Episode 05 - More Human, Not Less: Teresa Torres on AI, Agents, and the Future of Product Work

All right, guys, welcome to another episode of the Victor Cho podcast. I am so incredibly excited to be, I want to say, continuing a conversation that I had with Teresa Torres because she and I had just a great session the other day and she was gracious enough to agree to join and effectively continue the conversation on the podcast. So, Teresa, thank you so much for joining. For those of you that don't know Teresa, she is many things, an author. a coach, an instructor, an entrepreneur. She's the founder of Product Talk, which is probably the preeminent source for product discovery, product insights, author of the best-selling book, Continuous Discovery Habits, which is an Amazon bestseller, and sits on boards. She's a very busy woman, but at the very highest level, I think of her as, well, I'm just an amazing person, but one of the best product minds on the planet. So Therese, just a pleasure to have you here to have a conversation. Thanks, Victor. Very kind intro. Yeah, hopefully I didn't get any of those wrong. Wonderful. So let's start by just talking about, of course, AI is top of mind for everybody. And of course, yes, it changes everything. It changes every function field. I would love to start the conversation around. how are you thinking about this world of AI vis-a-vis product, product discovery? Because I know you're in the middle of it. So yeah, let's open up for discussion on that topic. Yeah, you know, it's funny. I have jumped in full tilt on AI, but not really with AI and discovery. And people ask me all the time, like, why aren't you writing more about AI and discovery? Why aren't you doing more with AI and discovery? And the answer is that I actually don't think discovery is the place to apply AI. There's some exceptions, like all these AI prototyping tools are amazing and it helps us get feedback on our ideas a lot faster. And I love that use case. I wrote a big blog post where I interviewed 11 different people about how they were using Lovable in their product development process. And we collected some amazing stories. So I love that use case. What I see a lot of teams doing is they want to not interview customers and just like use Gong or any of these other AI products that are just like taking customer feedback and telling them this is what your customers want or aggregating sales calls and saying this is what your customers want. I actually like all of that, but it does not replace customer interviews. And I don't, so that like realm concerns me. You know, here's what it comes down to. I feel like for those of us that like are building things, unless we're building software for one ourselves, I have this like really core belief that if we're going to build something for somebody else, we probably should spend some time with them. Like, that's it, you know? And that's a proxy. How do you feel about this? I heard this crazy idea, or maybe it's genius of, oh, we're going to, we're going to. create fake human beings. Yeah, I don't have to go through all the mess of interviewing a real person like we'll interview an AI person. And when I first heard that I'm like, that's not going to give you the insight. But like, what's your Okay, so there's like, there's like dozens and dozens and dozens of research papers on this topic. Now. I'm furiously trying to keep up. And this is why like, people want me to write about this. Yeah, it's all of it has a little bit of a yuck factor for me, like if we're building four people fundamentally we should spend time with those people and ideally we should be co-creating with those people like i would hope that we're starting to learn that we can't just like jam our software on other people right we can't just be like here victor use this you have to like the better approach is victory let me understand what you're trying to do let me understand your context yeah let me build something that like tightly fits with what you do not with what i think you do yeah um Yeah, that doesn't mean AI can't be useful. And like when I first heard about these synthetic user companies, my first response was like, this is ridiculous. Like, how do you want to build for people and not really understand who those people are? Like I just still that makes my skin crawl. I don't really get it. But I'm coming around a little bit. I think some companies are using those synthetic users as like a first gate. So we have 50 ideas. Which ones should we pursue in discovery? Super interesting. So you think there might be a role for it? I mean, I don't want you to explore 50 ideas. I want you to start with a customer need and explore a handful of ideas. So like, I still have some challenges, but I can see where maybe there is a role for these types of products and services. So I like, I don't want to poo-poo them. What I will poo-poo is if you're using them to replace actually talking to customers yourself and being curious about your customers and learning from your customers. I would really like, I know this is going to sound harsh, but I would ask like, why are you in the business of building for other people if you don't want to be getting to know these people? Yeah. No, I, I, I love that. I love that sentiment. Um, what do you think about, um, there have been a lot of, uh, I think prominent tech execs who have basically said, and let's take the software spirit, cause that's the world that I operate in. Right. Where, you know, their claim is look like. everything changes in an agentic world because human beings aren't doing the lion's share of interaction anymore, right? And that's not happening today, but you see the pieces coming in place with these agentic systems, right? And their rationale, whether it's Elon Musk saying, oh, I'm going to replace Microsoft with a bunch of agentic systems. Like, software is dead. Or traditional SaaS software is dead. How do you think about that from a look at if assuming we do move into a hyper agentic world in the future where a lot of those interface interactions are taking place by computers, not humans. Like how does that, how do you think about that in the world of building? Like now you're building for some weird agent human hybrid. Like I don't even know how to think about that. I mean, if that world really does come to fruition, which I think there's a big if there, maybe it's agents building for agents and we're not involved at all. Right? Like if we really are getting to a world where agents do everything on our behalf, well, why not just let the agent decide what to build for the agent? Like, and I think what's wrong with this vision is every new technology in the history of time. At the moment that technology emerges, there's these horrible forecasts of everybody's going to lose their jobs. Technology is going to replace everything. We no longer need accountants because of spreadsheets. Don't you know? Whereas what happens in practice is that I think technology allows us to abstract one layer up. It removes some of this tedious work. I still think humans are going to be involved. And like I said, I don't want to sound like an AI naysayer. I spend all day every day in cloud code. I'm a huge fan of AI. But I just, maybe? I also know not to predict the future. So that's the other caveat where I'm going to say a lot of maybes here. I'm more of the belief that AI is going to augment our work and not replace our work. But who knows? I do think this is a big enough, new enough technology that it could radically change everything. The internet radically changed everything, but we still have jobs. Yeah. So at this stage, if you were guiding product teams who are worried about this kind of agentic future, would you say, just keep an eye on it, but there's nothing you really need to action at this point in time? Or would you be taking... Like more proactive steps to get ahead of it. Assume, you know, if it does come, you know, it may come fast. I do think today product teams need to be prepared for an agentic world. This is already happening, right? Like if you're, like I ran into this, I'll give a personal example of this. I run an online course business. We train product teams on how to do good product discovery. I have a course website. My course website is hosted on a third party platform. The way that third-party platform shows courses is it dynamically creates the page. So all of my course titles, descriptions, pricing is not available to an LLM. Huge problem. Huge problem. Right? So what I had to do was I had to build static pages with that information specifically for LLM crawlers. Right? And so this is like, that's such a boring example, but this is the kind of stuff that product teams need to be aware of is. First of all, does our product have an API? If it doesn't, we probably need an API because there is going to be a world where agents are going to use your software, no matter what your software is. We are today in a world where if you have an API, agents can start to use your software. And I saw this last spring, like in April on Product Talk, we did a big series on APIs because I think every product team is going to learn how to add an API, how to have a well-designed API. You know, MCP is super hot right now. Some people are starting to sour on MCP because a lot of MCP servers are really poorly designed. It's a whole nother skill that I think product teams need to be aware of and to understand that it's not just a wrapper on your API because your API was designed to be used by deterministic code, which can process lots and lots of data. MCP is used by an agent. And if you spit out lots and lots of data, you just blew up the context window, right? So like, What we're talking about right now is not the future. It is the right now. And so product teams have to be aware of this. Like your software is going to be used by agents. It's going to be used in employee workflows where they're using things like MCP. Are you prepared for that? And what I like about that stuff, you don't have to predict the future. Like this is where the William Gibson quote really comes into play. The future is already here. It's just unevenly distributed. So find your customers that are living on the edge. and start building for them because that's where your market is going. Yeah. I love that advice. Yeah. Just in terms of how we think about our own product pipeline, as of at least two years ago, I gave a message to the product team of like, yeah, we're going to spend a bunch of time building interfaces for users. But we need to keep in mind, right, there is going to be a future world where probably the core value of what we build is going to be something that's consumed over effectively, right? Either an API pipe or some agentic system. So we can't build something that doesn't take us in that direction, at least, even if it's not here today. I also think it raises really interesting product questions that I don't think we know the answers to, which is if you build out an MCP server or you make your software available via API to agents. You're no longer the front door to your product. How do you build brand loyalty? How do you not become a commodity product? How does your pipeline not get replaced by your competitor's pipeline just because it's cheaper? And I think these are really important product questions to be asking. Like, what do you offer that isn't a commoditized product in an agentic world? Yeah. No, no, I had, I had, I was, if you think about like the, just the behavior on the internet today of like how, how much of it is people sitting in front and. moving mouses over a UI, right? There's so much energy to build all that stuff. I think the weight of that as it shifts more automated is going to be super interesting. You know what's interesting about that though, Victor? And this is why I feel a little yucky about the synthetic users. Like, I don't think the answer is going to be just build for agents. If we just build for agents, we're setting ourselves up to be a commodity, right? Like if we're just a pipeline that an agent is using, what's the difference between your pipeline and my pipeline? Yeah. I think... what is going to differentiate us? I don't have the answers, but I can tell you the approach is going to be become more human. Find a way to connect with your customers in a very human way, in addition to the agentic way, right? So that you can still build out that brand loyalty and not be commoditized. Yeah. Oh, you're preaching to the choir here. That is literally what we're building over at Movid. Yeah. Maintaining human connection in an AI world. What shouldn't change? We talked about a couple of things, right? One is like, yes, let's actually continue to converse and go deep with our customers. Are there any other first principles or just foundational things, regardless of where the world goes, where you're like, no, if you're going to be really good about product, these things are going to change? I think the theme for me, I remember I wrote a really early blog post about how to use generative AI in discovery because I was just getting hammered with questions. As I was writing that blog post, I wrote a sentence that I think today still really resonates with me is that, technology needs more humanity, not less. Yeah. Right. And I think, um, it's really easy to start to think like, I get it. Discovery is hard. It takes time to talk to customers. It takes time to recruit customers. It takes even more time to synthesize what in the heck did we learn from these customers? Like it is hard work, but I think it's some of our most important work. And I really, I'm starting to be a little cynical that maybe the market is not where I am. Like, I probably may be wrong about this. Like I think companies are going to willingly outsource this to AI. But I hope like I'm going to plead with individual people at those companies. Don't let that happen because I feel like human connection is going to become even more important. Having empathy for your customer is going to be even more important. Really understanding how you're going to differentiate is going to become even more important. So I do think a lot is going to change. And I think even our discovery practices might change, but at the core, I hope the thing that never changes is we remember to be human with humans. Yeah. No, it's music to my ears. There was a gent that I was chatting with who's had an interesting prediction around human connection and services and products, which was... AI will effectively kind of commoditize large chunks of what we do and make it incredibly efficient. So actually the case study was education. They were like, there's going to be a lot of really good AI driven education, probably maybe even at a higher quality level that we have today at a fraction of the price. And then there will be this super premium that we put on people. And his prediction was that no, folks that are super wealthy will have some of that, but they're going to pay to have the white glove human connection. And he had the same thought around like, psychiatrists or counselors, right? He saw this bifurcation of cheaper AI-driven versus kind of human connection at a premium. I didn't know if it was right or wrong. I thought it was an interesting premise. We're already starting to see some of this. I don't know if you've seen on LinkedIn, there's been all this buzz about the fact that both OpenAI and Anthropic are paying pretty healthy salaries for content writers. No, I haven't seen this one. Like 300, 400 K a year for content writers. And it raises this question of like, why aren't they just having their LLM write their content? And I think there is this recognition that like, if you want to create content that resonates with humans, there needs to be a human behind it. And like, maybe LLMs will write all of the like product documentation and how to content. And that's great. Let it do that. It's good at that. But I think humans crave connection with other humans. And I think we're starting to see writing that has a human behind it has a much stronger connection. And so even our LLM Foundation labs are paying a ridiculous amount of money to have humans write things, even though they have these free writing machines. Yes. No, that is so funny. That's amazing. And also, I don't know. That's a hefty salary for a writer. That's going to be like a writer's nirvana role. It is going to be a writer's nirvana role. As a writer, it makes me a little happy. No, that's amazing. That's amazing. Where do you think we'll ever get to a point where a completely AI-generated product ground up? Like, no human invention, where this product was the brainchild of, or maybe it's featured, but ultimately product. Do you think that's in our future? I'm curious. Yeah, I've been thinking about this. Like, you know, some people think I'm going to apply it to another domain first, then I'll try to come back to product. Some people think that like an LLM can never discover like a new scientific discovery because it's trained on the average of what humans know today. So where's this like new insight going to come from? Well, it turns out, I think just last week I read that Cosmos, this like science-based LLM. did discover meaningfully new science. So I don't think it's true that like an LLM can't produce a new thought. Yeah, completely agree. You just have to ask, you have to just give it the right, you have to ask it the right question, right? It can diverge like a maniac in terms of creative thinking. It's yeah. Yeah. And so like if an LLM can produce new science, like why can't it produce a product? Yeah. I also don't, like I subscribe to the school of thought of like, If folks, if listeners aren't familiar with like the philosophical mind brain debate that's been going on for centuries, I'm solidly in the brain camp. Like I don't, I'm not convinced that there's something unique or special about humans in this like biological sense. Like I don't, I personally believe consciousness is something that emerged from our biology and there's not this other, and I realized for religious people, this is a hard concept because with religion, there's this sort of spiritual soul component. But I do believe like the things that make us human, that we feel makes us unique consciousness and our mental experience. I personally believe it's emerged from our, like literally our wet biology. And starting from that foundation, I feel like there is no reason. Now maybe transformers aren't the right architecture, like that's to be determined. But I think there is no fundamental reason why a machine could not think, and produce the same thing a human could. So I do, but like getting back to your question, I do think we could get to a point where a machine does build a product completely from scratch. I think what's interesting about that is I think an LLM today could just build a product completely from scratch. The question is, does a human want to use it? And is it built for a human or is it built for other agents? And then my question becomes, how does that... LLM do discovery? Who's it talking to? Is it talking to other agents? Is it talking to humans? Because I don't think that this nature of like, we're going to just have a spark of an idea and it's going to magically work. I don't think that's going to change. I think fundamentally, there's still going to be this problem of like, who am I building this for? How do I make sure it works for them? Whether we're doing that work or LLM is doing that work or some next new technology or an alien is doing that work. I feel like fundamentally, there still has to be this match between what we're building and what somebody needs. Yeah, no, that makes a ton of sense. And I think we share the same perspective on kind of what emerges from super complex systems, whether they're biological, right, or digital. It struck me the other day, I should probably come up with a term for it, but like, it's funny, the things that we would need to build into the AI systems to make them seem more human are the exact same things that would make them way more dangerous. Right. Let's give it, let's give it a motivation engine. Right. Let's give it a desire to continue to increment towards some kind of progress. Right. These are, right. Let's give it the ability to modify its own wiring real time. Right. As it goes, like the things that I think will lead us to more ambient innovation are also the things, yeah, that we're like, we probably shouldn't build those in. It makes, um, like Asimov's work so prussian. Like, yes. You know, like even just his three rules and like this idea of like no harm to humans. Like I don't, I also don't fully prescribe, I'll say I don't fully prescribe to the, like the doomers who think that like, Genre of AI is going to end humanity. Like I think there's an assumption there that like intrinsically AI will want to survive at the cost of us, which I'm not, there's a few leaps in there I don't quite get. And maybe they're right. So like, I don't want to dismiss them, but like, I don't fully understand it yet. I also feel like we could also just unplug the machines. So like, I feel like we have some safeguards. Yes, exactly. Now, when we give it control of our whole power grid, maybe we lose that and maybe we shouldn't do that. Yes. Right. Like, but thankfully, I think we have a long way to go. And I have a feeling like for you and me, this is not going to happen in our lifetime. And I, again, I shouldn't predict the future. I could be wildly wrong, but I have a feeling we have some time to figure it out. Yeah, man, I completely would you. Yeah, I think the, there's a fairly simple set of controls that you can put around. We have prisons for sociopaths and bad people. Like we can put boxes around these things unless we are stupid and we let it go outside the box, right? That would be our, that'd be our issue. Um, so speaking of agents coding, I know you've been going deep into, and you've been having a ton of fun. with things like Lovable and just kind of exploring. But we'd love you to share kind of some of that journey with folks, because it was so fascinating when we chatted last. Yeah. How are you using this? I think with a lot of people, I started in the web browser just using ChatGPT. I had some really early, like really early kind of pivotal moments. One that stands out to me is when Hamas attacked Israel on October 7th. What was that? 2023. I woke up and read the news and was like, wow, I really don't know very much about the Middle East. And like, I got really curious and I wanted to learn more about the history. I mean, I knew that like, I knew enough, I knew like the broad strokes of like World War II and coming out of World War II and Israel and whatever. But like, I really just, you know, our whole adult lives, we hear about what's happening in the Middle East and I just wanted context for it. The challenge with that topic is you can ask questions that offend people without knowing it. Right. And so like, I didn't know how to learn about it. And so I just sat down and I chatted with chat GPT about it. And I know there's lots of flaws with this. Like chat GPT is not unbiased, but like, it was a really great starting point. And it allowed me to ask what I think were really like naive questions. So like we started discussing the history of the Middle East. And one question I had was like, tell me about the origins of like, what is Semitism? Like, where did this word antisemitism come from? And like Semitism and anti-Semitism is not specific to Jews. And so then that was confusing to me. I was like, how did we, how did this word get applied to just Jews and not like, and that like all of that is very complicated. And like, what a fun system to be able to get down to that level of curiosity of like. okay, but why do we use this term antisemitism and where did that originate from? And like, oh, interesting, that term doesn't just apply to this group, but colloquially we use it to apply to this group. Where did that come from? And then I got really curious about like, okay, all the things I just learned about this region, Jordan is right next door. Why doesn't Jordan have this, like in this quagmire of like political craziness? So then I started to learn about Jordan's history. And then I started to worry like ChatGPT is made by an American company. Americans definitely have a point of view about what's happening in the Middle East. So I was like, okay, ChatGPT, now tell me what somebody that lives in Saudi Arabia would think about this or somebody who lives in Egypt would think about this. And we just started to explore all these different perspectives. I started to explore citations. I literally would not feel comfortable doing that with a human. One, I would be worried that I was going to say something that like just put my foot in my mouth. I do that a lot. I don't mean to. I just am curious. And then two, like it was endless. I could ask a million questions and I wasn't offending anybody. I wasn't tiring anybody. And I did this for like four hours. I'm not joking. I did it for an entire evening. I tell people right now, like forego some of the nukes. It's so hard to find balanced news. I'm like spend an hour every day having a conversation with one of these systems because it's the best. Yeah. It's the best learning you can possibly get because it can micro-target the thing that you don't understand. And it's just such a great teacher, right? So cool. Great conversationalist. So that was kind of my beginning. And that was literally in the web browser. And then I knew that this is going to affect product work. I have to learn how to build with this. I have to learn. As I started to dive in more, I started to frame it as there's two paths. There's how do I build a product that is powered by AI? So like the LLM functionality is part of the service you're delivering to your customers. And then there's a second path of like, how do I as an individual augment the way that I work with AI? And so I started tinkering a little bit on both paths. So this past spring, I built my first AI driven product. I have a customer interview coach. So I teach a class on how to conduct effective customer interviews. Students do practice interviews interviewing each other. They submit their transcript and they now get really personalized, detailed feedback from my AI expert interview coach. Awesome. And this is something that like as a teacher, I wanted it to be really good. So I had to dive in like deep on how do we get reliable quality from a non-deterministic kind of messy system. So that was a super fun path. And I'm still like, I now have grand ambitions of creating basically a very broad discovery coach that can work with a product team all day long. Amazing. Is that all to be found on Product Talk? Yeah. The Interview Coach is a real product that does exist. It's part of our story-based customer interviews course, and that is on Product Talk. And then that broader discovery agent is something that's in development right now. And that'll take a while. To support that, there's going to be a whole bunch of little AI products that I'll release, and then eventually I'll get rolled into a centralized agent. And then on the second path... I was not doing a lot in this area until maybe this past June. Maybe like April, May, but then really in June. So April and May, I started just forcing myself. Anytime I did a task, I forced myself to ask, can ChatGPT help? Yeah. Like, what am I doing that ChatGPT could help with? And then eventually I got into Cloud Code, and I've gone deep on Cloud Code. And I'm now writing like a really... in-depth series on how people can get value out of cloud code. And I do code. I do write code. I can read code comfortably. Like I do have a maybe novice to intermediate. I should be nicer to myself. I have probably like an intermediate engineering background. I'm not an expert engineer by any means. But it's been enough for me to like start coding. It started with, I started using cloud code to help me with building my interview coach. And so I started using it in this coding paradigm. And for people that don't code, coding, a lot of engineers will pair program where there's like two people staring at the same screen and they're discussing how they want to work together. And that sounds really slow and wasteful, but really when you get that collaboration and discussion, they actually build more faster and there's less dead space. If you think about your work, you lose momentum a lot. You stare at the screen, what should I do next? But the two people working together, they have to keep momentum. It actually can be exhausting. So teams actually limit how much they pair program a day because it's exhausting. This is the best analogy I have for what it's like to work with AI as a thought partner. So this is now how I work with Cloud Code. I'm literally in Cloud Code from the minute I sit down at my computer until the minute I walk away. Every single thing that I do, I'm pair programming, even if I'm not programming, with Cloud Code. And I do this for like planning my week. writing a blog post, definitely for coding, thinking about marketing strategy, like literally analyzing interviews. I do it for everything. And it's part of the reason why I'm starting to write about it is I do think this is the future of work. I do think we are all going to pair work with an LLM and there's a whole bunch of skills required to learn how to do this well. And so that's, I'm just exploring that in depth. I talk about, I know cyborg is the wrong word, because when you talk about a cyborg, you think of metal slaved into your skin like Terminator. Yeah. But there is this augmentation via these systems that I think it absolutely is the future of work. I think it's why, at least coming out of college, they're like, oh, we want AI natives, which is the shorthand for people that can do this. I actually find it amusing that the people that are furthest along in this, I think, are not the young folks. I actually think folks... like you who are going really, really deep and have the broader expertise of what kind of systems are needed are going to be the most effective at making this melding. I'm curious, in all of that, was there an outlier success where you're like, oh my God, I just went from two days to two seconds. Were you surprised by any of the outcomes? I think writing has been the biggest lift for me. Pre-LLMs, most of my blog posts were 2,000 to 3,000 words. I would spend two to five days writing a blog post. Okay. Like, I would spend a lot of time on my blog posts. Yeah, yeah. Now, most of my blog posts are 5,000 to 7,000 words, so they've gotten longer. But, I mean, I challenge any listener to go look at them. They're not fluff. Yeah. They're not longer because they're fluff. They're longer because they're deeper and they're better. And I write them in about a day and a half. And I'm curious, how do you do that? Do you just outline? Yeah, I'll tell you it's an exhausting day and a half. So one thing I still need to learn with that pair programming analogy, I said engineers often have to limit how much they pair program. And a lot of that is because there's no downtime. And this is what I'm finding with Claude. Claude will always say, let's do the next step. And so I don't lose momentum. And so writers will, this will probably resonate. Like when I used to write a blog post. I'm an outliner. So I would always like jot down my thoughts, try to create an outline. And once I had an outline, I worked like one section at a time to write it. Sometimes I'd have to redo that outline midway through whatever. I would like finish a section or like finish a crummy draft of the outline and be like, okay, that's pretty good. I'm going to go check my email. Right. And then I'd get distracted for an hour and then I'd come back. And I would do that for like four to five days. And I would slowly chip away at making this blog post. Well, now with Claude. Like I come up with an outline and I go, hey, Claude, here's my outline. What do you think? And it gives me feedback. And then we retry outline. And then I'll be like, hey, Claude, I actually have this new way to structure it. Can you just recreate this outline, but using this structure? And then it just does the work for me, right? Like it does these translations. So I end up trying like five, 10 different ways of structuring the same blog post before I even start writing. Interesting. And so. Like I'm a big fan. I'm a big advocate of comparing and contrasting. Like we make better decisions when we compare and contrast. But like I never did this as a writer because writing is hard. Like I'm not going to put in the work to come up with three more different structures. But I do this with Claude. Like, hey, I'm using this analogy. What are five other analogies I could try? Or like I broke the problem down this way. What if we broke it down this other way? What do you think? So I iterate way more at the outline level. And that's, I probably spend. half a day just on that. And then once the structure is, once I'm happy with the structure, I start writing. And here's the difference. I write, I still do all of my own writing. Claude is not writing for me. Okay. So I write a section. I have Claude review it. And I have like, if people are familiar with Claude code, you can have these rules, files, and context. So Claude has my writing style guide. It knows all the context of my business. So when it's reviewing, it knows my target audience. It actually has access to my whole blog archive. So it knows what the related articles are and what we've already covered. And so it'll read a section and then it gives me really detailed feedback. It tells me what's working well. It tells me what could be clear. It tells me if I make any technical mistakes. And then we just iterate on the section and then it fixes my typos, which is amazing. And then it just says, are you ready to do the next section? It's so dumb that that's all it takes. But I'm like, yeah, I am ready to do the next section. Let's keep going. Whereas if I was on my own, I'd be like, I wrote a section. Cool. Let's go check email and browse the web for a while and go get a snack. I mean, those are foundational differences in the creative process though, right? So I use this term divergence a ton. And that's also where I use a lot of AI. Yeah. It's like, I think this is like option A, but give me, yeah, give me seven options, which again, before that would have been either a whole team helping you do that. But yeah, amazing at divergence. And then you're basically using it as a real-time editor and co-contact creator. Yeah, like these things did not exist before. It's really, really powerful. And sometimes it finds gaps in my outline. So like I wrote a blog post that's coming out this coming Wednesday called How to Use Cloud Code Safely. That's not the exact headline, but something like that. And I'm basically writing a cloud code series for non-technical people. And the challenge with this is Claude runs on your local machine. There is potential for quite a bit of harm. Claude could download malicious software. Your computer could be compromised. Like if you're going to use this tool, you have to be aware of that. So I was like, how do I write an article to like help people build awareness, but not scare them? Because like, just like we all learned how to uncover phishing scams in our email, I think we can all learn how to use LLM safely on our local machine. And so I started to think about like, where's their risk? And I started framing it as like, well, what can Claude do? Claude can read files, it can write files, it can write code, it can execute code, it can download things from the internet. And so I just started jotting down like these tiers of risk. And then I said, Claude, what do you think? And Claude was like, this is great, but I'd split this tier into three tiers, separate downloading packages from running code that I write, from installing plugins and skills. And I was like, oh, you're right. Those are three different tiers. Okay, cool. Right? And it's just, it's still me. It's still my content, but I'm being augmented by Claude, who's just helping me expand my thinking and fill in the details. Yeah. Amazing. Amazing. I'm going to, well, I need to ramp up because I haven't gone super deep in Claude, but, uh. I've absolutely been inspired to do it based on our conversations. I know a blog series that can help you. Yeah, that's right. We'll be sure to post that in the details and make sure that we amplify. What else? Two last questions because I know we're running close on time, which is kind of sad. One is just any personal tips, anything? I mean, we've actually shared a bunch, but anything for the audience where you're like, if you're struggling with just all of this complexity. I would go do X. Yeah. You've already rattled up a bunch. I've got a couple, though. I think the first one is don't try to keep up with everything. People ask me all the time, how does Cloud Code work different from Cursor? I actually have never tried Cursor. Not once. I don't feel the need to. I try new tools when I have a gap that I'm trying to fill. I'm not, I'm busy. I run a solo business. There's plenty for me to do day to day. I don't want to. And you know, some people argue if you're really a product person, you're trying every product on the planet. Baloney. You can learn product taste. You can learn what makes a good product just by using the products in your daily life. Yeah, I love that. So I think it's impossible to keep up with all that's happening. I don't even try. I just look at like, how do I? expand my comfort zone with this technology? How do I make sure that I'm always learning? How do I make sure that like I'm investing in what the future is going to look like? And then the other thing I'm doing, which I think also helps a ton is I'm sharing what I'm doing and I'm actively seeking out other people who are sharing what they're doing. Because this is the most fun part to me and it comes back to where we started, which is technology needs more humanity. So how do we explore this together and how do we explore it in community and how do we like help each other navigate this space so i share a ton on linkedin and then because i share a ton on linkedin other people have started like sharing with me how they're sharing on linkedin so that i'm finding kind of the right people to follow i'm in a few slack communities and slack channels around this stuff but i don't i'm trying really hard not to have a fire hose of data i just kind of pick and choose i love that guidance because i am guilty for sure of I think I was like, I should know what these five brands do, but you're absolutely right. I would be way better off taking that whatever half hour and going deeper in some other new experience that wasn't like, you know, 80% the same, but like, stay radically different. I I'm going to take that one to heart. That'll actually. So like, I'm really deep in AI. I built AI products. I'm now using it all day, every day. I've never used perplexity. I've never used cursor. Yeah. Right? Like there's a lot of these, I've never used Llama. I've never used, actually I've never even used Gemini. Right? Like this is kind of crazy to me, but here's the deal. Like I firmly believe we learn by getting our hands dirty. We don't learn by like spending 15 minutes playing with a product. We don't learn by reading 40 articles. We learn by like picking something and going deep with it. And I bet I could pick up Gemini command line interface tool in no problem. because it's really similar to Claude code. That's right. Right. And like I use Claude in VS code, which by the way, is directly analogous to cursor. So I probably could install cursor and be up to speed. No problem. Right. So like I try to look at like, what are these foundational things? In fact, I did this with discovery, like the opportunity solution tree is just what's the underlying structure? Like, how do we get people talk about first principles thinking like, what's the root of this thing that we have to learn? You know? I love it. I love it. That's the best piece of advice I've heard in a long time. I'm going to take that one to heart. Go deep. You know, the depth at this stage is more important than breadth. I will share one AI tool that I had heard about for a long, long time. But now that I have a podcast, I use it and I love it, is Descript. Are you familiar with Descript? Yeah, yeah. What I love about it, I know you have a producer, so you may not need it. But what I love about it is you can edit a video by editing the text transcript. Yeah. So like in my podcast, I can just delete all the opening small talk text and delete it. And the video is edited. It removes filler words, which is if you're doing video, the video gets jumpy and weird. So if you have a producer, use a producer. But my favorite thing about it is it identifies short clips. And I've used a lot of products. I've tried a lot of products that will identify short clips. And most of them are garbage. Descript actually does a pretty good job. Actually, we haven't used it for our own podcast, but we'll definitely check it out. Yeah. Any non-tech inspiration? Anything that you've run into? You know what's funny is I am a huge reader. Like I'm usually reading a book a week. And lately I have not been reading. And at first it was alarming to me. I was like, this is like the first time in my life I'm not reading. But I think the way we learn is changing. At least the way I learn is changing. And I'm just giving myself permission to embrace that. And I don't, I think when we talked personally, I shared this, like I had a, I had a pretty traumatic injury in the spring. I broke my ankle in hockey game. I had to have surgery. It's November and I'm still not 100%. Like it's been a really long road. And normally I have a very balanced life. I work some, I go outside and play a lot. I have a dog. I like, it's, I have feel like I have a very full diverse life. Whereas since this injury, I have been so AI focused. I don't do a lot of other stuff. I'm just starting to like get back to activity because of this stupid injury. But I don't think this is a bad thing. I actually feel like this was just a phase that like allowed me to put my mind somewhere else and not worry about what I couldn't do in the moment. But like, if you ask like, what has been inspiration that's not related to tech? I don't, I don't have a good answer right now. Cause I've been like deep in the rabbit hole. But I'm starting to come out. It's time to bring back some of that balance. It's also going to start snowing here and it's going to be ski season. So that'll help for sure. Beautiful. Do you ski? I cross-country ski. Cross-country, wonderful. Well, hopefully you can be back on the slopes soon and your ankles better. Awesome. Well, Teresa, thank you so much. This has been just such a treat continuing our chat, but having a new chat for some of the folks here. Thank you so much for all the, we got some great words of wisdom and just some great insights. Ah, thanks for having me. It was a lot of fun. Awesome. Cheers. In addition to running this podcast, I'm also the CEO of Immovid. In business today, it's impossible for anyone to tell what digital communication is real versus AI generated or modified. This is a huge issue. that we have actually solved at Imovid. Imovid is the only guaranteed authentic asynchronous communication platform that has been built for business. If your business relies upon relationship and trust building, this is a must-have platform for you and your teams. Please check out imovid.com at E-M-O-V-I-D dot com, or even better, if you can see it, scan the QR code and you'll get to see the product in action and even get a special podcast offer. Thank you again for listening.

What happens when one of the sharpest minds in product discovery meets the age of AI?

In this episode, Victor sits down with Teresa Torres—author of Continuous Discovery Habits and founder of Product Talk—to explore:
🧠 Why synthetic users won’t replace real customer conversations 
🤖 How AI agents are already changing how we build products 
⚙️ The power (and danger) of becoming too efficient 
...and why going deep, not broad, is the real unlock in today’s tech landscape.

For more information on the podcast, please click here: https://www.emovidcorp.com/the-victor-cho-podcast
To start a verified Human Conversation with Victor please click here: https://link.emovid.com/0wjoz4