20Product: Is the Design Phase Dead in a World of AI | Has Claude Code Crushed Anthropic Already | What Roles of a PM Are Less and More Important with AI | How the Best Product Leaders Tell Stories with Noam Lovinsky, CPO @ Superhuman

0:28This is 20 Product with me, Harry Stebbings. Check this out. He was responsible for all of YouTube's applications. But before we dive into the show today, you know what's wild? It's 2026 and so many product teams are still flying blind, buried in spreadsheets, chasing feedback across 10 different tools, trying to figure out what actually will move the needle. I've spoken with hundreds of product leaders and the best teams all do one thing differently. They build a system to capture ideas, validate them with real data, and focus their roadmap on the right things. That's why product teams at Canva, Deliveroo, Toast, Decathlon use Jira Product Discovery.

1:04It pulls ideas and feedback into one place with built-in tools to prioritize what will actually have the biggest impact. That's when a roadmap stops being an endless list of ideas and becomes a plan people actually believe in. Join more than 20 ,000 teams already using Jira Product Discovery. Head to Atlassian.com forward slash Harry. Oh, I like it. I get my name in there. Atlassian.com forward slash Harry. and start building the right thing today. After Atlassian helps your team build and ship great products, Intercom helps you support the customers using them. If you're looking for a way to transform your customer service, let me introduce you to Finn, baby.

1:42Finn is the number one AI agent for customer service, resolving up to 93 % of customer queries automatically. There is no other agent that can do that. Not 93 % of customer queries, okay? No other agent can do that. So why choose Finn? Finn is the best performing AI agent for CS. Finn doesn't just answer questions. It takes actions. It automates the most complex customer queries like refunds, transaction disputes, technical troubleshooting with speed and reliability. I wish my team was speedy and reliable. Beats every competitor in every head-to-head bake-off. Completely configurable and code optional setup.

2:21My word. I mean, the benefits just go on and on. It's easy and efficient implementation. It works on any help desk with no tedious migration needs. It's trusted by over 6 ,000 customer service leaders, including top AI companies like Anthropic, Lovable, Synthesia, Clay, Vanta. So if you're ready to transform your customer service team, scale your support, and give team members time to focus on the really high-level strategic work, learn more about FIN at fin.ai forward slash 20VC. While FIN scales your support without losing speed, Reforge shows you how to translate that scale into durable product-led growth.

2:59Everyone's shipping faster than ever. Cursor, Claw Code, Codex. AI is making code and writing code faster than ever. But here's the problem. Speed means nothing if nobody uses what you ship. That's where Reforge comes in. Reforge is building the product discovery engine that sits upstream of your coding agents. Not another prototyping tool, research repo, or AI interviewer, but a product that will, number one, ingest your customer data, number two, generate variations of product solutions, number three, validate the solutions before code is written, and number four, hand off winning directions to your team.

3:41Reforge kills product debt before it starts, because every unused feature you ship isn't just wasted engineering time, it's a maintenance burden. complexity tax, and surface area that you cannot shrink. Used by product teams at companies like Toast, Vimeo, Klaviyo, and many more, Reforge helps teams ship more features that actually get used. Try Reforge at reforge.com forward slash build and use the code 20VC, that's 20VC, for one month free of pro. You have now arrived at your destination. nom i'm so excited for this dude i've heard so many good things i was literally just making you incredibly uncomfortable beforehand normally with venture investors you say that you've got great references and they're like oh stop it tell me more but you're like legit like no it's making me uncomfortable which is it shows your humility but thank you so much for joining me today dude oh it's my pleasure thank you for having me not at all but i want to start with a little bit of a story because I spoke to Glenn at Redfin before and he said you came into Redfin when they'd had a couple of different product leaders and you answered a question that he always remembers.

4:53The question that you answered is, what is a product leader? And he said the description that you gave was very simple, but it was phenomenal. And so when I ask you, what is a product leader and a great product leader? What is that description? I mean, I think fundamentally a great product leader is a great storyteller, someone that is able to understand what the customers actually need, what problem that actually needs to be solved and can form that into a story that is just well understood and well aligned, not only with the market and the customer, but that gets everyone internally to row in the same direction.

5:36there's all these interesting debates right now about like roles collapsing and product and marketing and so on and i just really don't see these things as separate i see product and marketing and you know it's the same thing and so maybe that's where where this comes from i think that fundamentally a good product leader is just an excellent storyteller and the best companies and the best brands are excellent storytellers this is where we don't send schedules in advance because I just kind of lose interest in them the minute you say anything. Our teams are like, we spend hours doing this research, Harry, and then you just go rogue.

6:12But you said storyteller. I get that. But the challenge when you're a horizontal product is different customers resonate with different stories. How do you think about being a great storyteller when you have such a broad customer base with a horizontal product? That's an excellent question. Not to return some of the flattery to you really shows like kind of a depth of understanding that you have that I don't think is always common amongst this crowd. So I appreciate that. I think that you, you go to different things, right? You go to what is the feeling that you're trying to create? What is the, you know, after your, your features are delivered after you solve the kind of the, each of the micro problems, what is the overall feeling that you're leaving the customer with?

6:57How do they feel supported? How do they feel kind of more in the flow? Whatever you're going, whatever you're going after. I think a good a good example of that is like Instagram, like was Instagram about like the features of the specific problems or was it more about catering to the feeling of the need for vanity? And that's essentially was the kind of the I think the product insight is that, you know, vanity is a much bigger market than than we realized it was, you know, latent demand that kind of tapped into. that's the feeling that you're trying to address. What is a bad story that product leaders often tell, do you think?

7:35What do you see? I mean, maybe this is like close to my heart right now, but I'm pretty tired of the like, this is going to make you more productive story or this is going to make you like faster story. I think we like lean to things like time and productivity when we don't know what the value is, what the true value is. And so we're like, well, just let those, that's a, that's a catch all. It will save you time. People want to save time. Let's see if like, you know, time, time spent will be the, the, the thing that resonates, you know, rather than kind of going a little bit deeper and understanding like what problem are you actually tapping into beyond the feeling of, you know, wanting to be quick, wanting to be in the flow.

8:18And that's an anxiety problem. A lot of people suggest that we're going to lose the design phase in a world where vibe coding and prototyping is so much quicker and real. Do you agree with that? No. You know, it's interesting. So many of these things that we're like talking about as if they're new things. I don't think they're very new. I think what's changing is that the tool set is accelerating, you know, things that have been happening for a long while. Like, I'm sure in all of your investments, like the best teams are one where like talent is collapsed in a smaller number of people. People wear many different hats.

8:53It's the idea that like you are a designer and therefore you stay in these lanes and you do these tasks and you don't do engineering work or you don't do product work. It just like doesn't make sense. Like it doesn't work that way. And I get that as we scale, there's the idea that like everyone needs to specialize and kind of get deeper in their lane. I just I just don't think that that's ever been true. And now the tools are making it even less true because to specialize and do well in these in these areas to to be a unicorn that can do many things. You're not required to learn the syntax or the tooling or the tactics of doing any one of those jobs in the way that you have before.

9:31So is the design phase going away? No, like you still have to do that sort of thinking. And there are many different tools for doing that sort of thinking. And sometimes the designer today, I had the CPO of Duolingo on the show. He was fantastic. And they said about actually chess and the integration or introduction of that from two designers. And they vibe coded it in a couple of days and then brought to life a working kind of prototype that everyone could play with. Is there any excuse for a designer to bring a design anymore when you could vibe code your idea into reality as efficiently in the same amount of time?

10:11So, I mean, that's that's interesting. as efficiently in the same amount of time. I don't know that where you are in kind of the design thinking stage necessarily leads to building a prototype is as efficient. I think if what you're asking is, if you want people to empathize and understand your idea and feel your idea in the highest signal way, should you give them the highest fidelity approximation of that idea that you can as quickly as possible, that I would say, yes, that is necessary. As a designer, you know, you should be able to produce that prototype, produce that working product, if you have the time and the means when you're trying to get other people to empathize with your idea.

10:55But that doesn't mean that when you're starting your thinking process, that maybe a whiteboard is actually like the best place to start or a blank sheet of paper or a Figma canvas that works in the dumb old way that they used to work. I just think that the design thinking can still benefit from those different mediums, those different ways of doing the thinking because of the constraints that they apply, because of the space that they create. I think that's different than saying, hey, if you're at the point where you wanna like share your idea and get people to empathize with your idea as deeply as they possibly can, what's the best way to do it?

11:33Yeah, like have them use the thing and feel it in the highest fidelity approximation that you think it should be. And so that might be a working product or a working prototype. But I don't think that means that all design thinking should start in a wide coding platform. What tools do you see used most often within the product teams today? Is it Cursor? Is it Cloud Code? Is it Cognition? What are you seeing internally that's interesting or surprising? In terms of the AI coding, I mean, definitely Cloud Code is the far and away. And I think like just a lot of people make the transition from, I want to start with something that is more as more familiar UX like cursor and then move to the terminal.

12:15And there's, there's a lot of freedom in moving to the terminal and almost like, you know, having layers of the abstraction, even less visible, not worrying about what files it's creating, not worrying about inspecting the files, like kind of gaining that confidence. I think that's, that's what I'm seeing most. And then, you know, the next is starting with the kind of the more familiar applications for prototyping like Lovable or Figma make. But the transition that I typically see is that you start with kind of like that Lovable. It's more familiar. It feels less scary. You know, it's like it doesn't feel as much like coding.

12:51And you move to something like a cursor and eventually in the terminal. What do we do in product development today that we won't do in three years time? Write specs for humans. I hope most people aren't writing specs for humans any longer. I think writing specs for agents is really helpful and smart. And I think that also takes a different form. And the way that we write them, I think, then also becomes different. But yeah. How does the world change when you're writing specs for agents, not humans? And what needs to be altered? Some things are still helpful to be similar. Like, you know, who am I building for?

13:27What problem am I trying to solve? Kind of like what, like those fundamental things. But when you're writing for a thing that's actually going to do the job, I think you just you end up like creating different types of details and kind of also even just structuring what you're writing differently. It's good to use examples, build up like a context library of things that have worked in the past that you want to emulate, things that haven't worked in the past that you want to avoid. all of these things that when you're writing to a human, you just assume that they have that tacit knowledge, that, you know, since you've been working on these things together, they like they understand that, that you don't feel like you need to embed as much of like the context engineering effectively into the spec, that when you're going to have something that is actually bootstrapping and coding, you know, kind of based on that on spec and that in that plan file in a very direct and fundamental way, I think it changes what what you put in there.

14:23I also think it should change how you write it. You should use the thing that's going to code it to help you write it because it's going to end up creating a version of that output that is more compatible with what it understands to do the work. I have so many questions. I'm so enjoying this. If we're writing specs for agents, okay, are we going to see like a normalization of products, like a kind of plateauing of creativity? because the wonderful thing about writing specs for a human is that a human brings in the experience they had from growing up in a kibbutz in Israel where they think about whatever in a really different and cool way that influences how they think about collaboration features and they bring that really cool anomalous experience to the product process and you lose that hallucinatory element when writing a spec for an agent that executes it.

15:14Do we see that? I would say we will see a little bit of both. I actually think that this will lead to creativity and certainly like taste standing out more than ever, because I think you're going to have this like flattening where you are going to have a lot of things that just like, oh, you know, here we go. it's this it's the same dialogue it's the same flow like this this is what works these things have learned that this is what works and this this is what they're putting out but that actually i think gives more room for the standouts to shine for the things that kind of have that taste and creativity to shine and and fundamentally that's still like the human the human job is to figure that out so i get the maybe the analogy that to the describe that flattening just to go back to Instagram and sort of like what Instagram did to did to photos, right?

16:05We went through this period of like, every photo looks amazing. It's filtered, it's it's touched up, like, look at my amazing, shiny life, like it was sort of like this creative flattening of what like makes a good photo. And then, you know, slowly, organically, over time, what emerged is that the things that that people, you know, thought were more interesting, or more compelling was the things that like, looked a little bit messy or like a little bit organic or kind of like the trend of what was like in fashion changed after that kind of flattening period happened. I think we'll see something similar with, you know, applications.

16:43Can I ask, the joys of doing what I do is I get to ask really smart people for their wisdom to help me in my other job, which is also far more lucrative, I have to admit, than media, which is obviously using other people's money. But great lesson, kids, OPM, other people's money. You see, there's this brilliant wisdom that we give on this show. Vibe coding, I'm just stuck with it as a market. Do you think it will be an enduring market to offer non-technical functions, the ability to spin up development sites, you name it, faster? Will your lovables, rat blitz, base 44s be in every sales and marketing team, or is it a moment in time hype cycle?

17:21I think the idea that everyone can build is not a moment in time hype cycle. Like we've seen it time and time again, like you can democratize, you know, the act of creation. Many more people want and can create than we believe or than we currently observe and then that kind of expands. So I don't think that that is going anywhere. where the like value capture will be in in that stack i think is a different question you know is it in the kind of the deployment hosting and distribution of the thing is it in and will you be paying for the tool i think that that is a different question but i i think fundamentally will happen is that these vibe coding tools will like continue to march up the stack and they're not building an IDE.

18:13They're building a service, a thinking service that does things for you. My friend, which one's easier to do? Is it easier for ClawCode and Cursor to go down the stack and eat the consumer end? Or is it easier for the consumer end, Lovables, Base44s, Replets, to go up the intellectual stack and eat the developer end? I think that the hardest thing right now is figuring out the user experience that is going to scale to the largest number and average average users i think that we're already very much at the point where we have this kind of capability overhang that people talk about right where like what these models can do and what people actually can do do with them there's there's a big gap there so i think that that the question of, is it easier for cloud code to figure out the right user experience?

19:09Or is it easier for someone that is working more actively in the kind of that UX application layer, like a, like a, like a manis or, or, or like a, like a lovable to figure out the user experience unlock? I guess I would, I would probably bet on people working at that UX application layer already. But I think that the foundational labs are doing that as well and are trying to learn at that level as well. How much would you say of net new code created today within superhuman entities, all the different products you have, is written by AI versus by engineers? I think we're basically at the point where we're approaching about half that's there.

19:52I think that it can go much further than that. What do you think it is in 24 months? In 24 months, I hope it's like 90%. When it's 90%, what do we do? Do you have less engineers? Do you just create way more products? How does that change when 40 % more is taken? I think it's absolutely the latter, that we build more things. I just, this idea, I don't, I have never worked at a single company that doesn't have an infinite roadmap. That doesn't have like a, we're done here. Like we only need this many people or done or done here. I absolutely think we're going to go through a phase and we're going through a phase where the number of people we need and what a good ratio on a product team looks like.

20:40And all of that is going through like a fundamental shift. And that's going to that that causes obviously some disruption that that doesn't always feel great. But then I think once that kind of normalizes and we have a better understanding of that, our desire to do more is not going to go anywhere. And then we're going to like continue to scale, but basically, you know, divide up the work in a different way. Can I ask you, you said about the ratio on product teams changing. What do you think it's changing from and to? And how would you discuss that? So, I mean, I think that, you know, up until like a couple of years ago, you probably like, what is a, what is a kind of a decent product team look like if you were starting like a zero to one team?

21:20You know, maybe you say it's like, you know, depending on, on the problem, you know, five, 10 engineers at the limit, you know, it's kind of like 1 p.m., one designer, right? I think now you're, you're much more looking at, you know, it's like maybe 1 p.m., one designer, two engineers is kind of like, you know, what, what you need. And then also who is doing what is also very, very different, right? Like everyone is in the code. Everyone is building the thing. And you just have a small number of people that have their hands on a much wider part of the product pipeline. And I think that leads to better products.

21:57Totally get you. And so we have smaller teams, everyone still being in the code. When we get a little bit further along, how has testing and deployment changed in the world of AI? Or has this fundamentally remained the same? Well, I mean, I think that for one, like the AI can do a lot of the testing. Certainly the, you know, the first run, catching all the obvious things. I mean, even for on-call, like incidents, like when the kind of bad things happen, I think that AI can do a lot of the triage, a lot of the first run investigations so that by the time it gets to an on-call engineer, it's like, here's what I think is going on.

22:34Here's, I think, the three options are to fix this. like which one of these paths do I'm going to take go. And I think over time, as you build up that context and that memory, then it's going to ask you less and less. So I think the whole stack is going to be automated in the same way. How much faster does AI make your teams? Are you able to ship twice as much, three times as much? I know it's hard to quantify, but just to help someone who doesn't live in product. I think that maybe for our teams, like the fundamental thing that can get shrunk is the exploration phase and the rate of iteration through the exploration phase how quickly you can kind of get to this is the thing we actually need to build and we've you know we've learned that we've we've tested that we've iterated through that exploration phase that's shrunk quite significantly i don't know how to like put like a it's 2x it's 3x on that but i i do think that in the limit that speeds you up quite dramatically What does it really mean the exploration phase has shrunk?

23:35I think that the phase of going through, we've observed this problem. How do we build a solution for this problem? Okay, let's try to iterate through what a solution might be. Let's go and test that with some of our customers. Oh, that's not the right one. Let's iterate and try this other one. How much of that you can shrink? How many of those you can do in parallel? who can do that kind of full, full pipeline of that exploration, how many different people you need to do that. And so basically, it just it increases the rate of learning, like quite dramatically, which then kind of, you know, shrinks the whole kind of product development lifecycle.

24:16And I think if the business is fundamentally learning much more quickly, then you know, the whole thing has like a massive acceleration. There's also obviously the I think when we think about this, we like quickly go to like, I used to type all these things with my hands. And now like I can, you know, this thing just types them for me. Yes, there's that's that sort of speed up as well, obviously. And that saves you like quite a bit of time. But then again, I feel like you, you just end up doing more also of other things. You're, you're reviewing code more, right? And then we need to like make that process more scalable.

24:49You're then just kind of doing more things and in parallel. And so anyways, I think that the shrinking of the exploration and the rate of learning is, at least right now for us, one of the biggest accelerators. When we think about how it makes more efficiency and more productivity gains within engineers, if an engineer, say, is paid 250 grand a year, I'm just making up numbers, and it makes them 30 % more efficient or better or whatever it does. So it makes sense to pay 75 grand per engineer per You know, it's interesting. I just don't think that that's how things work fundamentally. I think what happens is engineers just get to spend more time on other parts of the product development process.

25:33And so they get to flex their product-y skills more, or they get to flex their data analysis skills more. And so the most valuable people will continue to be the people that can wear many hats that, you know, have a more like kind of fulsome skill set that they can express, which was before was just harder to express because of one, where time had to go. But to also the time that you're tooling required of you to kind of be able to express those skills. No, I'm a VC. OK, we were very simple. We're coin operating. We think in dollars. and great British pounds in the UK. If we are going to make money from AI, we need to fundamentally see the transition from software spend to human labor spend in a way like I mentioned.

26:2230 % of time, great, we'll pay 30 % of salary. Because otherwise, we're still paying the per seat 25 bucks a month, and then the TAM doesn't expand. Can you help me understand? Are we going to see the expansion of TAM with the movement of that software spend to human labor spend? or are we all getting way too high on our own supply and we're going to stay in a software spend world? I mean, I guess like TAM expansion that I see is that the number of problems, the number of things that you're able to solve with software and do for people is just going to expand. And so, yes, will you be able to have smaller teams, like kind of more productive, doing more?

27:04Will that change kind of like the HopEx accounting fundamentals that are like, yes, but I think then like what happens is that you expand through just providing more service, solving more problems. And does that lead to like basically fewer, bigger firms? I don't know, perhaps like maybe there are too many providers and what you actually need is fewer providers that just do more and cover much more, much more of the stack. Do you have to be a platform today? When we look at a superhuman with all the entities that you have your coders your superhumans your grammaries and when we look at like a notion i know notion obviously a competitor so forgive me for bringing in a competitor but it's like you know they obviously have calendar integrated now i think they're starting to do cool recordings they obviously have their cool kind of knowledge management system i think when you look at like an otter or fireflies they're aware of the need to move from product to platform is product no longer enough in a world where you need to be platform it depends how you define platform like I think there's one definition of platform, which is basically other people can build businesses on top of your product.

28:14I think that maybe that definition is not required for everyone. I think that your customers are going to build on top of your product is a requirement for everyone that is serious because customers' needs are always nuanced. And I think we're very quickly moving to a world where you're just going to have a lot more bespoke software. And so you have to have a platform approach that and how your customers can build on and extend your platform, your products in order to kind of meet their need. What changes then for product leaders in the world when you exist in a world where customers need to build nuanced, personalized, customized features, elements to your product?

28:57I think you just think of things pretty differently and the impact of things pretty differently when you know you have other people developing on your platform, right? Like how you roll out change, how you think about backwards compatibility, how you even kind of run experiments and kind of measure your experiments in terms of, you know, who's doing what with your product and kind of what you're making better or what you're breaking. like one of my favorite examples of that and this is more from like youtube is a platform and the definition that other people can build their business on top of youtube and when we would kind of experiment and you know just observe things in terms of watch time it's like hey watch time is is up this is a this is a big win and then we'd roll it out and we'd have like the the community like you know enraged and why is that it's just like what it's not it's not equally distributed.