Andrew Chen on Growth in the AI Era: Why the MVP Is Dead Again

The Works Feature: When Product-Market Fit Is Just Working at All

Andrew Chen has seen this movie before. At Andreessen Horowitz, where he co-leads the games fund, he's watching AI products attract millions of active users with what he calls "zero growth strategy." They can name themselves random acronyms. They send no notifications. They have no sophisticated onboarding flows. And they're exploding anyway.

The reason cuts to the heart of what actually drives growth in new technology categories. "It's like early internet where just by the product working at all, it has the works feature," Chen explains. "That in itself is so amazing that you can name your product anything, including just a random set of acronyms. You can have zero growth strategy, you can have products that send zero notifications and it'll attract millions of active users."

It's funny because it's like early internet where just by the product like working at all, it's—quote unquote has the works feature. You know, it's like that in itself is so amazing that you can name your product anything, including just a random set of acronyms. You can have zero growth strategy, you can have products that send zero notifications and it'll attract millions of active users.

— Andrew Chen

This is the beginning of the S-curve, where design differentiation doesn't matter yet. Chen points to quantitative analysis of the App Store showing massive volatility in the early years—flashlight apps, fart apps, Flappy Bird—all turning over constantly in the top 100. People were installing five new apps a week, hungry for new functionality. The MVP strategy thrived because the fact that something worked at all was inherently impressive.

But Chen sees the same maturation cycle coming for AI. The iPhone 1 to 3 jump was massive. The iPhone 11 to 13? Barely noticeable. Stable Diffusion and Mistral are already chasing their closed-source competitors. The novelty may be coming off faster than founders realize.

The 2021 Cohort's Reckoning: Why M&A Will Explode

Chen is predicting "a lot more" M&A in 2024, and his reasoning reveals what's actually happening beneath the surface of venture portfolios. Companies that raised on fancy 2021 valuations have spent the last two years executing the same playbook: RIFs of 25-30%, slashing user acquisition spend, raising flat rounds if they were lucky. All of it designed to extend runway from 18 months to three years.

That runway is now approaching its end. Pilot's data shows 60% of startups will need to raise before the end of 2024. But the growth rates aren't there. "A couple years ago, maybe people wanted to see 3 to 5x annual growth, but also were willing to accept a big burn, like a 7-figure monthly burn rate," Chen notes. "We now cut everything down across the industry generally on burn rate to be as low as possible. But what that means is often the growth rates that I'm seeing are more like 2 to 3x annually, and there's quite a few companies that are 2x or lower."

Those companies can't raise. They'd like to argue they could grow faster with more burn, but they don't have the data points to support it. The result: wind downs, soft landings, and opportunities for larger incumbents to acquire high-end talent that left during the boom years.

I think the market is still slow enough that I would expect quite a lot of wind downs, quite a lot of sort of soft landings. And then on the flip side, I think this is maybe a time for some of the larger incumbents to be able to get back some of the high-end talent that kind of left all the companies kind of in the last couple of years.

— Andrew Chen

Chen doesn't expect large, transformative M&A—the regulatory environment makes that difficult for mega companies. But small M&A and category consolidation? That's where the volume will come from. There's fragmentation everywhere, fueled by the indiscriminate funding of 2020-2021. Now comes the cleanup.

The Second-Generation Question: Who Actually Wins AI

High growth and high churn are defining the first generation of AI products. Chen sees the novelty effect clearly in diffusion models for image generation—the step changes are getting smaller, the improvements more incremental. The question is whether these early movers will adapt or get replaced.

History offers contradictory lessons. In mobile, first-generation companies like Flipboard and Foursquare were overtaken by Instagram and Uber three to four years later. But those first movers also got all the VC money, giving them resources to figure out the details. "If that top of funnel traffic falls off, then one argument would be the folks that are high churn that haven't figured out the product, there's going to be a moment there where a new kind of second generation might emerge," Chen says. "The other argument would be, well, that first generation got all the money."

The competitive dynamics may shift entirely. As models commoditize—Mistral chasing OpenAI, Stable Diffusion chasing Midjourney—the differentiator might not be answer quality but model size and local execution. Being able to run on a phone or computer without cloud latency could rewrite the game.

Maybe, for example, for LLMs, it's going to be way more important not just to give a pretty good answer, but for the model to be really small and to be able to run locally on your phone or on your computer. And so those are, might be very different competitive dynamics.

— Andrew Chen

Chen also sees opportunity in what's not yet happening: full workflow adoption. Enterprise customers are experimenting, using AI for concepting stages in 3D asset generation or isolated tasks, but they haven't moved their entire workflow over. The companies that build tools to enable that transition—not just models, but actual products—may be the real second-generation winners.

The Pre-Google Ads Moment: How AI Growth Channels Will Monetize

Chen draws a parallel to a specific moment in internet history that reveals where AI growth is headed. "We're kind of in the pre-Google ads on top of the search engine results period," he says. "There used to be a period where you could type in whatever you wanted to do a search on that, whether that was AltaVista or Hotbot or whatever, and there would be no ads. The search engines were just trying to do the very, very best to give you the results."

AI responses haven't been commercialized yet. They're optimizing purely for quality. But that won't last. The question is how they'll be commercialized, and what that means for distribution and growth strategy. Will it be sponsored responses? Affiliate-style recommendations? Paid placement in training data?

This ties directly to consumption patterns shifting toward short video. TikTok, Reels, even podcasts are now video-first, then transcribed to audio and text. AI models generating visual and creative content fit perfectly into these feeds. "All of these AI models actually are very, very tied in because the output is so visual and so creative. They just make for great compelling content in streams," Chen observes. "And so I think that feedback loop actually is driving, I think, a lot of the pickup for some of these."

The breakthrough won't necessarily be another GPT-3 to GPT-4 leap in general models. Chen expects incremental gains there. Instead, he's watching for breadth across media types: video generation moving from two-to-three-second idle animations to compelling five-minute YouTube Shorts, music, 3D assets, and the full multimodal stack required to generate something like GTA 6 from a text prompt. That game cost $1-2 billion to produce. The ability to generate anything close would be transformative.

The IPO Window and What Comes After

Only three meaningful tech IPOs happened in 2023: Instacart, Klaviyo, and Arm (though Arm was really a spinoff). Chen expects more in 2024, largely because it would be hard to have fewer. But the more interesting dynamic is what he hopes happens: a return to IPOs as growth vehicles, not exit events.

Companies are sitting on 2021 valuations with preference overhangs. For many, going public might actually be the best way to raise money, since preferences don't typically survive IPOs. If one or two adventurous companies go out and do well, it could trigger a wave.

The public markets are showing more stability and positivity. Late-stage companies have taken their medicine, cut burn, and are in better shape. Square and others went through a similar period of concern a decade ago before becoming great public companies. The question is whether the market will open for companies at $100 million in ARR with room to 10x, rather than requiring the business to be fully realized pre-IPO.

I'd like to see a world where the markets open back up where you can go out at $100 million in ARR and still 10x from there. I think it builds healthier organizations too. Instead of chasing exits, you're chasing compounding growth over time.

— Andrew Chen

Chen sees this as structurally healthier: companies focused on compounding growth rather than manufacturing exit events. Whether 2024 delivers that or just clears the backlog of zombie unicorns will determine if the venture cycle resets or just limps forward.