (35:45): You thought Excel spreadsheets and access databases were bad. Well, look out.

Rob Collie (35:49): And I'm making the joke with them. I'm like, "Oh, you should totally hire us. We'd actually help you figure out what to do with it." Then I go, "Oh, the problem though is that we'd make it very, very difficult for you to spend 20 million a year," 'cause that's the goal.

Justin Mannhardt (36:00): I don't know. I could think of a few creative ways to make the 20 million disappear, but ...

Rob Collie (36:05): Is this like Brewster's millions where you have to spend the money or if you can't spend the 30 million in a certain amount of time, you don't get the 300 million? You can't hit the spending targets if your people aren't using it, right?

Justin Mannhardt (36:18): This is a connective tissue, I think, between these kind of dilemmas is you could go back to BI, you could go back to stuff before BI. Throughout the evolution of technology, we've always said the equivalent of start with the business, start with the value, whatever you want to use. But I think it was easier to not necessarily do that with dashboards, because it was much easier to envision the in state. It was much easier to envision, "Oh, I've got a dashboard with slicers and filters." And I could go make something. It might not be the right thing 'cause I went about it the wrong way, but I could accomplish the deliverable. But AI is just so weird 'cause like you were describing that story with Kristy is like, you can spin up an app to make this bespoke process about artwork for your book that you'll do one time through the next however many weeks or months and then it's done. And so people are being told, "Use AI."

(37:14): And so then they're being given tools like Copilot that are kind of lagging or they don't have the full feature set to take advantage of it. And it's sort of an interesting struggle that I keep hearing about. People are like, "We have the tools and people kind of use them, but we're just not quite sure what to do." And so I'm really fascinated with this problem.

Garett Medlin (37:34): Even the training, even when they pay for someone to come out and do an all day training for eight hours, this happened to one of my friends.

Rob Collie (37:42): Oh, even worse.

Garett Medlin (37:44): Every mistake you could have thought to make to not get someone interested. I think four hours was spent on the inner workings of how AI and LLMs ...

Justin Mannhardt (37:53): We can't talk about the magic of creating the application idea you've had for years until you understand the fundamentals of neural networks.

Garett Medlin (38:02): Literally. Yeah, yeah. I couldn't believe it. I don't know what I'm authorized to say, but they got their money back from that one.

Rob Collie (38:08): Well, they should, right. Okay. So I've heard two stories recently that both made me hold my head and my hands, but at the same time going, "Of course." Someone contacted me on LinkedIn and said, "You've been talking about cowork and you've been talking about the Chrome extension. I've been listening to this local AI guru and this local AI guru has been telling ..." In our last meeting at the get together, she told us, "From my analysis, cowork is not safe yet. Cowork is a security risk and certainly the Chrome extension is, but stay tuned because I've got something next time we get together that will address this." This woman is lying to the audience in order to set up selling a product or selling a service.

Garett Medlin (38:50): Did she create Copilot Cowork? Probably not.

Rob Collie (38:53): So she's going to have some magic answer for, like, "If you have this talisman sitting on your desk that you paid for from me-"

Garett Medlin (39:01): The snake oil.

Rob Collie (39:02): Yeah, "it will make you safe from the hobgoblins in Claude Cowork." Good Lord. Claude is so security ... Like, you guys, I'm sure you've noticed this, that every first sentence from Claude these days is, "Okay, this is not malware." It's taken a look at the thing that you've given it and decided ... I didn't ask you to decide if it was malware, but no, it's like its first step all the time.

Justin Mannhardt (39:25): Oh, yeah. Big update in the system prompt.

Rob Collie (39:27): Yeah. "Please enter passwords for me in the Chrome extension." "No, I will not do that." "God damn it."

Garett Medlin (39:33): I've literally tried to force it to delete all the files on my computer just to see if it would actually do it.

Justin Mannhardt (39:38): He lives on the Frontier.

Garett Medlin (39:40): That's right. I have back up.

Rob Collie (39:42): I want you to delete all of my black powder for my musket. See if I can still survive. Anyway, so that is a scammer using fear and techno jargon and babble to control an audience. It's gross, so gross. The other story was a friend of mine was in Hawaii and was just hanging out with some person who works in cybersecurity. This person who works in cybersecurity has recognized that AI is coming for his job, his new shtick, is selling hardware boxes, whole computers, with local open-weight LLMs on them so that none of your data leaks to the cloud.

Justin Mannhardt (40:39): You mean my data that's already in the cloud?

Rob Collie (40:41): Yeah, yeah. Okay. Keep going, right.

Garett Medlin (40:45): It's probably cheaper than the 20 million tokens, but that's about the only advantage.

Rob Collie (40:50): There's multiple things wrong with this, but one of the things that he should advertise is like, "Hey, this is the suck box. This one's going to have an LLM on it that's not going to be nearly as good as that other stuff."

Garett Medlin (41:02): That's when you sell the paid upgrades, it'll get better over time.

Rob Collie (41:06): And at what point is that one box he's selling me for like $4,000, at what point is it ever going to have 16 GPUs in it? When I say, "Hey, go review this chapter for me," I need the biggest, baddest monster that's available to me to go analyze my work. I would never, ever, ever, at this point ... I wouldn't even downgrade to Sonnet 4.6 or something in this enterprise. I want the absolute best thing possible helping me because there's just such tremendous leverage on this. This box, to me, for multiple reasons, first of all, there's so many ways to make sure that it's not retaining your data. If you really want to get crazy with it, you can go to the Azure version 'cause everyone's already trusted Microsoft.

Garett Medlin (41:56): It's like the Fisher-Price LLM for this extremely security paranoid.

Rob Collie (42:03): Honestly, for the ignorant, it's not cheaper either, right. At $200 a month plan, take me 20 months to pay for this box. There's just no situation in which this is advantageous to anyone other than the person selling it.

Justin Mannhardt (42:15): I think it's a very small number of people. I follow some people that they run Gemma 4, which is the latest open source that ... But they're running it, Garett, on machines that have an insane number of integrated memory. These are not $4,000 computers. These are like $16,000 computers and they're burning through the equivalent of hundreds of thousands of dollars in token costs, but these are very advanced technical people that are interested in doing ... That is not the every person at all.

Rob Collie (42:49): This guy is selling to the uninformed.

Justin Mannhardt (42:52): Yeah, don't do that.

Rob Collie (42:53): It's the same thing. He creates a fear and then shows up with the antidote. Just like the woman who's controlling this user group, wherever it is.

Justin Mannhardt (43:04): Gross.

Rob Collie (43:04): Create a fear and then you be the one to solve it. How about solving a real problem? That takes more effort, doesn't it? It takes more effort, more courage, more originality, things that both of these human beings lack.

Justin Mannhardt (43:17): You rip the words right out of my brain, Rob, about solving real problems 'cause the tools ... Stuff like this, whether it's trust, security, or new capabilities with AI where, "Oh, it's maybe not so good at this thing, so I'm doing this integration kung fu." And then a month later it gets better at that. If there's incentive for Anthropic or OpenAI to resolve that perception, they will do it. If there is a real security concern, they have incentive to not have that be an issue.

Garett Medlin (43:50): Although, what's the number one security risk? People can't solve people pasting their passwords into OneNote or whatever. That's a real hard one to solve.

Justin Mannhardt (44:03): I think trust in AI will come as did trust in the cloud. We've already learned from social media that people will happily hand over their individual data in exchange for a good product experience.

Rob Collie (44:17): It's like, "Oh yeah, now you're taking a principled stand?" You gave away our country so that you could have Facebook.

Justin Mannhardt (44:27): My wife said something the other day I was like, "Listen, honey, we signed up for Facebook when you had to have a .edu email to sign up for it."

Garett Medlin (44:39): I do have some good use cases for local LLMs though, if that was of interest.

Rob Collie (44:44): Let's come back to that, 'cause I neglected to mention two things earlier in the Microsoft department. First of all, I have had some conversations and I will tell you that, can't get into any specifics, but I came out of the conversations over there recently going, "Oh, okay. Okay. Yeah, I see it. Okay." I went into those meetings with this impression like, "Man, Anthropic is just eating everybody's lunch." And I came out going, again with the, "Do not. Do not bet against Microsoft long-term." In the old fight club, on a long enough timeline, everyone's survivability drops to zero. On a long enough timeline, Microsoft always wins. Do not count these people out. They still have amazing field position. They figured a bunch of important things out that they hadn't figured out for in the early going. I kind of think they have more than enough time to turn the corner.

Garett Medlin (45:52): I don't think they're going to lose either ... I don't know that they're going to knock it out of the park as fast as we would want them to.

Rob Collie (45:59): I walked away from these conversations, again, feeling proud to be part of the Microsoft Wii.

Garett Medlin (46:06): Well, that's good.

Rob Collie (46:07): Again, right? We're talking about Microsoft.

Garett Medlin (46:09): Did they know you bought two MacBooks though? Did you tell them that?

Justin Mannhardt (46:12): There's MacBooks all over that campus.

Rob Collie (46:15): I don't know that I told them that, but yeah, I don't think that would be uncool anymore. I think that'd be fine.

Garett Medlin (46:19): Okay, cool.

Justin Mannhardt (46:21): Well, he's going to give them back since it works fine on Windows now.

Rob Collie (46:24): Yeah. Well, no, the battery died, Justin.

Justin Mannhardt (46:25): That's right. You're going to recycle it at the Apple Store.

Rob Collie (46:31): It's like the toothbrush. I'm giving you the MacBook and my toothbrush. But here's one thing that came up in those conversations that is just really just completely 100% open source knowledge that I just had never really tuned into. And it might be, again, even for both of you, like, "Oh, really, Rob, that you hadn't really thought about that?" It could be that obvious to both of you, but I've been thinking about Microsoft in terms of their home field advantage. They just have so much of your stuff. They have so much of your data, so much of your content. They also have so much of your application use. They have so much of also even the inertia of your users who are used to using SharePoint, who are used to using Excel, who are used to using all of these things. They have all of that.

(47:20): In our internal analysis, it didn't even make sense for us to code our own full replacement for Salesforce. The ROI tipped towards now only do a partial replace, right. Imagine if that plays out for us with 50 employees and something as really as simple as Salesforce. Imagine how durable something like the entire Microsoft ecosystem is in some Fortune 500 IT's operation. It is not coming out, folks, but there's an additional home field advantage that I think might be worth every bit as much as that, and I haven't been thinking about it, which is the trust boundary. This whole notion of data in the cloud and all that kind of stuff, it's an irrational thing, but it's so strong. The cloud where we're going to go to the cloud and, "Ah, it's just scary. People have my stuff." I'm like, "Yeah, but your email's already in the cloud." And they didn't know they were already there.

Justin Mannhardt (48:26): And arguably, that's easier to attack than the packets going to and from an LLM.

Rob Collie (48:31): But IT shops, organizations have made their piece like, "Okay, look, we will trust Microsoft or we will trust AWS." They make their deal with one devil. So now the biggest reason why Anthropic, for example, needs Microsoft is because Anthropic needs to run their stuff inside of Microsoft's trust boundary. It needs to run on Azure. It needs to run as part of the Microsoft ecosystem.

Garett Medlin (48:58): You need the private network.

Rob Collie (49:00): 'Cause doubling your number of trusted clouds is a terrifying move for an enterprise.

Justin Mannhardt (49:09): 1,000% is correct. I was talking with a CIO a couple of weeks ago about the same thing. It's hard for me sometimes to qualify the dilemma I find myself in here with what's going on with Microsoft's AI products. I agree with you, Rob. I find it very hard to believe they're going to come out a loser in the end. They've proven they can be, if they need to be a fast follower, they can do that. You don't want them in your space 'cause they can spend up a billion dollar R&D department tomorrow and figure out what's going on. I've always been perplexed by this sort of feeling of under delivery lately, you call it Copilot Cowork. So that sets an expectation relative to Claude Cowork.

Garett Medlin (49:54): Yeah. It's built in comparison, which is not ideal.

Justin Mannhardt (49:59): I've been talking to people like, "Oh, could you help us build a solution that could do this, that?" And they're like, "We absolutely could do that." "Oh, we got to do it with Copilot." It's like, "Ah." For me in my career timeline, it's interesting 'cause I came into the last era with Power BI and everything. It's like, "Oh my gosh, I have everything we need to do, something that's awesome inside the Microsoft four walls." So hopefully that story starts to change.

Garett Medlin (50:24): 'Cause they have come out with things that eventually didn't work and they stopped developing them, but it was nothing as monumental as AI. I mean, virtual reality, they tried phones and other things. Even their Microsoft Loop, which some people think is the better version of OneNote hasn't really gotten adoption and features within Power BI and Fabric that that happens to, they eventually kill them, but they know they can't afford to do that with AI tools, I'm sure.

Rob Collie (50:56): We mentioned you should do a product and a service known as Microsoft Bing.

Garett Medlin (51:00): Yeah. That's a good example of something that has stuck around.

Rob Collie (51:07): They just were never going to seed that front. Wasn't going to happen.

Garett Medlin (51:13): Every time I reset my Edge browser settings, I'm like, "Oh, okay. I got to change this default search provider again."

Rob Collie (51:22): I mean, they stuck around in that market long enough for that market to matter less.

Garett Medlin (51:28): That's what I'm worried about. Yeah.

Justin Mannhardt (51:31): This is, I think, the difference. There are some really cool things in the Microsoft ecosystem. I think Copilot Studio has gotten a lot better. I think Foundry's pretty capable product, but those aren't at the same kind of consumer approachableness as something like cowork at all.

Garett Medlin (51:49): Right. Or like infrastructure.

Justin Mannhardt (51:51): Well, hopefully there'll be some pretty exciting announcements and feature releases soon because I think, listen, companies aren't going to turn around tomorrow and say like, "Oh, we heard that Copilot isn't the best, so we're going to go just wholesale switch over to something else."

Garett Medlin (52:06): Yeah. It'd still be their first pick regardless, it seems like.

Justin Mannhardt (52:10): Most people I talk to that are responsible for that decision, the trust boundary point from you, Rob, it's just spot on. I want the goods. I want my skills, I want my connectors, I want my memory, I want all those things.

Rob Collie (52:26): What kind of skills? You want acquired skills? You want ... 'Cause those aren't here yet.

Garett Medlin (52:32): Can we just get the Microsoft certified sandbox version of the thing that we want? How can we get the best of both?

Justin Mannhardt (52:41): This dilemma, Garett and Rob, I keep coming back to this idea lately of the last few weeks. You got to remember, it's not just about the models. It's about the whole system that surrounds it. And like you were talking about in our last few episodes, Rob, the engineering they might be doing to control token costs or what's in the system prompts or the application interfaces themselves. It's not like you just, "Oh, cool. We have the latest Anthropic or OpenAI or whatever model, so it should work great." All these other pieces around it need to work well too. I mean, even the connectors, you can't use your email and cowork, Rob. So that's frustrating.

Garett Medlin (53:22): It's a very complex software product, multiple products, so many moving pieces and parts in them to make it all work.

Rob Collie (53:31): Jocelyn's cowork adoption's caught. It's like that multiple throw-in matches at the fire thing, but she's telling me yesterday, "Well, there's this plugin or something and it'll do Outlook web access. It'll access your email that way via Claude Cowork." I'm like, "Oh yes, this is what I'm talking about."

Garett Medlin (53:50): I might try that one, 'cause the email organizing, yeah, my wife is, I'm more of an inbox zero and she's more of an inbox infinite, so that could be a good hook.

Rob Collie (54:01): What do you think my unread count is in my inbox right now?

Justin Mannhardt (54:06): I would predict that you have reached the hard cap in the software of presenting the number.

Rob Collie (54:13): Well, yeah, I mean, it claims I have 511 unread, but we know that's not right. It's aged off the bottom.

Justin Mannhardt (54:20): Yeah. Like, the time to live has expired.

Rob Collie (54:27): I have 351 messages in drafts.

Garett Medlin (54:32): I couldn't live that way. I don't know. I don't know how you do it.

Justin Mannhardt (54:38): Different wiring, man. Just different wiring.

Garett Medlin (54:41): I see one little red dot on my task bar and I'm like, "I'm going to get rid of that thing immediately. Immediately."

Rob Collie (54:48): People are sensitive about different things. Jocelyn will have 400 tabs open in Chrome. Each tab is like four pixels wide. We were on a meeting with someone yesterday, the other meeting I did this week and someone was sharing their screen and immediately I see the hundred tabs across the top and I just start laughing. I'm like, "Oh, you guys are the Jocelyn school of Chrome tab management."

Garett Medlin (55:14): You don't need bookmarks if you just always keep the tabs open.

Justin Mannhardt (55:18): I worked with a guy, he was like that and I told him, I was like, "That would stress me out." And he's like, "Well, Justin, here's how I think about it. Each of these little one by one tab, these are all the little mountains that I need to climb today." And I was like, "Sure, good for you, man. Climb the mountains."

Garett Medlin (55:36): That's too many mountains.

Rob Collie (55:37): Just to bookend it, like, what does he do when you close all of his browser tabs and he can't get them back? That's when he takes the computer back.

Justin Mannhardt (55:47): The browser went away. I really enjoyed this computer.

Rob Collie (55:50): Ran out of browser.

Justin Mannhardt (55:57): Oh, man.

Rob Collie (55:57): Lost all my favorites. What favorites? Yeah, you know what I mean? Well, gentlemen, I think I'm going to go rest again. Doesn't that sound great?

Garett Medlin (56:05): All right. Rest well.