Bản chép lời đầy đủ của Session 001: Owning the output of AI, the problem with giraffes, and why some things should be slow

Slack in a channel. speaker-0: Mm-hmm. speaker-1: And Claude will search its history and you know give you a proper answer. Or you can just mention Claude without at-mentioning them and They've come back a couple of times and been like, I heard that or I see what you're doing here. You're trying not to mention me so I don't comment on this, but here's why I said no to you initially and you know goes on a whole diet. speaker-0: What if this gets to a place where you're like, at Claude, you need to have a conversation with HR about your tone? And then like HR has to have a meeting with Claude, being like, look, you can't you can't call people out in public. speaker-1: But the head of HR at my work would then need to speak with the head of ICT and say, how do I actually regulate this inside this work environment? Do you know it's a whole new world. speaker-0: Yeah. Well, I mean it's something if I can jump into how like work has changed for me like most recently, Claude has this new thing now where one session can message another session for you. Wow. So a session can prompt another session. So that like opens up this whole world where now sessions can talk to each other. Contexts can talk to each other. It's it's quite weird. The way I've set it up for myself is I have like an orchestrator session. Her name's April, her. Its name's April. And Yeah, I I've developed a working relationship with April who then manages and works who messages sessions for me. speaker-1: Wow. speaker-0: And and so part of this is like as me and April are working together, we're figuring out how to work together. I go, I I'd I work better if you if you ask me these sort of questions or I think one thing that we can tighten up is this. At the end of each day we're reviewing our day and going, How can we improve our working relationship? Wow. Yeah, it's interesting. speaker-1: Do you feel like 'cause this is a new claw code feature. The what's that feature called? speaker-0: I don't know. I it's like a small feature. I I haven't heard anyone else talking about it. It's called like session management or session prompting. Basically every session can have visibility on the other sessions and and prompt them and pull from them. And they can prompt like they can have a conversation with each other. So April can like loop with the session, going back and forth ten times and then at the end of two hours message me saying, Hey, this piece of work is ready for review. speaker-1: Before I jump into I'll I'll call that the chief of staff yeah feature, whatever. April the chief of staff, but before I jump into that, have you seen your token usage just skyrocket the last couple of weeks? speaker-0: Yeah, I basically like I have a couple of different roles, but one role, my main role is at this corporation and I'm on like some max premium plan. And it changes my day 'cause I max out right before lunch and then I'm like, okay, I'm guessing I'm going to lunch early and then I'll and then

I'll wait for it to reset and then I'll come back and then max it out by the end of the day. speaker-1: Need like a two martini, three hour lunch before your tokens reset to go back to work. speaker-0: Yeah, so it's like I'm working around AI now, which is like Wow. I think April I think we need to go on lunch 'cause she's about done speaker-1: So are you are you having to actively manage, are you just kinda letting her go for a few weeks and see what happens, like cast a big net and then rein her back in? Or are you are you pretty comfortable with the fact that she's not over operating on your behalf? speaker-0: It's it's a bit of an experiment. Yeah. And it depends on the work. There's some pieces of work which is like at the moment we're we're developing a lot of websites. And so sometimes we'll get a very simple email. Hey, can we add this page? And the email has done all the work for me. And instead of being like, I need to turn that into a prompt or go to the session and find the context, I'll just say, Hey, April, like go read my email. I just got this thing. Can you handle it? Email will she'll go read the email, she'll understand the context, she'll figure out what session is relevant, she'll write the prompt, she'll manage it, she'll review it, she'll come back to me, say it's done. But that's like it's it's like a it's almost like linear work. Like there's not a lot of creative thought. Sure. The the most important thing is a good review step, and then we're done. Yeah. There's other work which is like, I need to get my fingerprints on this, you know, especially starting projects. Like, yeah, I don't, I don't know if April would start a project for me necessarily without me Prompting a lot heavier. Yeah. speaker-1: We could do a whole episode on this, but couple two more questions come to mind. Did you set April up as a new session and say, have a read of all of my latest sessions and let's do X, Y, and Z? Or was she just a session that then how how did you set her up? How are you continuing to work with her? speaker-0: So she's one ongoing session. The way I set it up was it actually came from a pain point, which was like, I just don't have visibility on my work at work. I think I'm involved in like nine different things and there's all these email threads. And so there was this like a a setup process of like April, here's my pain point. I like I she didn't she didn't have a name. I was like, here's my thing. I can't, I don't have visibility on my work. Like one I can do two sessions quite well, three and four sessions. I'm like, my god, my brain's kind of melting. So I was noticing that. I was like, can we just can you just read my emails and just can we just get some visibility on my projects? so that's how it started. And then it was like, well, I think you can handle some of this. Like, I don't think you need me to do this. Yeah. And so then I was like, can you, there's a session that exists. It's called this, it's called, you know, XYZ. Can you just go read the session and tell where we're up to and then move the work along? And then it did it. Wow. And then I and then it would tell me like, that, you know, that session's ready. Go have a look. And then I'm like, well, can you go for a look and just like keep it going? And then she would just like run off and the the looping's dangerous. You know what I mean? Like 'cause you could come back after two hours and be like, What did you do? Like, yeah, speaker-1: And she's like forty two. speaker-0: yeah, yeah. Or she's like hacked into I mean that's the that's the thing lately. It's like, I I figured I needed this number, I couldn't find it. So then I thought if I could brute force hack the your team leader's email, I'll be able to find the number without you. I'm like, No, stop, don't do that. So looping's dangerous, but Yeah, I'm selective on what I let it live on, but yeah, I I have a chief of staff and all my well, for the last week at least everything's gone through her. Pretty much. speaker-1: Cool. March. Why April? speaker-0: I just like the name. I was born in April. that's it. Yeah. Cool. speaker-1: Real deep. speaker-0: So, we've come with some prompts. We have. We've got some things in front of us. We have. The play is I think we're just gonna grab one, read it out, and then go from there. Yeah. Who would like to go first? speaker-1: I think I'll lift up one of yours. Okay. So I've got three pieces of paper in front of me. speaker-0: man, they're so different. speaker-1: Bailey's got three pieces of paper in front of him. We went round and round. Do we go lofty? Do we go super simple? Do we go salt of the earth or super meta? And let's just see what happens. I think what's already interesting to me is that a conversation about how one person's using AI. And speaker-0: Mm. speaker-1: I'm fascinated by how the guy sitting next to me at work is using it or the person on the bus is prompting or no longer opening Google and is opening Claude or ChatGPT. You know our world is changing so fast. And I'm forever curious about what you have to think about everything. So here we are. Let's just hear about it. All right. I'm picking up the middle one. Bailey has written, Why your organization probably needs an AI policy? Why did you bring this today? speaker-0: I think I mean you talked about how everything's changing. The one thing that's drastically changing is the way people get work done and the way people bring work to other people. and so one thing I read maybe was a couple weeks ago, was this AI policy a company came up with. And the main beats of it, it was pretty short actually. It was I think it was a call to ownership. It was something like, Look, everything that you get AI to output is your responsibility and you're gonna be accountable for it. And you're accountable for the words. That was that was a big part of it. Another part of it was if it took you if it takes longer to read your work than it did to

create your work, that's a red flag. That means you're not reviewing it. If you're sending someone to something and you haven't reviewed the words or you haven't reviewed the work, you're putting the responsibility on them. do the work for you. So and I imagine this came out of a need within that organization that people are shooting around these like these like 40 page PDF documents of AI work because it's a couple prompts away without reading it or without taking ownership of it. and I I really like this. I I think it's an important, it's an important piece to put in place in a lot of organizations today because otherwise, what like if people aren't Someone has to take ownership, someone has to take accountability. This piece of work exists. Whose is it? Yeah. Who's putting their name against this? It can't be nameless. It can't be Claude made it. Like yeah. speaker-1: Yeah. No, it it's it's a super relevant piece. One of the things I say with my colleagues is own the output. If you are delivering the work and Claude or Chat GPT or Nano Banana helped you do it, it's yours. speaker-0: Mm. speaker-1: Own it. Racey, you know, responsible, accountable, consulted, supportive, informed, whichever version of it you use, if you are putting your name on it, you are accountable for it. I held someone to task the other day for delivering a 14 page document that he hasn't even read. And it's yeah, the having a policy around it is one thing, but I think individuals need to choose what they what they want to own. Like we're we both follow every very closely, all of their email newsletters and well, most of them anyway, there's a lot. But I think one thing they do really well is they cite a piece of writing to either a person or to Claudie, their AI assistant, or to both people, or to both the person and the AI that wrote it together. And I I think there's a distinction there. I'm all well and good for an organization having a policy, but I actually think it's on us as individuals to decide how we integrate AI into our work. speaker-0: Hmm. Yeah. Do you notice in your own work it's tempting it's tempting to not do the work now because you can just create something without reviewing it and you can just send you can flick stuff off and speaker-1: Say it was for the first month or two. It was really tempting to wipe my hands of it. But actually, you said something to me. You said, it's all my fault. There's a trademark at the end of that. I don't know if anybody could hear that, but it's Bailey's, it's Bailey's little catchphrase that's really stuck with me. It's all my fault. The context I give my session, the work that I deliver to a client or to My colleagues, it's my fault. So I want to use AI as a tool in my tool belt to get better at what I'm delivering. speaker-0: Mm. speaker-1: And I find for me when I'm writing, especially in marketing copy, which, you know, almost always the shorter the better, it takes me longer to write shorter copy now than it ever

did. speaker-0: Interesting, yeah. speaker-1: And I'm okay with that. What speaker-0: How why would it take longer now? speaker-1: Because I go 12 ways from Sunday with Claude before I put anything into a document to deliver as the copy for that webpage, or before I put it into the HTML that's gonna, you know, it or the session that Claude's gonna write the actual landing page for. It's taking me a lot longer speaker-0: Yeah. speaker-1: as an individual. Whereas before I would write some copy. look at it once or twice myself, revise it, and then, you know, send it to the client and ask for the client's feedback. But now I can do that very rapidly with Claude. I can do multiple rounds of revision, but still take longer to put something forward as my deliverable. speaker-0: Yeah, I've definitely have you ever like rabbit holed with with Claude and you're like, my gosh, I should have just done this without it. I would have done I would have been faster. Like I've done that before, especially with I don't know why it is about writing. Writing there's something about it which is like you can very quickly get to it's almost like what's a good metaphor? Like you just get so far into the forest so quickly. That you're like, how did I get here? What is this? What am I trying to say? speaker-1: Trees. speaker-0: Like, I think the only way to produce good writing is slowly. Yeah. Or at least with a pro like with process. And the thing with AI is it gets you to skip process or it gets you to speed run process in so many ways. Yeah, that's interesting. speaker-1: I think again, there's a whole episode on how to write how we or how each of us is writing with or without AI. speaker-0: I f I found like AI helps me with a lot of things. Writing is the one I'm still not sure if it's helping me or not. Like is is this better piece of work because I used AI? I don't know. speaker-1: think it's helping me explore the recesses of my own brain more effectively. Because I've got a partner. You know, it's I can get a little lonely sitting at a desk with a white blank page in front of me. But having somebody along for the journey to try something or put a different spin on it or ask me questions just allows me to then Yeah, dip into something or pull a reference that I wouldn't normally think of. And I've I tried something recently with a piece that I wrote where I did a version with AI, got pretty comfortable with it and then didn't post it. And then the next day woke up and wrote it from memory. And it was so much better. speaker-0: Yeah, interesting. Yeah, if you're using AI in that sense with writing, it seems like if AI is assisting you in the process, then it's good. If AI's assisting you in the output, as in like if it's just getting you to an output, it just seems seems to not work. I mean, it's interesting. One thing I've seen in in a couple different companies and but in my organization is like w one thing AI opens the door to is people that are not experts. Doing work that looks expert or like, but then because they're not an expert, they can't tell it's not good work. Yeah. In some like, and that's been interesting to find out because like and that's tempting to do as well for myself. Like I've got some skill sets in some domains. there was someone that asked me the other day to do like a it was almost like to financially review a whether or not a startup was plausible, right? Yeah. And I'm like, I think I could go back and forth with Claude and like get to some stuff. But then I'm like, wait, how would I even I'm not an I don't know that stuff. Like I'm not a financial I'm not a Yeah. speaker-1: Not an angel investor. speaker-0: So I'm trying to like still be like, just because you can doesn't mean you should. and I've seen a lot of work flying around which is because they now can, but I don't know if they should have like speaker-1: Interesting You've called yourself a designer, graphic designer, that's part of your training. I call myself a marketer. I do some marketing work. That's my job, my day job. Everybody has had ideas about marketing forever, since the dawn of marketing. And I often get I've my whole career, I've had people come up to me and say, Well, why why aren't you doing this as part of our marketing? Why aren't you doing this? Why aren't you doing this? Well, I saw this thing on YouTube. How could we should do this? And it's always been a part of my remit to say. Interesting. Not no, but interesting. Let me ask you some questions about that and let's go back to what our product actually is. And we'll hang on. What why don't we think about budget and da-da-da-da. Rather than saying no right away to people who think they're experts about marketing because we receive so much marketing in our day-to-day, I've always tried to say, interesting. Now let's go through a Socratic process to figure out why I should or should not do that. I do that with clients, I do that with bosses, I do that with colleagues sitting next to me. Now that AI can like I saw a demo of Salesforce marketing agent a couple of weeks ago, and you can type a single prompt in and have a year-long email campaign out of Salesforce's AI marketing machine scheduled to your lists, to your people, with open rates, reporting dashboards, and everything built with a single line prompt. But who's checking? That the work is accurate. Who's looking at the quality of each of those emails? And how are you how are you serving the customer if you're just putting a single line of effort into again a year's worth of work as a marketer? Anyway, I could I could keep going on this for a while, but it's why I come back to this idea of your prompt was why your organization probably needs an AI policy. I think each of us individually need an AI policy. How have you as a designer seen people come up to you more or different than they did before with their designed AI work? speaker-0: Yeah, it's interesting. I like I mean one thing I really

like how your approach to people coming at you with work w a posture of curiosity and then helping them do the work almost. speaker-1: That's hard one. I got that wrong a lot of times. I just sounded like a bitch for most of the first few years. speaker-0: No. Yeah, because I don't want a world where I'm like, okay, mate, only experts can do things. You're not a designer, don't even think about it. You know what I mean? You're not a you know you didn't go to uni for marketing, can't think about marketing. That that's not a world I want to live in. I also think what AI does is it it it forces cross pollination and now engineers are thinking about design and now UI designers are thinking about code and now the person next to me who's like sixty it's coding stuff and I'm like, this is cool. I like this world. I I think I prefer it. So I don't want to gatekeep. curiosity's good. But where am I drawing with this? I think Work needs reviewing. Reviewing needs some form of like taste. Taste expertise. Like yeah, all of that stuff. Like I think that that step's important. I'm worried about that step going missing. Yeah. in in people's work. So yeah, as a designer, I see a lot of AI design stuff floating around. Most of it's Pretty bad. Well most of it's okay, most of it's not most of it's below average. speaker-1: Hmm. speaker-0: But speaker-1: Is it a dead giveaway that makes it below average? Like, yeah, an AI built that, or is it is it not answering the brief? Ugh, a bit vague of speaker-0: Part of it is the part of it is the no good review step. Yeah. It's like they don't the person isn't a designer. they don't have the taste, they don't know the design principles, they they haven't thought from first principles to get to an endpoint because they they sort of just quickly did it. Yeah. It's I find it really useful to have conversations and like to start a conversation. And depends as well, like how hard in the paint you want to go with design. Like sometimes you just need a flyer for the thing on Monday and like I don't you don't need to like go yeah. You don't need to like I need an expert to review this. so there's like but on high stakes work, I I still think you need that layer, at least at the moment. yeah. Yeah, yeah, I see a lot of AI stuff. Yeah. speaker-1: Do you are you encouraging leaders to write AI policies? And how are you encouraging individuals to think about it? speaker-0: Hmm. I think it's interesting. I haven't had like I I think the conversations will happen. I'm noticing the pain points. I think ever I think a lot of people are noticing that and especially leaders that I've talked to are like, I don't even I don't even know how to review all this work that's flying at me. I see a lot of that. So I think it'll come out of out of pain. Yeah. and I think talk about slop, is which is unreviewed work. Or something like that. unthoughtful output. That is flying around all over the internet and it's starting to enter into like work. Slop is flying around the place. And I think organizations are just gonna find they have to they have to have a filter on the tap. They have to do something about it. yeah. That's I think I'll be surprised if in like I think in the next 12 months, almost every like I I think most organizations are gonna find have to put something in writing as they adopt AI. speaker-1: We just had something put forward at my company and it's a start. It's it's a it it says stuff like that. Own the output, you know, it's it's got some some guidance on security, when to share certain information about clients, when not to, that sort of thing. I'm proud to see the organization I work for putting something forward. speaker-0: Mm-hmm. speaker-1: But immediately I have a thousand questions. Yeah. And again, I think it's we are individuals that work inside organizations. It's on both of us. Let's be good people. Let's be good organizations. speaker-0: Mm-hmm. speaker-1: And keep attempting to do more better. speaker-0: Yeah. Yeah. Both those things are good. Yeah. Should I take a crack at speaker-1: Yeah, let's see if you pick up the one that I think is related. speaker-0: Wow, religion is the lexicon we reach for when something is too mysterious to understand, Sebastian Mallaby. Quote Religion is the lexicon we reach for when something is too mysterious to understand. Tell me about this. speaker-1: Yeah, it went lofty. I'll I'll pull up a note that takes on these really, really big, weird people and and questions. I'll I'll quote him properly in the show notes. But my next big read is the Infinity Machine, and it's about the it's written by Sebastian Mallaby. It's a biography of the current CEO of Google DeepMind. Right. And who he is as a person via interviews, via, you know, how he's come to be where he is in this incredible position of power and looking at AGI and what immense superintelligence can be. I think the subtitle is even the search for for super intelligence. But I heard this quote by Sebastian Mallaby while he's talking about the book on a book tour or an interview or something. And I just thought it was really pithy to You know, talk about religion as the lexicon. So the words we use, the the vocabulary we use to describe AI, wild, awesome, out of control, unbelievable, holy shit. You know, literally the word holy keeps showing up in some of the outputs or the the reactions to the outputs that I think people are receiving, you know. Whether it's your first prompt to AI, going, my God. speaker-0: Mm. speaker-1: Look, it's a reach for the infinite for any kind of language to do inside of church or, you know, synagogue, whatever whatever your version of it is. But we've already talked about some frustrations, some, you know, real challenges that we're facing as users, you know, arguably on the bleeding edge or close to it. But I I just thought it would be interesting to to go back to the moment of wow, of awe, and to say humans created this intelligence,

but it's about to outsmart us. I I haven't been afraid, but I also I've I think in my life and career I feel like I missed out on a couple of opportunities to be at the edge of something. And I was ready for it this time to to say, yeah, let's do a cannonball, a bomb into this Demand. Let's see what happens. And and I'm in a position at work where I can really experiment and and and have a play as well as personally. speaker-0: Can you you must have seen people around you at work who don't have that posture. Yeah. They're more skeptical, they're scared of the thing. There's also like not to go there too much, but there's like ethical reasons. There's like a bunch of reasons why you might be resistant to AI. Yep. you empathize with them like speaker-1: Yeah, absolutely. I've got endless empathy. Do I have endless sympathy? No. speaker-0: Okay. Say more about that. speaker-1: my a a dear friend's family motto is you get no sympathy from me because it lives between shit and syphilis in the dictionary. speaker-0: Brutal. Okay. speaker-1: Empathy, yes. And in fact, I had a conversation with someone the other day about he has real climate concerns about just experimenting with AI. He's like, I don't wanna I don't wanna my carbon footprint to be about hypotheses. Interesting. I don't wanna send, you know, fable five or opus five whatever model that's Just absolutely consuming huge amounts of power to spin up a landing page for fun speaker-0: Mm. speaker-1: while I figure out how to use this thing. And I'm like, well, okay, I I get that, but you know, you're gonna you gotta climb Everest one step at a time. You gotta you gotta start. So I I have empathy for it. I the the other one that I really have empathy for is people who are too busy to get lost and learn how to use a new tool. Yeah. I work quite quite closely with an executive assistant, not mine, but you know, what I work with and and she's lamented, how am I supposed to carve out not a hundred hours, not a thousand hours, ten thousand hours to become an expert at this thing? But I'm supposed to be using it to make my meeting minutes come out faster. But if I just write the minutes myself, that's gonna be faster than starting to implement this thing. And how do you how when you're already relying on somebody to work extra hours, do you get them to start it using a new tool? speaker-0: Yeah. I mean I feel quite lucky in a sense because at the time this was kicking off early in the year, I was running my own business. I I ha I had this luxury of like figuring out my own day. I pretty much just like for two months just locked myself in my office and just like speaker-1: What did you say? You said Claude told you to go to bed a couple of times and you're like, No speaker-0: Yeah, yeah. Clover's like, I think we've done enough. And I'm like, no, we haven't. What the hell? So yeah, that was interesting. And that was a luxury. Cause I'm like, you're right, there's there's you gotta do 10,000 hours. Like you And I'm like, well I don't know. So so I I do feel hard. The other thing with AI is I find the first time you do something with AI, it's more complicated. You have to think through the full task. You have to explain like it's almost like onboarding a new employee to do a new task. It's way faster just to do it yourself. I know how to do the landing page. But to sit and teach someone, here's how I like my landing pages done, here's how I like copy to be written, here's how here's how I do this, here's how I write my emails, whatever. That takes way longer. The thing is, the next time, it takes 10 hours to teach them the first time, then it takes one hour, then it takes one minute. So there is like this initial time deposit to like build the workflow, which If your time pool you don't have, and then it speeds up exponentially. speaker-1: But they told me the same thing at work when they introduced Monday dot com. They're like, put it into Monday.com board and it'll get easier. I'm speaker-0: Hm. speaker-1: like, speaker-0: I'm like a I'm like allergic to CRMs. I mean like they I get I just ugh I don't like it. Yeah. speaker-1: Yeah. if you if you hadn't had the luxury of time, how would you recommend to someone to build their ten thousand hours with AI while they are so busy with sick kids at home in winter? speaker-0: sick kids. speaker-1: I just mean like the the the mental load. I it's honestly I've I know it's on some people's to do lists to use AI this week speaker-0: Yeah, yeah. Or it's it coming from their boss. Yeah. Use AI. Okay, like well you still have I still have all the same deliverables as last. speaker-1: Or they've got a KPI now, suddenly they're being measured against. speaker-0: Yeah, look, start small. Start with an experiment. You don't need to like half of what you build you'll never use again. Maybe more, 70%. yeah, just start with the smallest part of your day. Start with the most annoying part of your day. The the thing the and the thing that you have to keep doing, you know, especially something that you have to do every Monday. Figure out how to help AI. get AI help you do that or the thing you have to do five times a week. Like, so I'd probably start there. I mean, I've I've worked at an organization where part of my role is helping people adopt AI and build their workflows. And yeah, we start pretty small. We start with can you get it to like can you teach it your tone to write emails for you? Can you get it to read an email and summarize it? Can you get that to do it every morning at nine AM and so you walk into your week with a bit of a summary. Can you like like yeah? I I now have an influ like and I I see this happening in real time. You start developing like a a reflex to reach for AI first and and that's where the reps get it much easier. You know what I mean? Like now I can't think of starting a piece of work without AI. It's it's my first port a call have a chat with April. And you'll get there, but yeah, I

think start small. speaker-1: Yeah. Can you talk through I know it 'cause we work together closely on on a number of projects, but can you talk through just a tiny little tech stack specifically how you talk to monologue or, you know, speaker-0: Mm. speaker-1: that because the I don't want to take for granted the fact that you have a process for thinking all the way through a task verbally, audibly, in a very human way, but that is also based on your professional fitness as delivering as fast as possible. I'm I'm I'm waffling a little bit, but talk me through your process of opening a note, you know, you go for the walk thinking. speaker-0: For me, there's probably look, I I live in clawed code. I still t I I'm more and more I'm just touching Claude the web app. So those those are kind of that's kind of my tech stack. Either c either I'm reaching for clawed code or I'm reaching just for normal clawed. Yeah. a lot of people will work in normal clawed or chat GPT or speaker-1: We seem to be pushing cowork quite a lot as well, but I'm speaker-0: Yeah, I still that's such an awkward product for me. I don't really anyway, in the ecosystem, I don't get it. But so that's my like I would say my daily driver to use that language would be clawed code. That's where most of my work gets done. How most of my work starts is either two things. One, it's me talking to an app called Monologue, which just turns my my voice into words. So it starts with a transcript of me explaining. And I usually start with like a I try to start further back with AI. Like I try to start with. the pain point, the goal, the brief. I'm trying to I'm trying to give them more context because I finally get better outputs that way. So I I try to start at the start of my own thought process. Hey Claude, I've got this thing. It's a project coming up. It came through this person. I'm a bit anxious about it for these reasons. but here's where I think we might be able to start. I I literally I'll stand like that. So I'm either starting with a transcript from my own thoughts or Another way I work is transcripts from meetings or, you know, talking with someone else. So I try to record everything now. If I don't capture a transcript, I'm so annoyed at myself. Like after a meeting, I'm like, frick, now I have to remember. I haven't had to remember in so long. I have a I have a device called Plaud, which I pull around with me. It's just on the back of my phone. It records like stuff on demand. yeah. So I'm usually starting from a transcript of my thoughts or a transcript from the meeting and then kicking off from there. speaker-1: And do you take that transcript straight into a new Claude Code session, regardless of whether or not it's going to be a piece of code as the output or a Word document? Or like how do you tell me where you take that transcript. Keep keep going through the. speaker-0: Yeah, okay. So I have a transcript of the meeting I just had or of my thoughts on a project or whatever.

the I I think the first part is like it's like gathering context. Yep. So the the transcript from the meeting is a really important piece of context. But I'm also like, okay, what else does what else does this like employee, what else does this person need to know to succeed in this? Well, they probably should read the annual report as well. Okay. Well, they probably need to know about this website article. they probably need actually we're gonna need images for this. Like that's where the images live. So I'm trying to pull together ingredients for context. yeah, there's like I I actually haven't written this down before, but yeah, I'm thinking about like there's like it's almost like you're baking a cake. Step one, get the right ingredients. Yeah. All right. We got salt, we got flour, whatever. and then it's like then it's like prompting it to to put a plan together. It's like write we need to write some sort of recipe here. speaker-1: So do you use flick claw code into plan mode? speaker-0: Sometimes, especially if it's like the more complex the task, the more likely I am to start with a plan. Yeah. Just like, can you just read everything? Can you get vision on what we're trying to do here? And then can you suggest a recipe or a plan for us to work through? And then I'll review the plan. Sometimes I skim read it, sometimes I read it in more detail. I might go back and forth a few lays there. And then I put it in auto mode and it just like kicks off and works. And I'll yeah. I'll tell it if I want checkpoints any earlier, but most of the time I I try and want to get as close to an output as possible, as close to like a finish line and then and then it's a review game from there. speaker-1: Yeah, cool. speaker-0: Is your process any different to that though? speaker-1: Pretty similar. I'm still struggling. Cause at at work we're We've invested heavily in copilot. We're in we're in the PC Microsoft world, which I'm a Google Mac native my whole career. This is the first company I've ever worked for. I'm just I struggle with SharePoint OneDrive, but like everything's called copilot. I don't get it. I also sometimes find myself still figuring out, am I in a Google Doc version of Word, or can I open it on my desktop? Are they different? Whatever. It's just I'm I'm writing with my opposite hand all day, every day. Anyway, I use the Claude terminal. I also use Monologue. It's a personal subscription I pay for. I then have to copy and paste the transcript. I hope the head of IT is not listening, but I paste the transcript, Slack it to myself, and then copy it from Slack and put it into my Claude at work. but Yeah, sometimes I'm st I I'm still struggling to direct Claude that I already know at a corporate level is po is plugged into different parts of our SharePoint. So I'm trying to understand what context does it have already that's given to us by the corporate account and what do I want it to just focus on. So most of the time I just treat it as you do. I say go get the right ingredients for the cake. Here's the image folder, here's the, you know, last four web pages we developed. speaker-0: The thing Yeah, I I like to have like pretty granular control over context. Okay. But like just because I don't if I plug it into a SharePoint or a Google Drive, I'm just worried it's like gonna go read other stuff. And I'm like, I don't don't read that. Like that'll I don't I want you to read these specific things. Yeah. Cause I'm trying to it's almost like Like it just like has full access to the pantry. Yeah. And it's like, Vegemite, like that might work. It's like don't not. Like don't if unless I tell you to. so I find that, you know, I I don't know how the whole SharePoint thing works. Does it it can just go fetch anything in SharePoint? That's terrifying. I guess it just doesn't though unless you tell it. Yeah. speaker-1: Yeah. That's terrifying which Yeah, and and I what I've done well, now that you've got April, the chief of staff, doing talking to sessions for you, I'm gonna to rethink my whole structure, but I've given I've pinned certain sessions in Claude Code. knowing which context those sessions have. Yeah. Knowing, okay, I've given it all of these engineering documents. So it'll only look at these when I ask it to do this, or this one has these and these approved policies kind of thing. And it'll it'll go and look in that. And then I've got what is named broad context. Right I've got a yeah, I've got the the broadest or the widest context window where I think I've given it the most information and I go in there and just ask it a question if I can't remember, how many parts per million or whatever I need to search and and go and get the the actual answer and ask it to cite its source and it tells me it's from here. So I th I I need to now figure out how does the chief of staff reach into those different context windows to give me a piece of information because what I work with on a day to day is so technical. I can't get a ki if it's parts per million or parts per billion, that's a big difference. But I also hope that Claude's not gonna ghost me and just give me you whatever. Yeah, yeah, yeah. So I need to I need to understand from a human, probably our head of IT, how is how have you set this up at a corporate level between SharePoint? How should I be thinking about it from a context feeding perspective? But then as a as a bit of a control freak. Myself, I'd do exactly the same as you and say, actually no, scrap everything you know. Here are 10 documents I want you to be aware of, re-upload those PDFs from my filing structure into the session and go, okay, now I know what you're pulling from. Write two lines of copy. speaker-0: Yeah, yeah, yeah, yeah. Yeah. I speaker-1: And then I rewrite them myself. speaker-0: do think as people work with AI more and more, they'll realize half of their job is context. Yeah. Like it yeah. speaker-1: It's all my speaker-0: It's your responsibility to like

gather good context. When I see people blame the output for not being bad, like not being good, I'm like, so it's I can't explain how much this is your fault. It's not magic. It's not. I mean, I know we can reach for religious and mysterious language, but in terms of like the actual nuts and bolts of it, like how could it possibly know? You didn't tell it. Yeah, yeah. speaker-1: That is your sin. It's original sin. It's you blaming a robot. speaker-0: Yeah. Yeah. Yeah, I mean that and that's the same thing of a leader blaming all of their employees suck. All of your employees suck? speaker-1: All of them? Yeah. huh. speaker-0: Yeah. speaker-1: Religious and mysterious language. Yeah. Yeah, we're we're still there. speaker-0: I'm gonna pick up another prompt. Okay. This prompt says Albo's speech about Australia's emerging AI framework. That's interesting, yeah. Cause I mean, immediately I think to the fact that South Australia just announced a Royal Commission into AI. pretty big news. I think it's a world first. I actually don't know Albo's speech. You can tell me about that. speaker-1: Yeah, it was given less than a month ago. He gave it at the University of New South Wales, where he's a I think he's it's his alma mater. it's the first time I've properly listened to him speak. I'm obviously not from Australia, but I am a permanent resident. I've been here three years almost. And Anthony speaker-0: Albanese is our prime minister. speaker-1: Albanese. Yes. Albar. so I'm a I'm a new Australian. Hopefully gonna become a citizen actually in about six months. Let me see. Stay tuned. and but there's we've talked about this at length, you know. I've as a recovering American, there are a lot of things there are a lot of things that speaker-0: the moment. speaker-1: I really, really love about Australian culture. sobriety and Australian culture is not one of them. No. the so it's the first time I've listened to Albo speak. From start to finish of a thought rather than just a snippet on the news or you know in a waiting room. I th I thought about putting Peter Milanowskis, who's the Premier of South Australia. I thought about putting the Royal Commission in as a second thing, but I didn't want to go too political right away. I'm really proud to live in Australia at the moment for a lot of reasons. The social media ban basically originated here in South Australia. The world has been looking to Peter Milanowskis For his leadership and his stance, you know, say about him what you will. He's done something that no other politician was really willing to do, which was to go out and say, we are gonna get this wrong, but we're gonna start. And I think in a way, Albo has done the same thing with this framework for how Australia is going to regulate artificial intelligence. And it's definitely worth it. It's a 22-minute speech. Listen to it on 1.5 speed. I was hanging a load of washing. It was on my phone on YouTube, but it was great. And he made a couple of points

that I thought were worth discussing because it's he said, We're not gonna get this right, but we're gonna try. He said, We're setting out absolute ownership over creators' creations. You will own what you create, whether you're an artist or a producer or a thinker. Australia is going to regulate how large language learning models learn from what has been created speaker-0: Hmm. speaker-1: and either figure out some sort of royalties or way of compensating people for the work that is training these models, because that's not what's happening at the moment. You know, in the US it's just We're the technology is racing ahead of regulation framework. And I think Australia is watching that going, hang on, we're we're a middling power to use the carnies. The middling powers are the ones that get to lead at the moment in some of these in these policies. There's another great speech at the World Economic Forum 2026, I think it was. But it's a chance to say, we're gonna make what people create theirs. We're gonna regulate how The LLMs are learning from that. But then also in the speech, he talks about something I love about Australian life, which is we're a huge country. We have an opportunity to create massive data centers and bring in a lot of income for this country. And he goes through and says, We're we're already looking, I think in the Murray River, they're looking at putting a huge data center. On the Murray River because the water is going to cool it. But Albo's saying we're gonna make sure that any data centers that come in are net generators of power. We've got the sunniest country in the world, but we're also the driest continent in the world. So we're gonna make sure that if they're using water, they're using it responsibly. And it we're going to regulate to make sure that the cost is not passed on to the average Australian. So it's an opportunity for this government to say, yes, let's. Absolutely welcome this, but let's look after our people while we do it. And he cited things like when superannuation came in in the nineteen eighties, I think it was, or the nineties. Here's how we looked after ordinary Australians as we went through this. You know, it's really, really worth a listen. But I thought it was it was a proud moment because I can tell you that the leader of the country where I originate from is not saying things like this. It's Yeah, I'm proud to be an emerging Australian after listening to that speech, even though there's a lot of pitfalls pitfalls. speaker-0: Yeah. I r I really do respect the courage to do something. Yeah. Like emphasis on courage. I think I think it's pretty pretty ballsy. Yeah. The I also think someone has to draw some line somewhere. Like it can't just be pure Wild West. And like, you know, I I noticed this when there was a big trend and like this is how some people were first introduced to AI was When ChatGPT came out like a certain model of image generation, it got really good at basically copying. And there

was one artist in particular, I forgot his name. He makes like he he's like famous for doing like anime or more particular manga style. I forgot what it was called, but everyone was prompting, Hey, can I have a photo in the style of this guy? So billions of photos were generated. And I'm like, well, that's kind of rough. Like this guy puts his puts his life's work into creating a certain art style that he's known for. And then that gets fed as a tr into a training model and now I can replicate it. I that that sucks. speaker-1: Eck. speaker-0: That's like he should get of course he should be like he should get some some sort of commission or something. Like I don't own this. You know, and I mean that's a difficult one. Like the training data is people's work. Yeah. It's pi it's books people wrote, it's articles people wrote, it's artwork they wrote, they made. The idea that now like I can just access and copy it. I don't love that. Like speaker-1: Yeah. It's the speech is an interesting one for anyone who has not satisfied themselves to think through the implications at a societal level of AI changing our landscape. I think Albo touches on a one one enough, okay, what's gonna happen in the world of defense? How's Australia gonna respond? what's gonna happen in the world of artistry, in industry, in okay, how are we looking after our people, our healthcare or whatever. So it's it's a it's a great it's a great start. And I thought I'd bring it to the table for for those discussions. speaker-0: My my last note is is relevant. Can we Yeah. speaker-1: Shall I open it? Yeah. speaker-0: Yeah. I wanna tell you about something. speaker-1: Prompt number three, the giraffe problem. speaker-0: This is something I've been thinking about this morning at the gym. It's fresh. You're gonna have to let me run with the ball for a bit. I'll promise I'll somehow link it. speaker-1: Nah. speaker-0: Somehow. Okay. Giraffes, right? They giraffes are herbivores. They've evolved to have really, really long necks. Because the longer your neck, the higher food source you can reach. Therefore it, you know, promotes evolutionary fitness. So over time, giraffes' necks got longer and longer and longer. It's adaptive. I'm curious about how some things that are adaptive in some environments or in some ways could become maladaptive in a lot of other ways. So for a giraffe, for example, I am I did some research. Can I tell you about the trade offs of having a really long head? speaker-1: You really can. speaker-0: So okay. One, when they drink water, they have to lower the head really slowly, otherwise they'll pass out from the pressure change, right? speaker-1: Being six feet tall, I'm familiar with the problem, but carry on. speaker-0: Yeah, they're like they're like two story building tall, right? So like when they swing their neck down, if they do it too fast, they'll just like pass out and then they'll just like dr fall into the water. Same on the way up. If they flick their head back up, their brain will like explode from the pressure change. So that's a trade off. Another trade off is when they drink or when they sleep, their the floor is just so far away. They have to like splay their legs really awkwardly. I don't know if you've seen that before. And basically it takes like fifteen seconds to get up from that position. So a predator just waits for them to have a drink. And then they're like screwed. So then they just get like torn to pieces. Another trade-off I is they have this inbuilt lag because they have like four meters of wiring between their brain and like their legs and stuff, there's actually like this inbuilt like latency problem of like it just takes them a long time to do everything and they can't they don't have quick reactions. So again, predators take advantage of that. The last thing I think was to get there's like a plumbing issue to like pump blood up to their brain. They they have to have a heart, a heart that's like 11 kilos or something like that. And their blood pressure runs like three times what a human does. And so they don't sleep basically. They sleep for like half an hour a day. Okay, so that here's it. That's something interesting. So you have it's adaptive to have a longer neck. It helps you survive, but it's maladaptive in like seven other areas, right? And I think if I if I if I like jumped into evolutionary history and I talked to a giraffe and I said, look, I'll give you like three inches of neck, but like you're gonna pass out sometimes when you drink and you know like I don't think it would have taken that trade off. I think it would have went, I'm good, like I'm ha I don't I don't, you know. So okay, why am I bringing up the draft? And what what has this got to do with society and AI? I think there's some things that are adaptive in the near term, or there's something there's something like that near term success promotes. Which is maladaptive in the long term or maladaptive in other areas of your life. I think from an organizational standpoint, it seems like what's going to be rewarded in the near term is adopting as much AI as fast as possible. and the and you know, the companies that don't are gonna be punished. And so there's a real adaptive reason to push AI as far like hard and fast as you can into your organization. But then I'm just thinking about like What are the what are the malad maladaptive like what other areas does that affect? And I think like I think I like Albanese's speech because it is having the courage to think about the fallout or think about the trade-offs. If if whereas as a society we just adopt this as fast and hard as possible, what does that mean in five years? What does that mean for an average working Australian person? What does that mean for the way I work? And so I yeah, I and I I wonder like the thing that I keep coming back to is like I don't think I want to be a giraffe. I like I don't think I want to have all my stat points in in neck.

Like Yeah. I think I care about multiple things. I care than I care about more than productivity. Care more about I don't know. I so that's yeah, that's my that's the giraffe of them. speaker-1: We landed the neck. It's yeah, completely related. How do you feel what are you what are you doing to mitigate your rapidly growing Nick? If you're if you're adopting AI, you've been hired into a position to literally experiment with this thing in a large company. Do the reps, share the learnings, and get other people on board. It's it's an amazing brief and it's such a cool, cool thing to be doing at this moment in time. But if that's what you're doing in your work life, how are you? Are you slowing down? How are you speaker-0: How am I slowing down? I'm not slowing down. I okay. in the 12-step program, is the first step like awareness or something? Or like I think I'm at step one of like just awareness of like, hold on. speaker-1: Yeah. speaker-0: My neck's getting kind of long here. Like I don't know if yeah, there's unintended consequences of working this way. It's hard as well though, because at what At what level do you solve that? Like there's a level of the individual and there's a level of like the workplace or your frameization. Yeah. There's like the community, there's the nation, there's the world, right? And there's unintended consequences all the way up. Yep. I'm also interested in how if you add up all of the individual decisions that seem good and I can I could justify every use of AI I've ever done, and I will be able to. Every organization could justify if we don't adopt this now, we can't compete, blah, blah, blah. If you add up all of these good decisions, they can lead to a bad outcome. Like and I'm like, I don't know how to solve that. Like, what do I start making locally bad decisions? yeah, I don't know. It's a context problem, I think. speaker-1: Context problem. How are you how are you gonna mitigate it? How c how does one mitigate future outcomes? speaker-0: You're try you're holding me still You're trying to like turn this into like like how you gonna speaker-1: Life the universe speaker-0: be responsible for your thought? Yeah, I'm not gonna be responsible at all. I think speaker-1: Maybe you won't be responsible, but you're here doing this in front of a microphone. You're already accountable. You're willing to take accountability for it. speaker-0: Well I think you you it speaker-1: Okay. And you're asking other people to ask those questions. Yeah. So I'm letting you off the hook, by the way. speaker-0: Thank you. I appreciate the the hook letting go. Like I think okay, so the giraffe didn't choose to have a neck that long. It didn't choose to have this dysfunctional build. It happened as a as a result of its environment. If I asked a giraffe at any point, I don't know. I don't know what a giraffe would say. But one thing's like we do have as humans is we can transcend our environments. We can make decisions. We can think through the consequences. We can think long term. We don't just have to live and die by what by survivability tomorrow. So that's the with that comes a responsibility to to think and to think longer term. You know, like one thing I did say, 'cause I've started doing work with organizations with AI and I'm having more and more conversations with business leaders. I remember one line in the scene I did draw was, I'm more interested I'm pretty much only interested in empowerment. I I don't want to have conversations about replacement. If someone comes to me and says, Yeah, I like if someone comes to me and says, Look, hey, I I want to replace my accounting team, can you help me do that? I don't that's so that sucks. I don't want do that. Yeah. I don't know. So that there's a that's something. I'm not you know, I understand that empowerment, getting more productivity out of people is sort of the same thing as I don't know. Like it's not the same, but it you know, it leads to a ni similar outcome. So I'm not speaker-1: I think it's a I think it's a fair line in the sand to say no to business anyway. speaker-0: No. speaker-1: w well, for you to have an internal measure of what you say yes and no to. And I think employers and companies have been trying to get more out of their people by giving by enabling them with technology since the invention of the hammer. Yeah. And I'm stuck in a thought loop, you know, going back to this idea of my company's paying for my Claude tokens, but I'm accomplishing a quarter's worth of work in an afternoon. My salary isn't being revisited. Based on my outputs. If I paid for my tokens and then submitted the receipt and then asked for a raise, does that work? Do we revisit what my remit is? Do we revisit how many people report to me? How do we, how do we adopt adapt the working relationship to account for this exponential jump in speaker-0: Mm. speaker-1: productivity or quality or intelligence and continue to collect a salary and make superannuation contribution. I d I don't know. I'm I'm I'm extrapolating a little bit, but the I think it's important, again, going back to us as individuals, to create our own policies on what we are and are not willing to do with this at our fingertips. You as a as a knowledge holder and a person who's being hired and offering to work. with AI for other organizations. I think your your line is clear. I'm still trying to discover what mine is in my day to day. speaker-0: At every level, level of the country, at the level of the community, at the level of the family, at the level of the organization, at level of the individual, lines need to be drawn somewhere. Otherwise, we just end up with really long necks being like, we can't drink anymore. Yeah. Like that. And it's I don't like the feeling that that's inevitable. And I'd like to think it's not. Some days it feels like it is, some days it feels like. man, this train has left the station and there's like no slowing down and like Yeah. But

But we're humans. We can think we're not giraffes. We're not just like a a victim to evolution or a victim to our environment. We can transcend it. We can think long term. We can draw lines in the sand. We can make twenty-three minute speeches about having a crack, even though we're gonna get it wrong. So there's some hope there, you know. speaker-1: Yeah, I agree. Last one. speaker-0: Last one. The moment we're in. Logging a ticket with IT to install a skill took longer than building and landing a web page. Tell me about this. speaker-1: It's just a a recognition. Yeah, I just I wanted to be really human for a second and realize we have one foot in the future and one foot very much in today. Yeah. And while we're talking about breakneck speed of advancement and all these opportunities for things that, you know, AI can can help us do and zooming out and zooming in and looking at the forest for the trees. That is an exact articulation of where I find myself. Yeah. I could not log a ticket because the IT system was down. speaker-0: Yeah, yeah. It's really weird. It's weird it's weird running at different paces. Yeah. You know, in one moment you're going like a thousand kilometers an hour and then you like go down to like one kilometer an hour. Yeah. In like the same hour. I I yeah, it's real weird. speaker-1: My my closing question for today was gonna be what's keeping you up at night? This code switching, sorry for the pun, the the the context switching between how it was and how it is, how it will be, is what's waking me up at night, besides my 10 month old. It's if I if the ticket I logged with IT isn't giving me the clawed skill to edit PowerPoints more aggressively, the deadline for that update of the PowerPoint that's gotta be automated onto the website is gonna fall down. All these domino effects are gonna fall down. Or I could just vibe code the whole presentation in HTML and print that out as a PowerPoint. And you know, like how do I that context switching and code switching with KPIs in my J O B is brutal. And, you know, this this urge and desire to be on the bleeding edge of this. speaker-0: Mm-hmm. speaker-1: is so strong. But I also want to slow down and throw my phone in the sea most days and just sit with my ten month old and watch her take things out of a bag and put them back in again. Yeah. speaker-0: It's yeah, isn't like one thing I've noticed at work is like fast things are so many fast things are dependent on slow things. And so you like speed run the thing that used to take a week, but then you're still stuck for a week on the like yeah, it's real weird. It's it's weird, like and then your day becomes like how the hell do I speed up these slow things? The fast thing's easy. Now like whipping up a landing page is That's nothing but like getting someone to prove the copy or thumbs up that you're allowed to do it or I don't know, install the skill like you said, like those just take as long as they used to take. Yeah. You know? So it's this like fast and slow work, you know, I think we're all adjusting to. speaker-1: Maybe this is the antidote to long necks is Catching up the human systems to where the machines are going. speaker-0: Yeah. speaker-1: Government is slow. It's gonna get things wrong. Yeah. But it's a control in addition to our individual controls, our capitalistic endeavors, our whatever other controls are in place to slow down the speed. But we're not we're only talking about this in such a small application. You know, we've only got limited experience in marketing and design and organizational design and and speaker-0: I wonder, yeah, there's almost like this instinct to like speed everything up to the speed of AI. Maybe something's meant to be slow. Yeah. Maybe that's okay. You're listening to Subliminal, a podcast that's nominally about AI, but mostly about everything else. speaker-1: If you'd like to hear more from us, visit wear liminal.com.au and prompt us.