Press Start to Continue - Gaming and AI

Note: by AI here I mean the 2020+ wave of generative AI. Games have been using AI for ages, and hypercasual mobile games have gotten it down to a science.

ChatGPT has been rightfully lauded for reaching 100M users in a couple of months - but Pokemon Go reached 100M installs in a similar timeframe back in 2016. Video games is a $180B+ global market. Almost double the movies and entertainment market, although I believe all media will converge into an interactive, self-assembling blob catered to you. In the meantime, what does gaming look like in the next 5-10 years?

Supply Side: games are faster & cheaper to build but upside from that alone is capped

This is the obvious low hanging fruit: make generating game assets and the process of game development cheaper and faster. This is the realm of the likes of Inworld AI (NPC personality & dialogue - recently pivoted its positioning to agentic frameworks, raised $125M); Latent Technologies (generative physics); and nunu.ai (game testing). Microsoft recently released peek videos of Muse, and Decart both generate open worlds in response to user movement. Extremely cool stuff! But not intrinsically a net new game mechanic (more on gameplay below).

If everything works, there’s still the question of live operations: creating and managing event-based activities such as discounts, seasonal content, quests etc. especially in mobile gaming. “Traditional” AI/ML has done a good job here, but generative models can take in a lot more unstructured data to find even subtler, hyperpersonalized signals and knobs to wring out more juice. Companies like SuperScale (raised $1.2M seed) and Pragma (Greylock backed, raised $12M) tackle this.

These are all largely deflationary: most likely it’s the game themselves, not their production infrastructure, who capture the bulk of the value created. Unreal Engine’s revenues was $275M in 2023 - around 6% of Epic Games’ total revenue. Part of the challenge in scaling independent engines is that the game developer market is an oligopoly, and those players are vertically integrated and use their own in-house tooling, leaving not much meat on the bone for large ticket $s. As for the fragmented long tail, Unity and Unreal account for 79% of games released on steam, and no other player has made a significant dent in changing that so far.

However, if you are able to incubate a new cohort of creators who didn’t exist before, you may be able to deepen the market by extending its long tail. Think of Canva making social media visuals easy to make for non-designers, and Wix for non-frontend engineers. Rosebud.ai, Jabali, and Breshna are some of the companies tackling this using generative AI to lower the friction of game creation. You can argue the pre-AI version counts Minecraft and Roblox amongst them.

But if you’re bringing new creators online, you also have to help them make money. In my opinion, you have to solve for distribution or monetization to entrench them into your infrastructure or risk becoming commoditized.

Demand Side: AI slop sea makes curation dramatically more valuable

Now that AI is making game production faster and cheaper, there will be a Cambrian explosion of games - including a ton of AI slop. Curation will become even more important. Steam lowered the transaction cost of purchasing games by moving it online, allowing it to hold an impressive 75% of digital game distribution in the US. But its Yelp-esque reviews aren’t sufficient to efficiently match game and player. A record 19K games were launched on Steam in 2024 (52 a day!) but 80% of them had almost no players.

The Intent Economy

The novelty TikTok introduced was on both demand and supply; on the supply side, it allowed new entrants to be discovered quickly even with negligible followers, but on the demand side, it pushed content rather than waiting for user intent signals like search keywords or followed creators. The motion echoes social media ads that spring on you mid-doom-scroll, except instead of facilitating impulse buying, controlling the content stream itself facilitates impulse binging.

Since the shareware days in the 80-90s, demos have been proffered as a taste test for future purchases. Mobile games have taken this one step further with playable ads, letting you try the game mechanic within the ad itself. Generative AI ads are increasingly commonplace, adapting or generating assets for the audience. But games ultimately still takes time; unlike images or even videos, I can’t grok the gist of a game from a quick glance. Brains show neural activity signaling scene understanding within 250ms after stimulus onset. That’s enough to decide if you want to keep watching a TikTok short clip, but not enough to know if something is fun to play, and our attention spans are ever-shrinking.

Perhaps the answer is hijacking intent even further upstream. Tiktok’s content push shapes the viewer’s interests and introduces them to new categories they would not have discovered on their own. I don’t know what this looks like in the future of gaming - perhaps games-within-games, perhaps mini-games within your ChatGPT conversation, perhaps it is ads adapting to your BCI or wearable signals, perhaps bundled with other entertainment mediums like music or movies as one cohesive experience. The point is finding the player before they even think of finding you - but making them think they did it on their own.

(A)I Told You So

In the 80-90s, internet forums became the water cooler of game discovery as shareware proliferated. Bulletin Board Systems and other gaming specific forums dominated, turning the single-player experience of playing games into a multiplayer social community.

As we form parasocial relationships with our abundance of anthropomorphized AIs, verified-meatbag-relationships become scarce and therefore valuable. Human curated (and ostensibly so) lists become the badge of belonging to a community that will be small but powerful gatekeepers of content. I don’t think human influencers go away; I think they become even more important vis-a-vis other forms of marketing. Even if/when AI influencers become the norm, the identity of being a fellow human follower of the Lunas of the world will still deeply influence who you listen to in decision making.

Product: innovations at the mechanics/system level is key

A lot of the new games focus on extending existing game systems, like AI Dungeon or the generative AI-powered NPC in Cygnus Enterprises, but I have yet to see a novel game mechanic emerge from these new capabilities. Infinite levels/conversations/enemies is not an automatic conversion into joy. No Man’s Sky procedurally generated an approximate 18 quintillion planets, but active gamer count hasn’t broken out past the 10-20K mark. It’s also not entirely new. Rogue was the first procedurally generated game built almost half a century ago in 1980 (driven by memory constraints on storing hand-crafted levels!)

The Neurochemical Flywheel of Fun

Maybe it’s worth first considering what doesn’t change. The hot, wet, three-pound blob that is the human brain is likely to still be in play. We talk a lot about dopamine, but I think there’s a broader neurochemical flywheel:

1. Challenge (norepinephrine): My brief stint in making mobile games was under mentorship of my startup’s head of product, Kriti Sawa, who spent a decade with King and Zynga. She shared about the “sawtooth” tuning of difficulty levels. The difficulty of a level would plateau, then suddenly spike - enough of a challenge to provoke that burst of accomplishment - then flatten out while you sail on that sweet triumph, before climbing again before you get bored. The contrast generates norepinephrine, excitement that increases alertness and memory formation.
2. Anticipate (dopamine): Dopamine ramps up and sustains during the anticipation of an action for a reward (study); in a study on gambling disorder, dopamine neurons’ activity in response to anticipating reward is often greater than the activation observed after the actual outcome (study). This is why tracking progress, seeing a loot area (but not knowing what the loot is yet), and other leading indicators are so compelling. This is also why variety/surprise is important - if you know a plus b equals c, the outcome is no longer as captivating.
3. Belong (oxytocin): In its first year of release in 1978, Space Invaders generated almost $700M in revenue in Japan alone, even causing a nationwide coin shortage. But what it pioneered was more than the specific mechanic, it was the creation of arcades as a third space. The advent of multiplayer turned gaming into a shared experience, which brings the players back in the mid term.
4. Build (serotonin): Create a longer-term identity or achievement. I would compare this to the metagame system in game design, the longer term goals you reach with shorter term actions. You don’t just play Fortnite, you are a Fortnite competitive gamer.

No amount of aesthetic, quantity, or speed can make up for fun. For example, I think that the reason why Axie Infinity failed to sustain its hype ($9B+ token market cap / 2.7M DAUs at peak!) is because the game wasn’t fun! Most of the neuron activity was in token price fluctuations.

So what does change?

UltraHyperSuperMegaPersonalization

We already know game systems adjust to players as pre-defined cohorts; it’s not hard to extrapolate bringing that to the next level of resolution, the individual. But what if we could slice that even further? It’s hard for me to name a specific favourite game, music, or show, because it is so dependent on what I’m in the mood for. Could game engines ingest my HRV, time of day, location, prior activities, and other datapoints to determine the difficulty of the boss, the frequency of lootboxes, the personalities of my NPCs?

So how can we treat “fun” as a learned latent space? Could we use player telemetry data (action sequences, session durations, social interaction patterns etc.) as an input space, and game parameter adjustments as the output space (difficulty curves, resource availability, NPC behavior etc.)? Could weuse RL using session length as a reward signal, or supervised learning on explicit feedback like ratings and reviews? Perhaps we could draw from behavioral economics to define the reward function quantitatively (prospect theory perhaps? given dopamine production is affected by deviations from expectations), or even use biodata signals.

The Rise of the Remixer

Emergent behavior has traditionally been rate limited by the cost of creating enough content to accommodate all possibilities. Now, both the player and the developer can be surprised by emergent outcomes from player interactions as they can now take actions that shape the gameplay itself. The player becomes a collaborator on almost equal footing with the developer.

I suspect that UGG (user generated games) will largely be remixes of existing/studio-developed games rather than new titles from scratch. Consider templates in Notion, Figma, Canva. Counter-Strike and DotA both began life as mods of Half-Life and Warcraft respectively. I write (bad) poetry, but the primary foodstuff of my reading diet is not my own writing; it’s the words of others. I believe that people don’t really know what they want when it comes to entertainment (and in fact, the novelty is part of the draw), and trusting audiences to define their own experience would led to faster horses instead of automobiles. As mentioned, personalization will be powerful here, but only insofar as it reshuffles primitives that the developer has already designed.

Time is All You’ve Got

Infinite generative capacity undermines traditional scarcity. As such, the last frontier rate limiting game consumption is time. Similar to the LiveOps mentioned earlier, LLMs can automate and scale experiences that have value from their transience. Event-based gameplay will become a core feature of the largest games. Consider a reality show-cum-game where audience and participant are the same: the thrill of your game is augmented by the fact that this is the only time it will ever be like this.

Borderless Immersion

(You were waiting for me to mention XR, weren’t you?) Erasing the seams between digital and physical life is going to happen across both utility and entertainment use cases, and games are a clear beneficiary of that. The point is to truncate the middleman between intention and action. Imagine an Animal Crossing that lets you negotiate rent with Tom Nook over a call, collect butterflies in a real-life park via AR glasses, and choose furniture on your tablet. Imagine haptic feedback from your Oura ring, Apple watch, Eight Sleep mattress. Imagine signals directly to your brain to recreate the sensory environment of zero-G horror in Dead Space. Similar to how Sandbox’s adaptation of the Squid Games into VR extended consumer’s relationship with the Squid Games IP or how Pokemon Go resurrected millennials from their office job stupor to reconnect with their childhood world, imagine a VR location that dynamically ports and creates experiences based on whatever story you bring to them on the spot. Going back to the hunger for authenticity in an age of AI slop, the physical connection - even the shared experience of playing VR in the same studio - could become key to immersive experiences.

Conclusion: Gotta Catch ‘em All

Nothing is true, everything is permitted.
Assassin’s Creed Motto

The platform that has the holy trinity will be the next gaming behemoth

Roblox is a fantastic example of a company that got all three right, owning production, distribution, and runtime. It is crucial that Roblox captured a new generation of gamers: ~40% of Roblox gamers are under 13 years old.

You don’t have to wait for today’s babies to learn how to use their opposable thumbs in order to rapture a greenfield audience. The smartphone form factor allowed for short-session portable games that lowered the barrier to entry for people who wouldn’t typically count themselves as “gamers”. As the lines demarcating “game” blur, so does the demographic profile of “gamer”. Who are the adjacent consumers you can bring into your fold?

Gamification still exists

The term “gamification” is so over-used and over-worn and my insides are groaning as I type this. But I still think it works! (I’m biased of course, given that was the entire thesis of my startup). Medical adherence continues to be a huge challenge in efficacy. Retention and engagement determines the sustainable quality of revenue for any consumer and prosumer application. A post for another time, but if we can think of gamification beyond points and leaderboards, I believe it’s a crucial pillar of any product design.

Of course, if I knew exactly what the killer next game or game infrastructure would be, I’d probably try building it. Agree? Disagree? Yes-and? No-but?