Pratik polled an entire airport bar on his way to give a talk on AI ethics, and every single person told him they don't use AI. Chris and Pratik spend this one on harnesses: the thing wrapped around the model that decides whether you ever actually use it. Pratik defines the term, then walks through Meta's new Muse and its cloud-hosted computer, while Chris reports back from a week with the new iOS 27, where Siri has quietly taken over a chunk of what he was paying ChatGPT for. They also land on the same answer to the daily driver question, which almost never happens. In this episode: - Pratik used four different AI tools before 11:30 on a Monday morning and still calls "whatever fits the problem" a cop-out answer - The one thing both hosts agree on: for most people, the twenty dollar ChatGPT is the entire answer - Chris rebuilds his glucose monitor workaround as an on-device shortcut and stops burning tokens to read his own screenshot - Muse negotiating loyalty discounts on two of Pratik's subscriptions, with the support rep on the other end assuming it's him - The Slack workspace ChatGPT created for Chris that he didn't find until the next day - Why coding got product-market fit and nothing else did: forty years of QA baked in, versus a feeling Chapters 0:00 Intro 1:20 Dreamforce and the bar where nobody uses AI 6:57 From the Dash: what should your daily driver be? 16:33 What a harness actually is 23:04 A week with the new iOS, and what Chris stopped paying for 29:37 How Slack quietly democratized AI at work 38:10 Meta's Muse and its cloud computer 45:06 Lockers, credentials, and giving up control 51:54 Why coding works and everything else is a feeling Send us Fan Mail
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
SPEAKER_04Claud Claud in Chrome was really good. Chat GPT computer use was really good. But there's always those one or two things that it runs into, right? I'm not seeing those instances with Muse. With Muse, if it runs into a roadblock where it needs you to do something or take over, it actually flags it to you and says, I just need you to move this slider. This actually happened to me this morning.
SPEAKER_00Please welcome your starting lineup for the Teak and Chris Podcast. It's Christopher Long and Pratique Desai.
SPEAKER_01Hello and welcome to episode 14 of the Teak and Chris Podcast. My name is Christopher Long, and joining me today is my friend. I'm going to speak on behalf of everyone and say it's going eh.
SPEAKER_04Yeah.
unknownYeah.
SPEAKER_04A lot of news. A lot of the news to digest. Yeah. And a lot of conflicting news to digest. So I get it.
SPEAKER_01Um I mean, speaking of news, that is just the easiest segue, is just to say speaking of and then plug in thing and then make uh an unrelated connection. But speaking of news, uh, you were just at a big event uh over last week. Is that right?
SPEAKER_04I was. I uh had the pleasure and uh privilege of attending um Dreamforce, which is Salesforce's biggest conference. Some would, I think, argue that it's probably the biggest technology conference pretty much uh in the world. So that's cool. That's exciting. Um I've heard of that. I've heard of that. You you might Chris, Chris, you might have heard of it. You might have actually attended in the past. Um we did miss you. We did talk about how uh your presence or lack thereof was was felt. So um uh we I think the the one story that I want to share here uh on this on this podcast is I I was actually doing a talk on AI ethics.
SPEAKER_04Um and I I think just that you know one that's super relevant, super timely. Uh but ironically, I had planned that like four months ago and just could not have timed that better. Right. Um so I as I'm doing this talk, I'm preparing for it. Um I'm sitting in the lounge uh for my flight to San Francisco, and I'm sitting at the bar. There's a bunch of people at the bar, uh business people, people traveling for personal reasons and presumably also professionals. Uh, there's a few bartenders um who are going to school.
SPEAKER_04So I'm like kind of seeing a diversification at the bar, and I'm like, does my talk track actually resonate? Especially at a technology for uh conference where there's gonna be people from all walks of life. So I look at the bar. Hold on.
SPEAKER_01We I have to acknowledge something.
SPEAKER_04Yeah.
SPEAKER_01I I I need everyone listening or watching to understand that this is you. You are that person that will talk to strangers at a bar. Okay. Is that normal? Uh I wanna I don't want to say it's not normal, but it is certainly not everyone. Um, so just congratulations on a life well lived and being able to talk to other humans in any moment.
SPEAKER_04Fair enough. Actually, it is a as an anecdote to the anecdote. Um, we we were at a wedding, my wife and I, this weekend, and we were sitting around the campfire and we're sharing how we all met. Um, and somehow we were in a group of people who had all met, you know, the the the campfire was going around. We all met IRL. We all met IRL. Um, so we were somehow it it was just felt so rare that we were in a in in a group of people who had all met in real life. Um so it's just uh it's an interesting uh take because you're you're you're pointing that out because it feels so weird.
SPEAKER_04Um to me, certainly. It feel it feels so weird, and and I feel like it shouldn't feel weird, but I I understand why it feels weird. So um I get that. Uh so going back to the first anecdote, right? I um I I just speak up and I'm like, hey, I have to do this talk on ethics for AI. I would love for for folks to kind of react. So literally the entire bar just stops talking and they all just focus on me, which is what I wanted, right? So it was great. And I do the talk and I went through it really quickly.
SPEAKER_04So don't think I was like asking everyone to listen to me for 45 minutes. It was only 40 minutes. No, I'm kidding. It was I went through it really in 10 minutes. Okay. And when I asked them, like, how do you feel? What do you think? And they're like, oh man, like, you know, it that addresses a lot of things we're seeing on the news. And I'm like, no, but like in your day-to-day usage, does this resonate? Does this resonate with how you're seeing AI manifest in your life? And the one common thread across the bar was, oh, we don't use AI.
SPEAKER_04And I'm like, what do you mean we don't use AI, you don't use AI? Like, I don't mean like crazy use cases. I mean like your day-to-day usage, you know, managing calendar and nothing. Right? Every single person on that bar was like, no, we don't, we don't use it, man. Um and I just found that fascinating because you and I have acknowledged, like you and I are living in a bubble, right? A super user bubble of this particular technology. Um, we don't know where that's gonna go, we don't know where that's gonna lead to, but we know we're in a bubble.
SPEAKER_04Um, I think it was a little bit of a slap in the face on how much of a bubble. I mean, again, there was business people at this bar who are like, yeah, we yeah, I mean, I've seen the button at work, but I don't use it. And and I just I I don't know, I found that fascinating. Um especially because we a lot of what we talk about is the acceleration of the technology and how fast it's doing things and capable of doing things. And then you have folks that have just never never clicked the button, which is amazing.
SPEAKER_01And very true, very common. We absolutely being in tech industry, uh, are living in some kind of special place um that most people do not spend much time.
unknownRight?
SPEAKER_01That is just that is just the case. Yeah.
SPEAKER_04Yeah. Yeah. The the other um the other item I'll share from Dreamforce was uh Dario um shook shook Benioff's hand and they had this awkward exchange where Benioff looked at at the at the team behind Dario and was like, oh yeah, that's great. You know, I I love that your team is really just all of the people that I trained for the last 20 years, and then you've been hired. So Dario thanked him and they they did this awkward back and forth. Um, see, you say awkward.
SPEAKER_01I watched it online I and I thought it was hilarious.
SPEAKER_04Um hilariously awkward, right? Okay, sure. It was hilarious because of how awkward it was. Um but this is this is the the tech industry. Um but again, it's it's going back to this idea of a bubble, right? Like yeah, it's this bubble of people moving back and forth just solving all these technology problems, um, and then us talking about it. So yeah.
SPEAKER_01Well, I mean, speaking of people in the bubble, we have a new segment. Uh, we're launching it today. It's called From the Dash. I want to say that one more time. So when I do the edit, it sounds really cool. From the dash. And basically, we have our little dashboard where you can ask us questions. And even though we're calling it from the dash, really, it's any comment you make about you know our topics in any of our channels. Like so we have TikTok. If you put a comment in there, we could pull that. Um, if you put a question straight into our website, we could pull pull that.
SPEAKER_01If you're talking to us on YouTube, we could pull that. Um, but today's from the dash question, and these are all anonymous too, at least if um you're putting it through the website, it's anonymous. Today's question is what AI system can be my daily driver as a technologist professional? So, I mean, right before the podcast started, like you were already joking that this may be a hard question to answer because it seems to change frequently.
SPEAKER_01But why don't you give us your take on it, pratique?
SPEAKER_04I think this is hilarious because I change my daily driver every 15 minutes. Um so every 15 minutes, I'm like, oh, I have a new daily driver. Um, and it's just because it this technology is moving so fast. I mean, I'm half joking on the 15 minutes. It's not really every 15 minutes, although today I switched back, but just today, it is 11:30 a.m. on September 21st. I have used four different technologies. I have used Claude to run skills around my expense reports.
SPEAKER_04I have used ChatGPT uh to interface with an article that I'm writing because I believe it's better at uh giving me feedback. I have used Muse to go solve a problem that I have personally that we can talk about later. Um and then I have also used Gemini to work on some images that I want to um massage before loading them up to LinkedIn. So, man, uh what is your daily driver? Uh the the best technology for the problem in that minute.
SPEAKER_04I I mean I hate I hate to give that answer. That's cop out. That's cop out. It's such a cop out.
unknownYep.
SPEAKER_04It's such a cop out. Um I'll try to take a shot at it in in terms of like what I think is my average usage per use case. How about that? Okay. Okay. If it's if it's to do with images, it's Gemini. I mean, I think your deli driver is Gemini. I I I do think Google has such a head start on images right now that like generally for the most for the masses, your Gemini, uh, your Gemini harness is going to be the best for images.
SPEAKER_01This is already hilarious because I think we're already at a difference of opinion. Completely different opinion, yeah. So so images, what else?
SPEAKER_04And notice I said harness, right? Like, yeah, don't don't don't go use a Gemini model on O Lama and expect results that the harness is really ultimately creating, right? Like it's it's the harness leveraging the model that's ultimately the the right answer here. Um I think for general purposes, I'm going to ChatGPT, right? If you if a new use case pops up that I don't have an answer to, I'm going to ChatGPT. If a use case that is a is a general purpose um uh problem, I'm going to Chat GPT.
SPEAKER_04And so I'll create the dichotomies with the rest, right? So the the why would I go to Claude? I think where I go to Claude is just things that I do often, over and over. Um, largely because not that ChatGPT can't do things like skills, they have their own version of it, but that's just where I ended up creating my skills because that's where it came out first. And so that's that stickiness factor, right? I created my skills, for example, expense report. It knows that if I point it at a folder and say, this is an expense folder, it's going to take a skill, open that folder up, read through all the expense uh receipts, rename them, then fill out an expense report that it knows that that I need for my work, and then give me an email that I can send to describe what the report is and request reimbursement.
SPEAKER_04Like that skill already is created. I just haven't moved it anywhere. I don't really know why I would move it, right? I just I could just use Claude. So if I need a skill based, I'm using Claude. Uh if I'm doing coding, I still think Claude Code is probably the best, especially because of the the teaming aspect that still Claude has had lead on. Um codex is getting good. I I do I do A B test Codex. It's getting really, really good at it. Um but Cloud Code is is where a lot of my repositories are connected.
SPEAKER_04A lot of things that I created in the last six to nine months was originally created in Cloud Code. So I do a lot of continuation there. Um again, that stickiness stickiness factors. As I'm saying this, by the way, I'm realizing that I'm explaining my claud usage with stickiness, and that starts to make me think subconsciously I'm moving away from Claude. But anyway, um then where does uh where does this new to new tool that I want to introduce come in? Muse. Um I'm starting to move a lot of things to Muse now, and and not everyone has access, but with Muse having computer access, where it can go off and like run actual uh prompts where it can take action by loading up a website and doing things.
SPEAKER_04Remember, a lot of the internet is not designed for L Webs, right? Um, a lot of one of my favorite features in Claude was Claude uh desktop usage or you know, claude in Chrome browser because it could actually take the internet, browse the internet, and act as if it was you trying to take action. So for example, I had Muse um go off and opt me out a bunch of uh spam caller websites, right? Um I had it go through my entire Amazon account and say, like, go through, click through, because Amazon doesn't give you the ability to export your orders, go through and find all the areas where I should have been refunded and create an Excel sheet so that I can I can make sure that I was actually refunded all those things, right?
SPEAKER_04Um so much of the internet's not created for LLM consumption right now or agent consumption. And Muse, with its ability to go and act like a human as much as possible, I think is is a little bit of a game changer. And then last but not least, because I spend a lot of time on a plane, I have to tell you what I use on a daily driver when I'm on a plane. It's it's the free messaging where you can click free messaging, get it on any plane. Um, you can then use WhatsApp. Well, I used to use Meta AI on WhatsApp as my daily driver on a plane, uh, but now you can access Muse through WhatsApp, where on a cloud-based server it has access to a computer.
SPEAKER_04So essentially you get free internet because I'm just using Muse through WhatsApp, which is free messaging, and then it's going off and browsing literally the entire internet capability. Um, I would say that's my that's my answer, but again, I'm gonna preface September 21st answer. Might change in about an hour. It's also not an answer.
SPEAKER_01I think it's an answer. What's your answer? It's uh I'm gonna answer the question. The best daily driver for most people is the$20 chat GPT.
SPEAKER_04Fair.
SPEAKER_01I think I it is that is it. That's that's the answer. All other answers are incorrect.
SPEAKER_04So the I I actually, you know what? I usually tend to not agree with your disagreements. Yeah, I I'm gonna I'm gonna fall on the sword. I actually agree with you. I th I think that is the only answer.
SPEAKER_01Yeah, I I the capabilities for what you get for$20 is fantastic. Um, it can do the computer use things that you're talking about. Like this weekend, it helped me set up some Slack stuff, and it did everything like in the desktop desktop Slack app for me, and was going to the web, and I just watched it bounce back and forth between the two. It like it was just doing all the things, clicking all the buttons. Now, again, for the$20, like I couldn't do that very much.
SPEAKER_01Like you you kind of get an hour a day um or an hour per your five-hour limit on that twenty dollars if you're doing computer use or something like that. But for everything else, just asking it questions and getting answers, making spreadsheets, you know, when you connect your Gmail, it's a lot of people have Gmail, right? Or Google stuff. You can connect it like directly, um, and it will just start to help you with your email. Like nowadays, it even gives you prompts when you open up ChatGPT. Like you, like it'll just say, like, respond to this email because it seems important.
SPEAKER_01And I haven't said do anything with my email. Um, so just for the value for the people who are not super users, I think that$20 is fantastic. Um, my journey over the past couple years started with ChatGPT as a daily driver, then I went to Claude as a daily driver. Uh, now I'm back. ChatGPT is a daily driver, but the interesting part to me is you mentioned harnesses.
SPEAKER_01Uh, the new iOS came out, and this is going to lead us into our topic of the day. And I'm already switching a lot of the things that I lean on ChatGPT for. I'm already handing those over to Siri on my phone. So let's get into our topic of the day, which really is harnesses and where are they at right now? How are they changing? Where's the value? We want to hear about Muse because it's new. I want to talk about my experience with iOS and then kind of ask you some comparison questions with Android and things of that nature.
SPEAKER_01So critique, please tell us about harnesses.
SPEAKER_04Yeah, so I think for I'm gonna try to, and you know, we've we've received some feedback that we try to uh water down some some verbiage. So I I want to address that before I actually give you a definition because I I know the feedback has been no, don't call it Apple and Android, call it Kotlin and Swift, for example, on a previous episode. At the end of the day, like if for the folks that know it's Kotlin and Swift, you know what I'm talking about. And for the folks that don't know it's Kotlin and Swift, if I use that terminology, you wouldn't know what I'm talking about.
SPEAKER_04So um, you know, I want to address that because I'm about to give a definition where I do think I'm gonna use some terminology that will upset some folks that are purists. Um but just know that I'm I'm trying to I'm trying to appeal to the masses and I want to make sure the definition makes sense. So um I think if you if you dissect really any um any technology out there that is using what we call AI, right? I think colloqually we all understand what I mean when I say AI. Um but there there is a there is an underlying model, which is the large language model generally.
SPEAKER_04It's a large language model that is the artificial intelligence that we're calling that is, you know, at it at its core, it's just trying to guess the next best character. When I say character, it's literally just saying, you gave me this question or what we would call prompt, and I'm going to respond by trying to guess a series of characters that actually make sense for your prompt. And I'm just doing that based on a ton of training data, based on a ton of training sets, data sets that would make sense.
SPEAKER_04Now, that's why there's there's this concept of hallucination, because sometimes it guesses wrong. And I'm gonna harp on the word guess, right? These are probabilistic probabilistic models. These are probabilistic models, they're not always accurate. Hallucinations are a problem, and it's because they're guessing. They're just getting really good at guessing. Now, in a previous episode, we actually talked about uh me setting up a model on my laptop, and when I talked to it, um it missed one thing.
SPEAKER_04Every time I talked to it, it had zero context, right? I it would restart the conversation every turn. And so now there's an additional thing on top of models to make this work for you, which is context. Now we're getting into the concept of a harness, right? Context is it understanding the previous things you asked, whether it's within that thread, within a previous thread, or just in general things you've told it that it has now created what we would now call the concept of memories off of, right?
SPEAKER_04A harness is gonna provide that. Harness is basically saying, I have this horse, this engine, that is the model. I'm gonna put a harness on top of it so I can theoretically ride this engine in an appropriate and in a very good optimized manner. So this harness is creating uh context for us, it's creating memories for us, it's also creating the right way to for this model to be prompted. So there's there's a ton of instructions in the harness on how this model should react to certain prompts, um, which also allows for safety guardrails, right?
SPEAKER_04All the safety guardrails that we've talked about in the past. Uh, and and we would presume that Enthropic and Chat GPT are using. Um harnesses also allow for us to create our own instructions. So always be concise, always use bullet points, never use bullet points, don't use m-dashes. Um, it also allows us to create instructions in packages. So that's what skills are, right? Enthropic skills. But I want to actually move away from anthropic harnesses, which we call clawed, or chat GPT harnesses, which we call chat GPT.
SPEAKER_04And I want to move to harnesses that we've actually been using every day. Right? Harnesses are not a new concept. We've always had harnesses. We've had harnesses, for example, for mobile applications. Mobile applications out there were not usable unless we had a harness. And in that case, the harness was a phone. Right? Uh we've had harnesses at home every day of our life, uh, most of us, I think. Although I now that I'm in a bubble, I'm like, do I live this weird life?
SPEAKER_04Um, Alexa at home, Google Home at home, those are harnesses. They're driving mobile applications for you. They're driving you know, whatever it is that you do with them for you. Um, so now if we think of those as harnesses, all of a sudden AI is starting to pop up in them for us to drive. And I I think this is interesting because Chris, I think where you're going with this and what you just touched on is you've got a harness that you've used on a daily basis that you're comfortable with. Um I still, you know, I still have a lot of hate for iPhones, and and I'm gonna say that out loud um because I I I stand by it.
SPEAKER_04Even you are an iPhone.
SPEAKER_01Yeah.
SPEAKER_04Just FYI to the population of the world. Um, you are an iPhone user, and that is your harness. You're comfortable with it. You and I have this debate all the time. Like you love iPhones, I love Androids. Um, but respectfully, we're always on our phone. And you're moving your usage to what is more accessible on your phone. So I'm gonna put this back on you. I one, I hope I did service in the terms of the definition. Yeah. Um, for the masses, please, please do not uh destroy me on on uh on your comments.
SPEAKER_04Like I'm trying to appeal this definition to the masses, but Chris, talk to me. Did I do a good job on the definition? And two, talk to me about the idea of you just moving more and more to harnesses that you've historically always used.
SPEAKER_01Yeah. So yeah, you I think you talk through what a harness is very well, right? So if you had to put it into one sentence, what what would you say that it's it's the world an AI lives in? It's the world your experience lives in.
SPEAKER_04It is the world your experience is created by.
SPEAKER_01Okay. Okay. I like that. All right. So yes, and we've talked about this before in prior episodes, is like what happens when the thing that's always with me, and for most people, that's a phone, what happens when that can really do some of the things that these um foundation model harnesses have been capable of for the last you know 18 months or whatever. Um starting to see, at least in my universe, which is the Apple universe.
SPEAKER_01I've been an Apple fanboy like since I was young. Um and we finally got our first glimpse of this with iOS uh what is it? 27. I think it's 27. Um it was just released this last week, so I've spent a week with it, and already I'm offloading many things that I was doing in ChatGPT, and I have a convenient list here. I can talk about some of those things. So the harness of my iPhone has all my text messages.
SPEAKER_01So in the past, when I would ask my phone, hey, you know, what did Pratik say to me? It could only ever just run a basic query and find the most recent message and read it verbatim.
unknownAll right.
SPEAKER_04So for the audience, is you need that with my messages because I tend to message in just large wall of text. Yeah, I need summaries.
SPEAKER_01You're like, I'm not gonna read any of that, summarize. Yeah. So, but I like using my voice, I'm able to ask Siri, hey, uh, what did Pratique say to me recently? Yeah, and it's phenomenal. It tells me about the last text, it's doing it all in summary. It's saying these were the topics, blah, blah, blah, blah, blah. Um, and it went back multiple messages, right? It considered recently, basically, I think about a week. And just fantastic. And I was never able to do that anyway with ChatGPT. So that's not something I offloaded, but that is something that all of a sudden my phone can do, and it's great.
SPEAKER_01Um, another thing my phone can do, which I wasn't using ChatGPT for, is just manipulate whatever app I'm using. So a great example is Maps. I was driving and I said, hey Siri, add a stop before I get to my destination, find it, and it just does it. And it it has all the tools built in for it to do all the things in most of the apps, right? So we just added the thing while I'm driving. Um, look at my email. So here's one I I can't offload, right?
SPEAKER_01So I did have my email connected to Chat GPT, but now I can ask my device where I have all my email accounts already connected and it's easy, and I'm not paying extra to connect anything, right?
SPEAKER_02Yep.
SPEAKER_01And I can say, go back into my email, look for this order. I bought this thing, I don't remember what size it was, I need to know, and it just finds it and it's fast. Yep. Um, the some of the larger things that I'm doing that I've done with Chat GPT for a reason because I couldn't do it anywhere else. So I think I've talked before on an old, old episode about um my glucose monitor has an app, but the app doesn't talk to Apple Health.
SPEAKER_01And because I'm an Apple fanboy, I want my data in Apple Health. So I had to create an Apple shortcut that opens ChatGPT, takes a screenshot of the glucose monitoring app screen with like my current blood sugar level in it, sends it to ChatGPT. Chat GPT evaluates it, figures out what the blood sugar is, sends it back to my phone, and then my phone adds that to health.
SPEAKER_01So that is taking up tokens from ChatGPT that I would rather be using for things like computer use, right? Like I want to be saving these things. So now on my iPhone, I went into, we have an app called Shortcuts, and I just in natural language spoke to the app and I said, I need an app that does this.
SPEAKER_04Yep.
SPEAKER_01And it made it, and now on my phone, totally local. It never goes anywhere other than my device. It opens the app, it takes a screenshot. The AI now on my iPhone is smart enough to pull the blood sugar out and add it to health. It's faster, and again, it never goes anywhere, it never leaves my device, which is amazing. Um, it can do all kinds of things for medication because I track my medications in my device. It can tell me if I've taken a dose, if I've logged it, um, what do I need?
SPEAKER_01What time is my next dose? All that kind of stuff. And the biggest one, and I'm not sure how I feel about this, because I got in the habit over the last year of using my ChatGPT subscription basically is super Google.
unknownOkay.
SPEAKER_01Right? Whatever question I had, instead of going to Google, I went to ChatGPT and I said, here's my question. And it does an amazing job at finding answers and options and providing links, you know, so I click through and go read something myself. I have now started trying to do that with Siri on my iPhone instead. And it's working. It's working. So like that is a huge chunk of what I was doing with ChatGPT. I'm now just doing on my phone, and I'd rather do it on my phone because it's easier and it's not costing me anything.
SPEAKER_01Like if I wanted to cancel ChatGPT, I could because I I have now gained that ability just on my device, on that harness. Now, I won't cancel ChatGPT because I use it for more complex things. But for a lot of people, yeah, for a lot of people, maybe the device is catching up. Like I think, I personally think this is how the bubble gets burst. I think this is how everybody, the bubble we're living in of who uses AI right now is a very small bubble.
SPEAKER_01I think this, these things, these features coming to our personal devices is how the masses are going to start to use all of this. But I am curious, that's my experience on an iPhone. Has it been that way on Android? And I'm just behind.
SPEAKER_04Yeah, so I'll I I'll answer that question. But I want to dissect a few things, right? Okay, we've talked a lot about um AI in enterprise. We know that that's like an initial target for Anthropic and ChatGPT and all these companies. Um, argu I I would make the argument that I didn't really see deep, deep use case penetration of the organizational like enterprise until Slackbot, right?
SPEAKER_04Or Slackbot type technology. I I truly believe AI started to get more democratized when Slackbot came out. And the reason I believe that is because, yes, you found the fringes and definitely coding, right? Like we we saw coding everywhere. Um but that does that's not democratization, that's just one domain being really deeply penetrated. I I when I talk about like democratization, I'm talking about AI being used by the people that you wouldn't have never guessed being used. Um and I saw that pop up with Slackbot.
SPEAKER_04I wanna I want to draw the parallels here, so just bear with me for a second. Okay. What Slackbot allowed people to do was the very same experience where their their world, like their experience at work, largely was through Slack, right? I I don't think anyone's gonna argue that. Their experience was largely through Slack. So if their experience was largely through Slack, Slack was the perfect conduit to create the AI experience, right? Almost by definition. Because instead of talking to Christopher Long, I just talked to Claude or whoever.
SPEAKER_04And then if if we're saying like Slack was almost the perfect harness to unleash democratization, democratization of AI, um, the second thing you said is also relevant, which is I'll just I'll just make it all you can eat. Right? And why can they make it all you can eat? This is where it becomes really interesting. If you think about an anthropic, they can't make it all you can eat, right? Not not with not without putting certain limits in place, because that would get ridiculous. They can't because they make their money on how much you eat.
SPEAKER_04With Slack, the harness, they make their money on you buying a Slack license. One feature of which has to you know attract you in buying that license is this consumption or all you can eat um ability of AI. And you're excited about that because you already use Slack, you already want to continue to use Slack. That's where your experience at work was already being created. Now you just added a new feature that actually feels valuable to you when it comes to AI. So we democratize AI that way.
SPEAKER_04Now let's bring it back to what you said, right? You you touched on all those things. I already was using this mobile device, it's already the thing that I use as a daily driver. I love the fact that it's just free. Just consume it. Why can Apple make it free? Why can't Chat GPT make it free without certain limits? And why can Apple make it free? Well, because Apple sells phones, Apple sells devices, and you're already paying a premium on those devices, right? You're you're certainly paying a premium on those devices.
SPEAKER_04So they're adding a feature to keep you locked in. You're you were already an Apple fanboy, now you're more of an Apple fanboy because they're they're now giving you an all-you can eat feature uh for things you already do in a device that you've been using for years. So I think there's a parallel here. And if Slack and SlackBot was the democratization of AI in the enterprise for the masses, or as much as we can define masses right now, then I I agree with you. I think getting good agentic AI in the harnesses that everyone uses on a daily basis, right?
SPEAKER_04Like my not not my dad having to download Chat GPT and figure out what to do with that. My dad using an iPhone, he's 75 years old. We had to teach him how to use an iPhone. Took a lot, took a long time. Well, now if if if there's just an additional area where he he says, do the thing I've always done, but all of a sudden AI is trying to do it for him and do it more efficient and more optimized. Like that is the that is where I think penetration uh will happen in terms of the masses for personal use. Um I'll get into your question now, right? How how is it on Android?
SPEAKER_04Um, if I don't quote me on this, but I I want to fact check this. I believe Apple is using Gemini models to create that experience, right? Like that that is where those experiences are coming from. I believe they have a partnership with Google. So um to you know, to actually answer your question, Android has been doing this all along. Um now you have actually the option if you have Claude or Anthropo or sorry, ChatGPT, you can swap out models. The feature set gets diminished a little. But Google Assistant has been there uh at my side for a long time.
SPEAKER_04It's it's weird. I I've never talked about it, right? Because I'm always talking about my experiences as a desktop experience as a laptop user. Um, but my phone, I really don't use anything other than just Google Assistant and whatever Google Assistant is giving and serving. Can you add tools to get to everything on the phone or most things?
SPEAKER_01Yeah, okay.
SPEAKER_04Oh, yeah, yeah. Tools, um, it can take actions. I mean, it all those things, right? Um, those experiences are there. You can and I believe because it's an Android, you can start to create your own experiences if you build and go into developer mode, which I haven't played around with yet. Um, although now I think that's the next thing I'm gonna do is start to play around with the phone version.
SPEAKER_01As an iPhone person and say, Thank you, Android. Thank you for sharing your models and your techniques, because uh we are uh thoroughly enjoying them, and I think it's going to make our lives better. And maybe now we can all just get along.
SPEAKER_04Insert joke about RCS messaging here.
SPEAKER_01Um I think uh uh that that's a green bubble joke for people who may not know what you're talking about.
SPEAKER_04I I I think uh if if you don't know what I'm talking about, they they should Google it because we don't have time to spend on it. But I think it was a phenomenal joke. Um you're welcome, Chris. I think here's the thing, right? I I also wonder, like with the partnership, is there potential uh for multiplayer? Right. Actually, ironically, with the RCS messaging uh joke, like is it possible for us in the future, you know, when when we use our mobile devices, for me to kick something off as an agentic experience and then say, non-Android user or Android user or whoever, like we're going to multi-team this and we're gonna team through our phones.
SPEAKER_04Not because I want to do that for work, although I think that would be cool, but because I want to do that in life, right? Like if we're going on a trip together, right? Chris, you and I are going to Vegas, or you were taking two family trips or you know, family trip to New York or whatever it is, like I would love to create that teaming agentic experience. Um I don't know why I've got expense reports on my brain, but it we we're trying to figure out like how do we you know divvy up all the bills, uh hotel bills, Airbnb bills, restaurant bills. Uh every time we're in New York, Chris and I get dim sum, right? So um we gotta pay for those dumplings somehow.
SPEAKER_04Like the agentic experience around that could be really interesting and and start to offload teaming tasks uh in a personal life. All right.
SPEAKER_01So I think we've really brought to the forefront the idea that uh the the device that's with you is gonna be the thing that brings everybody into the fold. Um it's really gonna be the thing that drives adoption, and I think people and we've talked about this some before, everyone's going to be using AI whether they know it or not. Right? Because it's it's coming into everything. And the device, the personal device, is where that's going to happen.
SPEAKER_01But um, this is about harnesses, and there's a brand new one that you got to tinker with, and I've I've heard good things about it. I mean, my primary concern, and you I can tell you already know how you're gonna set this up, but um, at some point I think we have to talk about the trust factor with this particular harness because of the reputation of the provider. But yeah, tell us all the cool things up front.
SPEAKER_04Yeah, I um do any of them have a good reputation? Okay, so fair. The Facebook, so I'm so excited. Facebook just launched, or Meta, sorry. Uh Meta just launched Muse. Um, and there's a few reasons why I'm excited about this, and I'll get into them, why they're really pertinent to me. Um, but Muse is an agentic experience, just like all of them, um, where it will take actions on your behalf if given the prompt and given the ability to do so, approvals and credentials that it might need stored securely.
SPEAKER_04So um it's it's like that experience, right? What makes it different is that it is a cloud-based LLM harness combination that also has a cloud-based computer. So with Chat GPT and Anthropic Cloud, you have to provide the computer, right? Either your laptop is the computer or you're giving it remote access to something. Um, that's not the case with Muse. With Muse, um, I can access it from a phone or a laptop, and it's going off into its own cloud-hosted computer and accessing things there.
SPEAKER_04Now there's a few concepts that I want to play around with here, right? LLMs are great, but the internet is not designed for LLMs, right? So if I go to um any particular place, like recaptures are a thing, and sliders prove you're a human, and yada yada yada, and like some layouts are antiquated and you have to try to figure out how to do it. There takes uh humanist context to actually navigate the world, the internet world. Um, and the reason I think that's relevant is claude claude in Chrome was really good.
SPEAKER_04Chat GPT computer use was really good. But there's always those one or two things that it runs into, right? I'm not seeing those instances with Muse. With Muse, if it runs into a roadblock where it needs you to do something or take over, it actually flags it to you and says, I just need you to move this slider. This actually happened to me this morning. Just go in into the browser, it's remote hosted. You can at you can get in there and just slide the slider for me so we can move on and I'll just take action to the next thing, right? Um, it it is it is fascinating how good it is at this.
SPEAKER_04Um, or if it hits a recapture, it goes, hey, I've hit a recapture on to you, just go in and authenticate and I'll I'll move on. So a few use cases I've already done with Muse. Uh, number one, I've had it go through um a few subscription accounts, go in with stored credentials that I I shared securely, right? And I know you're gonna put probe into that.
SPEAKER_01I'm just listing. I'm in listing mode.
SPEAKER_04Yeah. The I went in, it went into my subscriptions, it went to customer support, and it just started negotiating a loyalty discount. Um, it was successful on two of them. So great. I got I I did very little work to just save that money, right? Just trying things out. Um the the second thing I did was I said, hey, I get a lot of spam calls. I want you to just scour the internet. One, see if there's things I can sign up for, like do delete my data or don't do not call lists or whatever it is.
SPEAKER_04Um also see where potentially my phone number is out there, just kind of riddled on the internet. It did both of those things. It signed me up for a few things, like do not call lists. Again, it's checking in. It's always checking in. Do you want me to do this? Yes, go ahead. Do you want me to register you with the FTC? Like, yeah, go do that, right? Do you want me to delete you from this database? Yes, quest that. Um, and then I went out and said, like, where else can I find your phone number? And it did a bunch of research. Um, I've always wanted to do that. I have I don't have the time to do that, right?
SPEAKER_04Um, it went off and did that. I tried that with Claude, it was not very effective, right? It was really good at browsing the things I said, browse this website, but like opening up a ton of different browser instances, navigating through all of them on its own. Like, I I wrote one sentence to Muse. I I didn't really do anything, right? Um, and I tried the same thing with Chat GPT. It wasn't the the same efficacy wasn't there. The use case I'm now gonna do with Muse um is I'm actually planning a trip to to Quebec City for Christmas.
SPEAKER_04We want to do the Christmas markets, yada, yada, yada. And I've given it all the requirements. Leave on this day, plan this, so we're gonna you know, stop here for food, and we have a two-year-old, so she needs to do naps, and we need Airbnbs, and let me know where where we need tickets and go ahead and actually try to purchase the ticket. Like I've I've said, take this as far as you can. I'm not giving it my credit card. I will happily go in that browser and put my credit card and do the final purchase. You know, search for all the Airbnbs, create lists of like tell me everything that is gonna make this itinerary perfect, but do all those things as well.
SPEAKER_04Um I am mind blown at the difference of the experience between Chat GPT, Claude, and Muse as it relates to that use case. Mind blown. The last thing I'll do, and I'll talk about it in terms of this use case. I'm always on a plane. Of course, you know this. You actually messaged me yesterday, and you're like, why are you messaging me on WhatsApp? I'm like, because it's currently the same. Because it's currently the I'm on a plane. It's the only thing I can message you with. So um, you know, if if you're like me and you don't want to buy Wi-Fi every single time you're on a plane, if you're like me and you don't want to buy Wi-Fi every single time you're on a plane, um, there's free messaging.
SPEAKER_04As part of the free messaging, you can use WhatsApp. And so I mentioned earlier I was using meta AI when I was on a plane because that's what I had access to. Now you can use WhatsApp as the as the conduit to Muse, which is a harness, to use meta AI. I can dispatch Muse to go do these things. And here's the here's the real conduct. Kicker. Because it's a remote computer, I don't run into the roadblock of not having Wi-Fi because I'm using free messaging to go dispatch my dispatch my agent.
SPEAKER_04And this is this was the promise of OpenClaw, by the way. OpenClaw was originally designed for the plane use case where the where the dude was on plane and was using WhatsApp. I dispatch Muse, it has access to a remote computer somewhere, does a bunch of things, and comes back and just checks for approval or feedback. So I planned this entire Quebec City trip where it was searching Airbnb and searching all these uh things to do in Quebec City and Montreal and yada yada yada from just a few messages on WhatsApp.
SPEAKER_04The harness that I use already, right? And for the international community, WhatsApp is the primary harness for most people in terms of messaging. It's happening in my workflow on my device, accessible everywhere, but with a remote.
SPEAKER_01Yes. All right, let's get into it. So security. Well, not necessarily security. So I do have a couple questions because I've heard a lot of people give glowing reviews of Muse. Um the use cases you described to me, the one that is much more interesting to me is the one where it went and negotiated deals on your behalf. The planning a trip, I'm sure it did an excellent job and probably did a couple things that maybe weren't obvious to ChatGPT to do.
SPEAKER_01Um, but like you can do that really well with any of the other um solutions, right? Like you can do that. The part where it's actively messaging on your behalf using your accounts, like like you could give it your email and it can start emailing somebody back and forth, and you you're just out of the picture, it's just doing it.
SPEAKER_02Yeah, oh yeah.
SPEAKER_01So the person on the other end, right? Person on the other end, I say person, who knows if it's a person, but the whatever you're communicating with. The muse agent on their end. Yeah, they don't know it's not you. Correct.
SPEAKER_04Right. I have no idea. They muse was in my login, so on my portal for the subscription service, pulled up a chat with customer experience and started messaging as if I as if I was doing it on my laptop. From the laptop's perspective, from the agent's perspective, from the website's perspective, and the portals perspective, it's pratique. We know that it's Muse using my credentials and using an LLM to power the next action.
SPEAKER_01Yeah. I mean, that that part is, you know, what's the word? Like it creepy's not the word, but it like, is it disingenuous? Is it does it matter? Like it's still your intent, and now this is a tool doing it for you. Um, there's just so many questions there. But I'd really kind of like just to focus on today about like the tool and the harness. So when you're setting all this up, does it feel safe?
SPEAKER_01Like, how do you how do you give it access to your email to run on a different computer? Like what specifically did you do? What did you give it?
SPEAKER_04Yeah, so in terms of email, it has a connector. So it's no different than chat GPT, it has a connector. So you're not giving it credentials. Okay. I presumably the connector is based on OAuth, right? Um just like ChatGPT's integration would be or um or Claude. Now, not everything has a connector. Okay. Obviously, right? So the difference between uh ChatGPT and Muse, and I think ChatGPT and Claude are starting to start to do this too, but with Muse, you you have this locker.
SPEAKER_04It's a locker. Okay. And in that locker, you can put your username and password combinations for things like subscription services, yeah. Uh for accounts that don't have connectors that you want it to have access to by by but and again by pulling up a website, you know, let's call it Amazon.com, and literally navigating to the sign in, putting your username in, putting the password in through the locker, and then signing in. Does it feel safe? Yes.
SPEAKER_04Okay. Does that mean it's safe? I will go on the record and say, I didn't put anything in there that I felt it could really cause me harm. Okay. Right. So the subscription the subscription services, I'm already at the top tier. It can't sign up for higher tiers. Like I'm already at the top tier. Anything in negotiated, unless I negotiated up somehow, anything in negotiated would have been down from there. Okay. But I was very strategic in what I shared in the locker. Okay. Um and and I and I I say that not because I believe I know something dangerous, but because I I just don't know yet.
SPEAKER_04And and I don't think the world knows yet. I believe they wouldn't have launched something unsafe. Yeah, we use anthropic and open AI every day, and all of those leaders are up there on the stage saying, this isn't safe. So I so I I don't know how to reconcile that. Um I I don't know how to reconcile that at all, but um, I'm I'm protecting myself to the best of my ability. So I did not share anything that I feel uncomfortable sharing. Um in terms of like the the the last piece in terms of safety or security that I want to mention is that it it does a very good job of making you feel like it's checking in always.
SPEAKER_01Okay.
SPEAKER_04Now, I don't have audit trails, I don't know exactly what it did. I mean, I could literally go in, check every browser if I wanted to, I won't, and I don't think anyone's gonna realistically do that. I don't know what it actually did. I know it checked in with me, I know it told me what it did. I'm believing it. I feel like it's telling me what it did and the truth. Um, so it does feel like that. But again, we're using the word feeling because we don't know.
SPEAKER_01So yeah. This is I know we're talking about harnesses, but this is something I want everyone to hear. Is so I did something very similar this weekend, and I mentioned already that I had Chat GPT computer use help me set up some things on Slack. And I was in that motion of it's working, I'm doing something else on my computer, and every now and then it would pop up and say, here's where I am, or I need I need you to click this particular thing because I can't get to it, or whatever. It worked through that whole workflow, worked through the whole process.
SPEAKER_01Um okay, great. Task complete, thing done. That was Saturday. Sunday, I go into my Slack and I notice there's now a new workspace. So through that process, it didn't, it it's not costing me extra money. Right, right. Like it didn't, it didn't it didn't endanger me in any way, it didn't, you know, wreck anything or break anything, but it did through the process of setting up this thing in a already existing workspace, create a new workspace, um, which through the flow of it doing the things was not apparent.
SPEAKER_01Yeah. Um and so I like that's why I just have so many questions about a process that's running off in the cloud. Um and you are it is checking in, it's checking in. Yeah, but it but man, you are really in the for better or worse. Like, I'm not trying to say there's a better or different way to for it to do this. Um, but you have to realize you are giving up a lot of control.
SPEAKER_04You just are you're yeah, that's all if you think about like LMs are a technology looking for a problem to solve, right? I I think that's where we we don't have like true product market fit in most places. And I think the the closest to product market fit that we'll we'll be able to agree on at a scaled level is coding. Right? Unless you can tell me anything else that feels really scaled. Um, coding is the one.
SPEAKER_04And inherently we mentioned earlier that LLMs are probabilistic systems. So by definition, a good product market fit for a technology that's probabilistic is a tech is a is a problem or a domain that has the QA baked into the process. If you think about coding versus every other use case we've just talked about, coding historically for 30, 40 years, there's always a stage gate where you QA the thing you created.
SPEAKER_04And you had the ability to QA every aspect of it. It wasn't like, oh, we just QA a little bit and trust the rest. You were QAing, regardless of who wrote the code, you were QAing every aspect of it. And so when we when we started creating code through LLMs, like that mechanism was already there. That audit trail, that QA, that like trust factor um was so much easier. Like the the burden of proof was always on the QA process versus the actual actor taking the action.
SPEAKER_04With every other use case that we're targeting, whether it's on the enterprise level or the personal level, there's not a lot of QA baked in. Right. Right? And so we're just kind of like taking the action and hoping. Yeah. Um, and I and I think that's what that's the fundamental, in my opinion, though, that's been the fundamental difference between product market fit on coding and product market fit everywhere else. Everything else is a feeling. I'm using my iPhone. I've always trusted my iPhone, so I feel like I can trust this. I'm using Alexa, so I feel like I can trust this.
SPEAKER_04Um that's that's the fundamental difference, and that brings me back to that bar a week ago. Okay. Literally today, a week ago, at around the same time. I just told everyone I was drinking at 11 o'clock on a Monday. But um right around the same time. You didn't. Well, now I did, yes.
SPEAKER_01You hadn't told anybody until you said that.
SPEAKER_04Until just now, yeah. Um that moment, I think I think the reason we were going around a circle and everyone goes, I don't, I don't use AI, is because I think ultimately the barrier right now is the feeling is that you can't trust it. And none of those none of those folks were coders, by the way. I think it would have been different with the with the coders. But on average, every single one of them said no. So I think it's a feeling.
SPEAKER_01All right. That's probably a good place for us to wrap up the harness talk. Um I find trying all of these different things so fun. Yes. Um, I mean, we you and I are early adopters, we are, you know, techno geeks, like we love this stuff. I've learned so much though by just trying all these different things out. So if you were, I mean, I know we we started kind of our entry into the harness discussion with that question of what would your daily driver be?
SPEAKER_01Um, what if we close after having the discussion with if somebody wanted to try something new as a daily driver, um, what would we recommend they try just because it's fun and interesting? Would it be Muse? I I don't think so.
SPEAKER_04I don't think I don't think it would be Muse. I think it'd be you know pick up your phone and just turn on the AI. And it's probably already on, quite frankly. Right. So if you're if you're using iPhone, just talk to Siri. I think that's your introduction. Siri's or has already been upgraded to the latest AI technology. Um if you've if you're on an Android, talk to Google Assistant and just just start doing the things that you yourself would feel burdensome, planning trips, reading through messages, um, coming up with responses.
SPEAKER_04Like what do you do on an on a daily day, day-to-day basis? I think the harder thing is not knowing where to start. I think start is just your phone, whatever phone capability is, is is the harder barrier is coming up with ideas on what to try. So just what do you do that's so repetitive, so often occurring? Try those things with your phone.
SPEAKER_01All right. Well, thanks, Ratique. This was a lot of fun. Everyone out there, if you would like some ideas for how to start on your devices, let us know. Ask us for some ideas. If you have ideas, if you've been trying things, let us know. We would love to dig into those and talk about those more. If you enjoyed today's podcast, please tell your friends. We are available on Apple Music, on Spotify, on YouTube. We are also now, we have a page on Buzz Routes so you can get our podcast, the audio version on some other podcast providers.
SPEAKER_01Uh, we're also on the TikToks. That's how you say it, right? The TikToks? The TikToks, yeah. And Chris.
SPEAKER_04Don't let your hear you, don't let your daughters hear you say that.
SPEAKER_01Yeah, yeah. Yeah. Not no, not if you want them to think you're cool. Um, but we've already established that we are. We are very cool. That's what I meant. Yes, we are very cool. So we are on the TikToks, Teak and Chris, T-I-K-A-N-D-C-H-R-I-S. So look us up on there. Uh, we've had some already have been having some great discussions on TikTok. But thank you so much for your time. Take care, and we will talk to you again soon.
SPEAKER_04See you next time.
SPEAKER_01That's the buzzer, folks. Thanks for listening to the Teak and Chris podcast. Do us a favor, like and subscribe.
Chris and Pratik compare notes after weeks with Meta’s Muse, from its single ongoing chat to memory and soul files. They share practical uses for calendars, email and ongoing goals, then discuss payments, credential...
Is AI exhausting you as much as it’s exhausting us? Chris and Pratik are tired of keeping up with AI. Their solution to dealing with AI fatigue involves more AI, which they acknowledge is a snake eating its own tail....
No guest this week, which means Tik and Chris had all week to argue about whether this episode should exist at all. Here's the problem. Nvidia's CEO says AGI has arrived. A researcher walked out of Anthropic the same...
Tik and Chris finally do the thing they've been teasing - they bring on a guest! Craig Martin, a water treatment project engineer from central New York, joins the show to pull back the curtain on something the AI world...
Chris kicks things off with a story that'll either make you laugh or cringe: someone paid $100+ for an AI-generated song that you could make yourself for much less. From there, Tik and Chris pull that thread all the way...