How AI Is Already Transforming Games: 17 Examples Analyzed Through Octalysis (Part 2)

In Part 1 of this series, I focused on the broader relationship between AI and gamification.

This article looks at the practical side: 17 real examples of AI in games that change how players feel, adapt, and stay engaged.

The key question is not whether AI is present. The more important question is which motivational forces the AI is strengthening.

About the Creator of the Octalysis Framework

Yu-kai Chou created the Octalysis Framework after studying gamification since 2003 — years before the term entered mainstream vocabulary. As a Human-Systems Architect & Behavioral Designer, his framework has been applied by LEGO, Microsoft, Porsche, Coca-Cola, Salesforce, and MrBeast, impacting over 1.5 Billion Users.

Chou has taught the Octalysis methodology at Harvard, Stanford, Yale, Tesla, Google, BCG, and IDEO.

His work has been cited by Harvard, Stanford, MIT, Forbes, Wall Street Journal, Wired, US Department of Energy, NIST, NSF, NCBI, US Department of Education, ClinicalTrials.gov, and 3,700+ more academic publications. Explore his books here.

Why AI Matters in Games Through Octalysis

AI matters in games because it can make systems feel less fixed and more responsive.

That responsiveness affects motivation. It changes whether a player feels surprise, ownership, tension, mastery, creativity, or social attachment.

When you analyze AI systems through Octalysis, you stop asking whether the game is using AI and start asking what emotional and behavioral effect the AI is producing.

One useful lens here is whether the AI is creating more White Hat motivation, more Black Hat pressure, or a volatile mix of both. That distinction matters because the same adaptive system can feel empowering in one game and manipulative in another.

The top half of the octagon is where AI tends to feel generous. The bottom half is where the same adaptive tech turns into dread, urgency, and fear of loss.

In the 17 examples that follow, notice how often the headline tech (behavior tree, utility AI, LLM companion, procedural generator) is almost beside the point. What I care about is the Core Drive the tech is pushing on and the mechanism by which the push happens. When you can name that mechanism, you can copy the effect into products that have nothing to do with games.

A quick legend before the examples. The Octalysis Framework maps human motivation across 8 Core Drives, and I refer to them by number throughout this post for tightness. Core Drive 1 (CD1) is Epic Meaning & Calling. Core Drive 2 (CD2) is Development & Accomplishment. Core Drive 3 (CD3) is Empowerment of Creativity & Feedback. Core Drive 4 (CD4) is Ownership & Possession. Core Drive 5 (CD5) is Social Influence & Relatedness. Core Drive 6 (CD6) is Scarcity & Impatience. Core Drive 7 (CD7) is Unpredictability & Curiosity. Core Drive 8 (CD8) is Loss & Avoidance. The cocktail effects (CD7+CD3, CD4+CD8, and so on) are where the strongest game AI lives, and that is the lens for the 17 examples below.

Advanced Enemy and NPC Behavior

The first cluster is enemies and non-player characters that feel less scripted. Four games show the arc, spanning two decades of behavior-AI work.

1. F.E.A.R. (2005): Goal-Oriented Action Planning

Monolith’s Replica soldiers still hold up as a high-water mark for combat AI. They use Goal-Oriented Action Planning, which means the AI does not run a fixed script. It is handed a goal (“kill the player”) and searches for a valid plan given the current world state. When you duck behind cover, one soldier suppresses while another flanks. When you take out the flanker, the survivor calls for backup and goes to ground.

Octalysis read: the dominant drive is Core Drive 7 (CD7): Unpredictability & Curiosity, because the same room replayed feels different every time. But the magic is the cocktail. CD7 pairs with Core Drive 2 (CD2): Development & Accomplishment when you out-think a flank you have seen before, and with Core Drive 8 (CD8): Loss & Avoidance when the enemy audio chatter (“he’s in the vent!”) tells you exactly how exposed you are. The audio is classic Evolved UI: the AI’s internal state leaks into the world as sensory feedback so the player can feel it without being told.

2. Alien: Isolation (2014): Two-Layer AI as Engineered Dread

The Xenomorph in Alien: Isolation runs on a two-layer architecture that is worth studying on its own. A Director AI always knows exactly where the player is. The Xenomorph AI does not; it only has sensory cues. Hints arrive from the Director in the shape of “go search this area,” never direct coordinates. That gap between what the system knows and what the alien knows is the entire emotional experience.

That gap is pure CD7 + CD8. You cannot save-scum your way out, because no encounter is scripted. The AI learns your habits inside a single playthrough: if you hide in lockers repeatedly, the alien starts checking lockers. This is personalization used against the player, and it is some of the most honest Black Hat Unpredictability design in the medium. The 2024 VR spin-off Alien: Rogue Incursion carried the pattern into PSVR2 with even tighter sensory loops, which is why it lands as viscerally as the original on a platform where you cannot look away.

3. The Last of Us Part II (2020): Social Recognition Weaponized

Naughty Dog made a specific, unforgettable choice. Enemy NPCs call each other by name. When you shoot one, another screams a name (“Jordan!”), runs to the body, grieves. The dogs have names too. The player is denied the usual shooter mental shortcut of seeing mobs.

Mechanism: Core Drive 5 (CD5): Social Influence & Relatedness used inverted. CD5 is usually how you build attachment; here it is how the game builds guilt. CD5 + CD8 + CD7 cocktail, but the CD5 direction is unusual — in most shooters, CD5 creates belonging with your squad; here it creates recognition of your victims. Few games have the courage to use CD5 against the player. That is the design choice worth stealing, even if the shape of the reuse is nothing like a shooter.

4. Red Dead Redemption 2 (2018): Persistent Social Memory

Rockstar layered a world-memory system under ambient NPC AI. The bartender remembers your tip. The gunsmith recognizes you from a crime three in-game days ago. Your horse bonds through repeated care and panics if you try to mount a stranger’s horse. Townsfolk gossip after you ride out.

Octalysis: Core Drive 4 (CD4): Ownership & Possession dominates because “my horse” and “my reputation” become load-bearing identity. CD5 layers in (the world recognizes me) and CD7 adds the live wire (what will this stranger do?). The Alfred Effect is the explicit Game Technique at work: small personalization signals compound into a sense that this world knows me. Rockstar spent hundreds of millions teaching us the principle; the cost to copy it into a loyalty program is a database column.

Procedural Generation and Generative AI

The next cluster uses AI not to make characters smarter but to make worlds and stories bigger than any human team could hand-build. Three examples, each a different slice of the same motivational engine.

5. No Man’s Sky (2016, ongoing): 18 Quintillion Planets, One Core Drive

Hello Games set a deterministic procedural seed system so every player, on every machine, generates the same galaxy, with what the studio famously marketed as 18 quintillion planets. Each planet has procedurally composed biomes, creature morphology, flora, and weather.

The Octalysis read is almost pedagogical. Core Drive 7 (CD7) is the beating heart (what is on the next planet?), but it is the pairing that made the launch-era criticism finally dissolve. Core Drive 3 (CD3) comes in through naming discoveries and terraforming; Core Drive 4 (CD4) comes in through your base, your freighter, your multi-tool. The famous marketing line “no two players will meet by chance” was Core Drive 7 (CD7) stretched to an existential scale, and when the creativity loop arrived in post-launch updates the game flipped from hollow to evergreen.

Hello Games’ 2024 “Worlds Part 1” update added new atmospheric and water generation systems. The pattern did not change; the fidelity did. That is the honest lesson: procedural systems age well when you keep buffing the surface without rebuilding the spine.

6. Minecraft (2011, ongoing): Blank Fill at Infinite Scale

Minecraft’s terrain uses Perlin noise plus structural rules for caves, biomes, villages, and strongholds. The AI is not flashy. What it produces is a guaranteed-novel canvas every time you start a new world.

Core Drive 3 (CD3) dominates. This is the Blank Fill Game Technique at planetary scale: the empty plot of land is the motivator, and the procedural world gen quietly ensures your plot is never identical to anyone else’s. Core Drive 4 (CD4) stacks on (my base, my server, my children’s creations). Core Drive 7 (CD7) comes in through caves and ravines, where “what’s down here?” replaces story. Notice that Minecraft has almost no scripted NPC intelligence, and it does not need any. When CD3 is the load-bearing drive, the world just has to stay surprising enough to protect the canvas.

7. AI Dungeon (2019): The LLM World, and the CD3 Cautionary Tale

Before ChatGPT, Latitude’s AI Dungeon was the first mass-market LLM-driven text adventure. The AI is the world. Any input becomes valid story. That is Core Drive 3 (CD3) at maximum amplitude, paired with Core Drive 7 (CD7) that never quite repeats.

What happened next is the part worth remembering. In 2021, safety-layer changes and content filters made the AI refuse, redirect, and break character. The user revolt was immediate and brutal.

Lesson: Core Drive 3 (CD3) requires permission to matter. The moment the AI says “I cannot do that,” the creativity drive collapses, and no amount of CD7 novelty rescues it. Every product team shipping an LLM feature today inherits this constraint.

The 2024 title Suck Up! (a fully LLM-driven vampire social game) held the line and kept CD3 intact, which is why it found its audience while more restrictive LLM experiments quietly disappeared.

Dynamic and Personalized Experiences

The third cluster adapts the experience itself to the player. These systems do not generate worlds; they generate relevance.

8. Skyrim (2011): Level Scaling and the Radiant Quest System

Bethesda’s Radiant Quest system procedurally composes new quests from templates (“kill bandit X at location Y for NPC Z”) and the world scales to your level. Every random dragon attack is tuned so the encounter stays meaningful at hour 10 and hour 300.

Octalysis: Core Drive 2 (CD2) is the intended drive (challenge stays live), Core Drive 7 (CD7) rides along (which contract will the Dark Brotherhood offer next?), and Core Drive 4 (CD4) stacks through owned homes and gear. The risk is also a teaching moment: Skyrim is criticized for scaling so tightly that mastery loses its edge. When an adaptive system matches the player too perfectly, CD2 hollows out, because mastery requires the world to sometimes out-class you or be easier than you. A system that always meets you at your exact level becomes a treadmill. The fix other open-world games found after Skyrim is locked zones, areas where the level stays fixed and CD2 has somewhere to breathe.

9. Detroit: Become Human (2018), and Baldur’s Gate 3 (2023): Visible Choice Webs

Quantic Dream built a branching narrative with reportedly hundreds of nodes per chapter and then did something unusual. After each chapter, the game shows you the entire flowchart — every path you took, every path you missed, every character who lived or died in your playthrough.

That flowchart is the design choice. The AI sits underneath (evaluating state, routing outcomes), but the motivational hit is the reveal. Core Drive 4 (CD4, my path, my story) is multiplied by making the ownership visible. Core Drive 5 (CD5) activates when you see which characters your choices preserved, and Core Drive 1 (CD1) kicks in because the branching forces you to confront the meaning of each choice retroactively. The Game Technique is Narrative Autonomy with Consequence Maps.

Larian’s Baldur’s Gate 3, still drawing a long tail of 2025-26 players, extends this pattern by letting companion approval systems react to every micro-choice. You never see the flowchart, but the companions tell you when you moved them. Two different expressions of the same CD4+CD5 cocktail, and both work because the game trusts the player to notice.

Innovative AI Systems

The fourth cluster is where the AI architecture itself became the design. Each of these introduced a pattern the industry has been chasing since.

10. Middle-earth: Shadow of Mordor (2014): The Nemesis System

This is the clearest case in the article. Monolith built a procedurally populated antagonist roster. Each Orc captain has a generated name, face, voice traits, strengths, weaknesses, and (critically) a memory of every encounter with you. Kill one, another rises and inherits rank. Die to one, that Orc remembers, promotes, taunts you at the next meeting with a callback to how he killed you.

This is the pure Core Drive 4 + Core Drive 7 + Core Drive 5 (CD4 + CD7 + CD5) cocktail, and it is worth naming each pull. CD4 appears because the antagonist becomes your antagonist; no other player has Ratbag the Coward the way you do. CD7 appears because the roster never stabilizes; new captains climb when old ones die. CD5 appears because Orcs taunt you by name, because they recognize the player as a specific person in their world, not a camera. Game Technique: Dynamic Personal History.

Warner Bros. patented this system in 2021. That patent is why no shipping game has reproduced the Nemesis pattern at its original fidelity, despite a decade of designers trying. The industry lesson is not really about AI. It is that a motivational system strong enough to become its own IP can be locked up the same way visual assets can. Designers working on live-service products should notice which of their own loops would be worth protecting that way.

11. Grand Theft Auto V (2013): Context-Aware Pedestrian AI

Rockstar’s pedestrian system uses behavior trees layered with contextual modifiers. An NPC reacts differently based on your wanted level, weapon drawn, vehicle you arrived in, time of day, and neighborhood. Shoot in Vinewood Hills, you get LSPD helicopters fast. Shoot in Grove Street, you get a slower, rougher response. Most of this depth is invisible, which is the point.

Core Drive 7 (CD7) is live (will this cop call backup?), CD5 threads through (the city knows what kind of person I am), and Core Drive 2 (CD2) activates for skilled players who read the signals and time their heists. The reason the system feels so alive is that the player does not know where the AI stops and the scripting starts. Opaque design is a feature when CD5 is load-bearing.

12. Halo: Combat Evolved (2001): The Thirty Seconds of Fun

Bungie designed Covenant AI around behavior trees with explicit social roles. Grunts flee when an Elite dies. Elites rally nearby Grunts. Jackals drop shields only under specific fire patterns. Jaime Griesemer’s famous “thirty seconds of fun” loop is what this AI enabled: a short, repeatable encounter shape that felt different every replay because the social script gave each unit a readable personality.

Core Drive 2 (CD2) is the dominant drive (mastery through pattern-reading) with Core Drive 7 (CD7) right next to it (this encounter might go sideways). The deeper lesson is readability. Halo’s AI is not the smartest on this list, but it is the most legible. Players could learn the Covenant’s rules in an hour and spend the next ten years finding edge cases. That legibility is a form of White Hat CD2 design that most modern “advanced” AI sacrifices for opacity.

Strategy and Simulation Games

In strategy and simulation, the AI is the product. A bad opponent collapses the whole experience. Three different approaches to the same problem.

13. Civilization VI (2016): Leader Agendas as Personality Engines

Firaxis gave each leader two agendas. A public agenda (visible in diplomacy) and a hidden agenda (revealed only through play). Teddy Roosevelt hates wars on his continent. Gandhi’s famously canonized nuke obsession — a wink-back to a long-running community myth — was wired into his hidden agenda. These are not tuning parameters; they are characters.

Core Drive 5 (CD5) dominates (rival relationships feel like actual relationships), Core Drive 7 (CD7) adds the hunt for hidden agendas, and CD2 kicks in when you learn to manage each leader’s psychology across a 300-turn game. The Game Technique is Social Personality: give the AI a knowable soul and the strategic AI’s limitations become features, because “Teddy would do that” is a satisfying explanation for any AI move. The lesson generalizes, because every AI agent you ship to users benefits from a knowable personality, because users forgive mistakes that feel in-character and resent mistakes that feel random.

14. XCOM 2 (2016): Permadeath and the CD4-to-CD8 Chain

Firaxis combined procedural maps, Advent behavior trees, and a pod-activation system where enemies spawn in clusters that patrol until you break line of sight. Every mission is new. More importantly, every mission has permadeath on a unit roster you customize and nickname.

Core Drive 8 (CD8) is the dominant drive (permadeath + mission timers), CD7 is the runner-up (fog of war, unpredictable pod composition), and CD2 is the mastery layer. The signature move is how the game engineers a CD4 to CD8 Chain: you nickname a soldier “Sharpshooter-Dad,” give him a custom voice, level him to Colonel over twenty missions, and then the design dares you to risk him. By the time the fear hits, the ownership is irreversible. That chain is a reusable behavioral-design pattern outside games. Loyalty programs, professional reputations, and long-term savings accounts all use the same CD4 to CD8 architecture to keep people engaged when the content itself is thin.

15. StarCraft II (2010): The Cheater Tiers

Blizzard’s AI ships with seven difficulty tiers, and above Hard the AI literally cheats. Insane, Cheater 1, Cheater 2, and Cheater 3 all grant the AI extra resources, faster production, or map vision the player cannot earn. Blizzard publicly admits this.

Why does this work? Because Core Drive 2 (CD2) is the entire point. At high tiers, the player is not competing with a fair opponent; they are racing against a rigged system and still winning. The cheat is the challenge. This is the Deep Blue principle in miniature: when the player’s mastery exceeds what a fair AI can produce, the honest design move is to admit the cheat and let CD2 absorb it. Any live product with a skill curve runs into this eventually. Sandbagging your top 1% of users ruins their CD2 experience, and fake-random difficulty feels worse than declared difficulty.

Generative AI NPCs and New Frontiers

The last cluster is the frontier: LLM-driven companions and narrative engines. The technology is real, but the motivational question is still half-answered.

16. Convai and the Cygnus Enterprises Moment (2023 onward)

Cygnus Enterprises shipped with LLM-driven NPC conversations, powered by Convai. NVIDIA’s ACE tech demos at GDC 2024 and CES 2025 showed the same pattern with better voice synthesis and facial animation: talk to the blacksmith, get a novel response, carry the thread across a session.

Core Drive 5 (CD5) is the goal (the character feels like a real presence), CD3 rides along (your conversation shapes outcomes), CD7 stays active (you never get the exact same answer twice). In 2025, the honest assessment is that this is still technically impressive and motivationally uneven. The lead problem is not AI quality. It is context persistence — the companion that does not remember your promises feels worse than the scripted NPC who at least stays in character. Games that solve persistence first (even with a small memory window) will outperform games that chase raw response quality. The design lesson echoes the product lesson: a small amount of reliable memory beats a large amount of novel generation.

17. Hades II (2024, ongoing): Graph Walker Narrative AI

Supergiant’s second dungeon roguelike went into early access in 2024 and has been iterated throughout 2025-26. Its narrative system, the Graph Walker, carries reportedly more than 21,000 voice lines and dynamically selects dialogue based on your run state, death count, character relationships, and prior choices. It is not an LLM. It is a hand-authored system using AI-style selection logic, and that distinction matters.

Because the content is authored, Core Drive 5 (CD5, companions remember you) and Core Drive 4 (CD4, Melinoë’s story is yours) hits land clean, with no uncanny-valley moments. CD7 layers in (which line fires this run?) and CD1 surfaces because the relationships across runs accumulate into real meaning. Hades II is the current best-case shipping example of generative-feeling narrative without generative-tech fragility. Supergiant chose hand-written scale over LLM novelty and the players can feel it. That trade-off is a design decision every product team evaluating AI features should study before they ship.

Even older dungeon roguelikes like Diablo IV (seasonal variance, adaptive enemy packs) and Persona 5 (Confidant dialogue trees that respond to long-arc relationship state) show how adaptive variability keeps long-term loops fresher than static content alone. In every case, the AI is earning its place by serving a specific Core Drive, not by being clever for its own sake.

The Core Drive That Dominates AI in Games

Across these 17 examples, one pattern stands out: Core Drive 7 (CD7) appears almost everywhere.

That is not accidental. AI matters in games largely because it makes outcomes harder to fully script, predict, and exhaust. Every one of the games above uses AI to push on “what happens next?” in some form.

But CD7 alone is thin. The strongest systems pair unpredictability with a second drive that gives the surprise somewhere to land:

• CD7 + CD3 for evergreen creativity systems (No Man’s Sky, Minecraft, AI Dungeon).
• CD7 + CD5 for living social worlds (TLoU2, RDR2, Civ VI, Hades II).
• CD7 + CD2 for earned mastery (F.E.A.R., Halo, StarCraft II, XCOM 2).
• CD7 + CD4 for personal stories and rivalries (Shadow of Mordor, Detroit, Skyrim).
• CD7 + CD8 for tension and fear (Alien: Isolation, XCOM 2 again).

Games that use AI only for randomness almost always feel shallow. The randomness has to attach to something the player cares about, or it becomes noise. That is the single most important lesson from the 17 examples, and it generalizes far beyond games.

What Game Designers and Product Teams Should Learn

The bigger lesson is not about games. Games are the market’s research-and-development lab for engagement design. By the time a pattern ships in a 10-million-player title, it has been stress-tested against human motivation in ways no A/B test can match. When the same pattern shows up in enterprise software three years later, it usually works.

Four practical takeaways for product teams:

1. Name the Core Drive before you pick the tech. “We need AI for personalization” is not a design. “We need CD4 ownership to compound across sessions so churn drops” is.
2. Pair CD7 with a second drive. AI-powered novelty is cheap now. What is scarce is the companion drive that gives the novelty meaning. Pick yours before the sprint starts.
3. Treat memory as a Core Drive feature, not an engineering task. The Nemesis System, RDR2’s bartender, and Hades II’s Graph Walker are all memory systems. The industry keeps rediscovering that persistent state is where CD4, CD5, and CD1 actually live.
4. Audit with the 8 Core Drives, not the tech stack. If your AI feature does not pass a simple CD-by-CD audit (“does this make users feel more powerful, curious, connected, or meaningful?”), it is probably a feature in search of a motivation.

Want to apply this to a real product? The fastest way is the Octalysis Framework — pick the two Core Drives your AI feature should pull on, then look at the table above to find the game pattern that already proves it can work.

For the broader view, read Part 1 of this series. For the underlying motivational structure, start with the Octalysis Framework. For the Game Technique library behind the examples, see how the 8 Core Drives combine in cocktails.

Frequently Asked Questions

What is the main Core Drive AI strengthens in games?

Most often it is Core Drive 7: Unpredictability and Curiosity, because AI helps games produce less fixed and less exhaustible outcomes. The strongest systems always pair CD7 with a second drive (creativity, mastery, ownership, social, or loss) so the novelty has somewhere meaningful to land.

Why is AI in games more than just better NPC realism?

Because the real value is motivational. AI changes tension, mastery, surprise, ownership, and social response, not surface realism alone. F.E.A.R. in 2005 did not look realistic by modern standards, yet its AI still ranks among the best because it produced the emotional effect the design wanted.

What happens when AI only adds randomness?

Randomness by itself almost always feels hollow. The stronger systems pair unpredictability with another Core Drive. Without that pairing, CD7 drifts into the slot machine pattern: novel, addictive, and ultimately empty.

Which of the 17 examples should product teams study first?

For CD4 and CD5 systems, Shadow of Mordor’s Nemesis System is the gold standard; study the memory architecture. For CD3 at scale, study AI Dungeon’s 2021 collapse to learn what breaks when creativity loses permission. For narrative without LLM fragility, study Hades II’s Graph Walker. Each one teaches a different structural lesson.

Why should non-game product teams care about these examples?

Because games often prototype the future of adaptive engagement. The same principles shape onboarding, learning, loyalty, and personalized product experiences. A loyalty program with persistent social memory is RDR2 applied to retail. A learning platform with CD4-to-CD8 chains is XCOM applied to upskilling.

Is the Nemesis System still the gold standard for AI antagonists in games?

Yes, and that is partly because Warner Bros. patented the Nemesis System in 2021, so no other shipping game has reproduced it at the original fidelity. The pattern remains the clearest CD4 + CD7 + CD5 cocktail in any commercial game, and the design lesson generalizes far beyond games.

What is the difference between AI-driven novelty and shallow randomness in games?

AI-driven novelty pairs Core Drive 7 (Unpredictability) with a second Core Drive — creativity, mastery, ownership, social, or loss — so the surprise lands somewhere meaningful. Shallow randomness uses CD7 alone and feels hollow because nothing the player cares about is at stake. Pairing CD7 with a second drive is the single most important lesson from the 17 examples.

What is the best practical takeaway from these 17 examples?

Do not add AI for hype. Start with the motivational outcome you want, then use AI only when it helps strengthen that specific Core Drive more effectively. If you cannot name the Core Drive before you pick the tech, you are almost certainly shipping a feature with no motivational spine.

Related Reading

• How AI Is Transforming Gamification: Part 1, The Overview
• The Octalysis Framework: Complete Gamification Framework
• How the 8 Core Drives Interact: Cocktail Effects, Chains, and Traps
• Stop Building Horses: The Mouse-to-Horse Game Loop Design Principle
• How Kimi AI Cheated Me Out of $200: Here’s Exactly How

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?