Building with AI
Building with AI: raising the standard of play
There’s lots of ways to be, as a person. And some people express their deep appreciation in different ways. But one of the ways that I believe people express their appreciation to the rest of humanity is to make something wonderful and put it out there. And you never meet the people. You never shake their hands. You never hear their story or tell yours. But somehow, in the act of making something with a great deal of care and love, something’s transmitted there. And it’s a way of expressing to the rest of our species our deep appreciation.
That quote is the spine — not a decoration at the top of an essay about systems. Building with AI is not a fashion wave, and not a slide that says “AI-powered.” It is the same act Jobs named: make something wonderful, with care and love, and put it out there for people you may never meet. Everything else in this note — boss fights, Nemesis, live ML stacks, the Create / Compute / Curate paradox — is evidence and craft in service of that spine. Not a replacement for it.
Games gave designers that duty long before today’s power budgets of future compute. Boss fights that demand excellence. Adaptive systems that read skill without humiliating it. Character systems that remember. Live stacks that personalise journeys and offers. Those are how the industry has sometimes tried to transmit care through play — raising the human standard of excellence so that what ships can still feel wonderful. The large language model did not invent the duty. It only changes the cost of Create while Compute scales — and leaves Curate as the scarce human work.
First principles: the fight that brings out the best
A great boss fight is not merely a damage check. It is a designed encounter with a player’s skill, attention, and courage. The best ones sit just beyond current competence — readable, learnable, fair under pressure — so that clearing them feels like the player becoming more than they were an hour ago.
Dynamic and differential difficulty grew from the same impulse: systems that observe performance and adjust pacing, telegraphing, health budgets, or aggressiveness so the encounter stays in the zone where mastery is possible. Long before GPU clusters and transformer inference costs dominated industry conversation, designers were already asking: how do we tailor the experience so human skill shows up at its best?
That question is how “make something wonderful” often sounds in a combat encounter — product care wearing a boss-fight skin. Behaviour trees, utility AI, perception graphs, state machines that look alive — these were not demos for investors. They were craft for players who would feel the difference in the tenth attempt, not the first trailer.
Nemesis: memory as a design material
Monolith’s Nemesis System in Middle-earth: Shadow of Mordor (2014) and Shadow of War (2017) remains one of the clearest proofs that adaptive intelligence in games is a design achievement, not a model-card achievement.
Enemies remember. They rise when they defeat you. They scar, rename, climb a hierarchy, and return with a personal history. Rivalry becomes emergent narrative. The player’s excellence — or failure — writes the next chapter of the opposition. Warner Bros. later secured U.S. Patent 10,926,179 (“Nemesis characters, nemesis forts, social vendettas and followers in computer games”), which is still active for years to come. Whatever one thinks of locking such a system behind IP walls, the craft lesson is unmistakable: personalisation of challenge and story was a solved design problem for human drama, using rules, memory, and hierarchy — not generative fill.
Nemesis did not need a trillion-parameter model to make players feel seen by their world. It needed designers who cared enough to make the world remember them.
The live business already ran machine learning
Parallel to systemic combat AI, the business of live games spent a decade industrialising prediction, matching, and segmentation. That stack is often invisible to players — and easy to forget when the industry’s attention swings to chatbots — but it is applied intelligence at scale.
AppLovin · Axon
AppLovin’s Axon (sometimes styled Axon AI) is the company’s AI recommendation / advertising engine: predictive algorithms that match advertiser demand with publisher supply, bid on impressions against return goals, and continually train on engagement signals. In mobile UA, the practical story many operators know is stark — creative and goals in; machine matching out — with deep learning generations (notably Axon 2.0 mid-2020s) compressing sparse behavioural IDs into dense spaces so the network can learn high-order relationships without hand-built targeting trees.
Agree or disagree with any particular monetisation outcome: the technical point stands. Games and game ads trained the industry to treat attention as optimisable signal long before generative models wrote copy. Personalisation of who sees what, when, under real auction load, is not a 2024 invention.
Unity Vector · audiences · segmentation
Unity’s growth stack makes the same lineage visible from another angle. Unity Vector is positioned as the AI core of Unity’s user-acquisition platform: matching players to games using behavioural and “game DNA” signals across Unity’s ecosystem — aiming at install quality, retention, and monetisation fit rather than vanity CPI alone.
Alongside UA, Unity Analytics Audiences and custom audience builders let studios segment by behaviour — recent purchasers, new players, churned, unengaged — and drive Game Overrides: tailored content, configs, and journeys for named groups. That is segmentation-based tactics in product form: personalised offers and player journeys as an operating system for live service, not a one-off A/B curiosity.
Put simply: Nemesis personalised rivalry. Axon-class systems personalised match and monetisation signal. Vector and audience tooling personalised who the game serves next. Different layers — useful history, not the spine. They matter when they help someone put something more wonderful into a player’s hands; they fail the moment optimisation replaces care.
When Create is cheap and Compute is growing — who has Curate?
Name the paradox plainly. Create got cheap: drafts of art, dialogue, code, offers, briefs, and analyses arrive in minutes. Compute is growing: more tokens, more GPUs, more surfaces that can spit out something that looks finished. Curate — taste, judgment, refusal, the hundred iterations until the thing is delightful — did not get cheaper. If anything, it got scarcer.
If you are out there shipping, you are already fighting the slop. Not as a slogan. As the daily work of deciding what deserves a player’s minute.
We have seen this movie. Hypercasual lowered the barrier to shipping a game that looked like a product — volume rose; the fight moved to taste, retention, and whether anyone still cared after the install. Short-form video did the same to attention: creation exploded; the scarce skill became editing, pacing, and knowing what not to post. Generative AI is the next wave of the same curve — not because games never had intelligence in the loop, but because the cost of a plausible draft just collapsed again.
Games have been fortunate in one human way that many industries envy: makers and players still gravitate toward each other. Rooms that remember teammates for decades. Communities that form around a title. Players who feel like family when the service is honest. That gravity is not automatic. It is curated — through care in the loop, the live plan, and what we refuse to ship. The Create / Compute boom only makes that humanist duty louder. Let’s keep makers and players finding each other.
This is where many “AI transformations” quietly die: they celebrate Create and Compute, and forget Curate — the Jobs standard of care and love transmitted through the thing you put out there, for people you may never meet.
Holding the spine while you build
The spine does not change with the model generation. It is still: make something wonderful and put it out there — care and love transmitted to people you may never meet. In games, that often shows up as raising human excellence in play — and as keeping makers and players gravitating toward each other. The checklist below is only how we keep Curate from losing to Create and Compute.
- Ask whether it is wonderful — before whether it is shippable. What skill, joy, mastery, or belonging should this raise for a real player? If the honest answer is only “engagement minutes” or “model coverage,” you have left the spine.
- Learn from the lineage; do not worship it. Adaptive difficulty, Nemesis-class memory, ML UA, segmentation journeys prove games were early. Hypercasual and short-form already taught the Create boom / Curate scarce lesson. Case studies in care under constraint — not a substitute for taste now.
- Treat personalisation as responsibility. Offers, difficulty, and journeys that know the player can serve them — or exploit them. Wonderful and extractive are not the same word.
- Curate is how wonderful survives Create. One hundred iterations is not nostalgia; it is how delightful things get made when drafts are cheap. Humans in the loop are not a failure mode. They are how care stays in the product — and how makers and players keep finding each other.
A note from Yardstick
Games were always about the people who made them and the people who played them — teams that remember teammates for decades, companies that become communities, players that become family. That is the humanising heart of how we offer Custom AI Transformation and partnership craft: commitment, people experience, and craft expertise you cannot train a model on — then tools that amplify those materials when the drafts get cheap.
We will not sell autonomous “AI PM” theatre, vanity chatbots without a product problem, or claims we cannot seal. Care decides; compute follows. Loop, economy, and live-plan work live under how partnership works and AFITMRR™. Either way, the ask is the same as the spine of this essay: make something wonderful — and put it out there with care.
Make something wonderful
For people you may never meet — and for the rooms that stay. If that is the work you are on, we would like to hear where you are stuck.