Your Questions Answered: Vibe Coding, Consumer Defensibility, and the Future of Agent Management
17:30 Share

♪

Hey, everyone, and welcome to Generative Now. I am Michael Magnano. I am a partner at Lightspeed. And for today's episode of the podcast, we're doing a bit of a mailbag episode where I'm taking your questions from X and LinkedIn. We put up a post a few days ago and a bunch of you sent in some really, really great questions. And so we're going to answer them now right on the pod. I have not read all of these, so I'm going to be a little bit surprised. Actually, our producer, Danielle, is going to read the questions and I'm just going to respond to them. So let's do it.

We've got a lot of really great questions here. So let's dive into the first one. So this first question from X Tanay is asking about the app layer. This is kind of a two-part question. So for the app layer, are we going to see more opportunities in new markets or existing markets? Yeah, probably a cop-out answer, but I would say both, right? I think in existing markets, we're going to see products get a lot more efficient, right? And so I think as a result,

existing businesses will get bigger, right? We'll see like bigger margins as a result. You know, we will see those businesses be able to invest back into their business and

and then they'll potentially grow as a result. I think new markets, I think anytime there's a new technology, we're always surprised at the upside, right? Things that maybe couldn't be done before now can be done as a result of technology and new opportunities emerge, right? New teams, new startups, new businesses emerge. And so maybe a cop-out answer, but I kind of feel like it's going to be both. If you had to choose one, which would it be? I'd say mostly like existing markets is probably...

probably where the bulk of the opportunity will be. But I think software always shows us that there are opportunities in new markets. We just can't predict what they are, like what they're going to be. And I think AI is the same way. I think like we're probably trying to put a little bit too much prediction on what the great products are going to be.

But they always surprise us to the upside. Yeah. Okay. And then second part of this question, are we going to see SaaS TAMs 10x with agents or are we going to see them decline because of commoditization, vibe coding? I'm an optimist on this. I kind of feel like we're going to see market expansion. I think, you know,

Humans have shown an insatiable desire for software. And I think, you know, the Internet has shown us not just with software, but with with other products and other, you know, mediums, let's call it on the Internet. The more supply we have for these things, typically like the humans fill the demand for it. And so I think vibe coding and tools that are internally built and all this, like I think it's just going to create demand.

more opportunity. We're going to see software fill holes that it couldn't previously fill. And I think we're going to see new products and services emerge that people are willing to pay for. And so I think the TAM, not just in SaaS, but also in consumer is going to get a lot, lot bigger when we have that much, that much more software that's being created as a result of AI.

Okay, so this next question is from LinkedIn. We have Amiya asking, for consumer AI startups building at the application layer, what strategies are proving most effective for creating genuine defensibility and raising switching costs beyond first mover advantage or better UX? The answer here is going to be less exciting than I think people think. I think this is a question a lot of people are asking. It's like, hey, what's defensible in the app layer for AI startups? And

The answer is kind of boring. The answer is it's like the same things that were always defensible.

for for companies even pre-ai and that's things like networked effects that's things like data modes um those are things like you know a really really strong brand and oftentimes these things do accrue to early movers or first movers that are able to get the networks and the network effects in place before anyone else or to get everyone using their product then spinning off some sort of data flywheel before anyone else can do it now i would say um

even though the tactics are the same, that doesn't mean that sort of the risk level is changed in AI. And I actually think that the risk level is changed in AI because we're

Partially because of the last question we just discussed, which I know we were talking more about SaaS, but we said that there's more there's increased pressure, increased competition because more and more people can create products now. So whereas previously maybe you were building in a space where you only had a couple of competitors, but now you're going to be operating in a space where you potentially have hundreds or thousands of competitors.

editors. And that will force you to move even faster and to make sure your emotes and your network effects and all this stuff are even stronger than previously. Look, you got to build great products. Like I think the bar is now higher. Uh,

It's higher than it has ever been before because anyone can build products now. We're seeing people crank up products on the weekend that are overnight generating like hundreds of thousands in MRR. So, yeah, more competition. This next one is from Sarang, and we're talking about the evolving roles at AI companies. So he points out that there's two evolving roles of agent engineer and agent manager. So Sarang.

Agent engineer is now a job at Sierra. They are individuals who work with customers to design, build and ship agents using Sierra's platform. Sounds like engineers who build agents. An agent manager, he links to a job post at Basis. So this would be managing intelligence embedded in software. So can you please discuss the finer aspects of these roles and how are they potentially evolving from your point of view?

I'm not an expert on these two blog posts, but, you know, quickly scanning through them. My take is that agent engineers are effectively a new type of software engineer where there's less attachment to specific tools and processes and setup. And there's more there are more things that the engineer needs to master, like in addition to just

programming in Python or programming in JavaScript or whatever, like the agent engineer needs to also be great at prompt engineering, needs to have like a strong grasp of vector database, vector databases probably also needs to be very, very adaptable and willing to learn the next sort of AI breakthrough that comes maybe only a couple of weeks from now. And so it seems like it's like more of like a adaptable engineer who is sort of less precious

about like purity of software engineering and sort of being comfortable handing over a lot of the reins to AI. So I think, look, I think that always makes sense. I think, you know, when you're an engineer, when you're in any field, you constantly have to be willing to adapt and learn the next thing. And so that's my take on what an agent engineer is. And then for agent managers, this is somebody that's sort of orchestrating the agents, right? So if an agent engineer is building agents, then there's a person that's sort of

managing all the agents and deploying them and making sure that they're actually executing on the tasks that the manager needs them to execute on. Now, I think this is actually really, really interesting and kind of different from what we've seen before, because in traditional management, we can talk to humans to better understand what they're doing. We can read the comments that they put in their code. We can read the documentation that they've created around their code. We can do retrospectives when projects go wrong.

we can read the meeting notes, right? Like these are all inputs that we can assess about from humans. But I think when you have hundreds or thousands or millions of agents out there

performing tasks on behalf of managers or organizations more broadly, it's going to get way, way harder to understand what is actually going on on the ground. And so I think this job, the sort of agent manager, is going to be extremely complicated. And I think that there are going to be lots of tools and perhaps even markets spun up

around this exact workflow? How do we know what the agents are doing? How do we track their work? How do we measure their efficiency? I wouldn't be surprised if we start seeing software and products and tools and services that measure this stuff for us so we don't even have to do it. And then they summarize the work for us and tell us exactly what's going on. And so to me, agent engineer sounds like, you know,

an engineer that's sort of staying at the cutting edge of what's being used to build great software. And obviously, AI is a big part of that. But agent manager, to me, sounds like a whole new workflow and sort of a whole new world of tools and orchestration to keep track of these non-human agents, which

may number in again in the thousands or maybe even millions one day. That's kind of how I think about it. Next one is from X from Akshay. So the best verticals for consumer AI. Personally, I see a lot of opportunity in the education space, but not a lot of people are talking about it yet.

I think people are just starting to wake up to the opportunity of consumer AI. I think we see a lot of opportunity in consumer services. So services that consumers traditionally take advantage of for all sorts of things could be, you know, could be your taxes, could be, you know, bookkeeping, could be legal services, all these things that regular consumers need. They can probably now start to get a lot more efficiency when AI is a big part of the workflow, right? And that means that it could be more cost affordable, which obviously will drive adoption, right?

But I think education, which Akshay points out, is another big one. Full disclosure, I am a recent co-founder of an AI education company called Oboe. And I think, you know, we and many others, it's not just Oboe, there are a number of companies out there who see the same opportunity. And the thinking is really, look, AI is this magical, magical co-pilot that knows everything, right? If it could also know everything about you and what you've learned and what you're

goals are, how you learn best. I mean, you could effectively have, everyone could have their own personal tutor from helping them go from point A to point B on a learning journey. And this could obviously apply to, you know, every level of education from sort of grade school to higher education, to professional development.

to just hobbyists who want to learn things just for fun. And so, yes, agree that education is a massive, massive opportunity. Another thing we think and talk a lot about, you know, there's this concept of vibe coding, which a lot of people are talking about right now.

And I think vibe coding is really interesting because it's going to bring a lot new creators into the ecosystem. And there's probably going to be not only a whole market built around those those creators catering to these creators, the tools, the marketplaces, the analytics, right?

But the things that come out of there are going to be really, really interesting. And you could almost imagine a whole new marketplace of ideas and products and services emerging from that that haven't existed before. It's possible we get, you know, a whole new type of app store similar to the iOS app store in the end, the Google Play app store we have on our phones, but for products, tools and services built by regular consumers. And then within that, within the sort of broad category of vibe coding, another thing that we're seeing people do, which is super fascinating, is games.

The barrier to create games is dropping dramatically. So we're seeing lots of people make these super casual games that they could just crank out very, very quickly. We totally believe that there's going to be a consumer market built around that. So those are just some of the things that at least we at Lightspeed are thinking about in the realm of consumer AI. But again, when there's new technology in sort of a category as big and as broad as consumer, I have no doubt we're going to see lots of surprises there.

and things pop up that we totally had not expected. What's your take on Ashkay saying that the AI education gold rush is here? Would you say that this is an AI education gold rush? I think whenever we see a number of companies and lots of smart people going after a space that

that is not yet fully developed, it's usually a sign that something interesting is happening. Our next question is from Counterplot. What acquisition would atone for Apple's lag in AI? To me, it feels like to really catch up, you'd have to you'd have to grab one of the large foundation model companies, you know, like an OpenAI or an Anthropic or an XAI. I

I have no idea what the regulatory hurdles would be on something like that. And obviously the price tag would be absolutely enormous. But, you know, it feels like if they want a model that is not only great, but it can actually do things on behalf of the consumer on the iPhone, it feels like it would be hard to get that without something that's so, so advanced and so far along.

And so that's why I mentioned something like that. Now, they could go in another direction, right, where maybe they try to get something that could

helps spin the flywheel of the app store faster. So maybe that's something like a replit where people could like very easily build apps for the app store. But I think that's different than say, you know, fixing all of their problems with Siri and introducing sort of the next generation AI to live on the iPhone and help people get stuff done. Super fascinating though. We obviously don't see Apple make

sort of gigantic acquisitions very often. But it does feel like

given their sort of lagging position in AI and that, you know, they're one of the most valuable companies in the world. Maybe this is the time where they break out the checkbook and do something totally unprecedented. So next question here, what's the potential of AI-assisted dispute resolution or arbitration? That's pretty interesting. I hadn't really thought about that. I think I'm not a lawyer or anything, but I have to mention that in dispute resolution and arbitration, a lot of emotions arise.

often get involved. I think AI has the ability to do this in sort of an emotionless way. Sort of the legal landscape is a place where AI, you know, we mentioned sort of AI-assisted services for consumers earlier. I think, you know, anything in the legal landscape could be a perfect opportunity for AI to come in

and do things in a far more efficient and cost-efficient way. I think about products like Polymarket, we're now starting to see sort of real world events

be sort of documented on the blockchain for everyone to be able to see publicly. And then, you know, those events and those sort of resolutions can be tied into other events. Maybe, maybe there's some tie in here where we can use sort of AI to settle disputes or help us settle disputes, maybe with the assistance of humans to help verify. And then sort of, we've, we verify these things, um, kind of publicly on a ledger, um,

Yeah, I don't know. I'm sort of thinking about this in real time, but I do think there's potential for AI to assist in this field for sure. So next question is from Daniel asking about the state of VC. What type of LPs are still deploying? What are the average check sizes and have capital calls changed? The only thing I'll say about this is like there's there's a tremendous amount of appetite for LPs.

for tech from investors right now, investors of all sizes. I think everyone sees the excitement that's happening with AI and everyone wants to get involved. I think, you know, LPs, GPs, founders, obviously, I think everyone is seeing this sort of a generational opportunity for a new technology to create, you know, new businesses, new technologies, new outcomes.

And I think, you know, everyone, everyone wants to be in the game. I think the question is, where do you bet? Do you bet, you know, in sort of big institutional funds like Lightspeed and others? Are you betting on emerging managers? Many of the big institutional LPs are placing bets up and down the stack, you know,

you know, to sort of, I think, maximize their chance of a return. But despite everything that's been going on in the markets over the past couple of weeks, you know, obviously very early to see how that's all going to play out. I think there is still plenty of demand and appetite for investors to want to bet on AI. And VC is the way to do that. Before we wrap, I'll just say, like, I've been really, really excited about consumer. I think

There's so much interesting stuff going on, frankly, as a result of things like vibe coding. I think, you know, it's been a few years now since we've had access to these these these models and these superpowers. And I think you're now starting the capabilities as a result of the software, but also the imaginations of the builders are now catching up to a point where you're starting to get starting to see people do some like really creative and inventive things.

And for me, that's always been like the most fun part of the cycle. When I think back to like the mobile era and the early days app store, it was like, wasn't the first year, wasn't even the first two years. It was like once people really started to wrap their heads around the camera and the GPS and all like the unique and interesting things you could build, like that's when the real creativity started to flow. And like, we're starting to see that now in the, in the consumer space application layer for sure. So yeah.

I'm really excited about these next few months. Will be interesting to do another mailbag episode in a few months from now and see how the questions respond to that. And if anyone else has any other questions, feel free to mention me on X at Magnano, drop them in the comments on YouTube or any of the Lightspeed channels and we can respond to you there.

Thank you so much for listening to Generative Now. If you liked what you heard, please rate and review the episode. It really does help. And if you want to learn more, follow Lightspeed Venture Partners on YouTube, X, or LinkedIn. Generative Now is produced by Lightspeed in partnership with Pod People. I am Michael Mignano, and we will be back next week. See you then.