The [Save Game] newsletter is back! After TinyLetter shut down I took a hiatus, but people keep asking me to bring it back, so here is an extra meaty one. If you are new to this series, please consider subscribing! It’s free, and in this age of social media atomization I have been finding more joy than ever in the circle of newsletters that brighten my emails.
Last week I went to the 2026 Game Developer’s Conference (GDC) and filled my notebook up with interesting lessons from all sorts of developers. I have pretty complicated thoughts about this year’s conference as a whole compared to the previous years, but I’d like to keep things focused on the juicy substance of game development talks.
Please keep in mind that all of my notes are, at best, paraphrases of the original author’s words, and many of them have lots of editorializing by me based on my interpretations and connections to my own experiences.
Without further ado, here are my favorite insights from a random scattering of talks:
Flavors of Challenge: The 8 Flavors of Difficulty and How They Can Be Combined to Make Better Difficult Games
By Brett Moody
Moody proposed a taxonomy of different kinds of challenges that games can contain, in order to better understand why some games motivate people to overcome adversity while others don’t. This talk was very exciting to me, because I had been assembling my own chart of “flavors of fun,” so I wanted to see where our observations aligned and differed.
He identified eight flavors of challenges:
- Reasoning
- Critical thinking (Pentiment)
- Deductive reasoning (Return of Obra Dinn)
- Pattern recognition (Chess)
- Physicality
- Hand-eye coordination
- Strength
- Balance
- Accuracy
- Agility
- Randomness
- Input randomness (starting seeds, like Balatro or Spelunky)
- Output randomness (differences in outcome, maybe like an attack dealing random amounts of damage within a range?)
- I didn’t understand this split very well, but for research I picked up Uncertainty in Games by Greg Costikyan at the GDC gift shop, hoping it will have some insights.
- Endurance
- This was one I hadn’t considered at all! Moody defined endurance challenges as the ability to “sustain concentration or focus”
- Guitar Hero’s 8+ minute long level based on Dragon Force’s “Through the Fire and Flames” is not the most complicated song, but is an endurance test
- Celeste and Super Meat Boy are a series of individual sprints, so the game is lower on the endurance flavor, while Getting Over It With Bennett Foddy is pure endurance
- Out-of-game resources
- Time (very prevalent in board games, but all games weigh these facets differently)
- Setup (waiting for player input, can build anticipation)
- Play (taking your turn, doing an action)
- Tedium (time it takes for actions to complete)
- Waiting (time before you can input again)
- Money (Poker, coin-op beat ‘em ups)
- Knowledge (trivia games)
- Time (very prevalent in board games, but all games weigh these facets differently)
- Representation
- Engaging with deeper themes can be difficult (Molleindustria games used as example)
- Obscenity (Kane and Lynch 2: Dog Days)
- Interpersonal Skills
- Reading Non-verbal cues
- He gave a bunch of other examples, but sadly I didn’t have time to write them down
- In-game resources
- Scarcity (Survival games)
With those eight flavors in mind, Moody offered various strategies for keeping players from feeling hopeless when faced with difficult challenges:
- Pair skill growth with virtual growth
- Consistently award stronger equipment, or stat growths
- Common in RPGs or Dark Souls
- Parallel Challenges
- If you get stuck on something, have somewhere else to go
- Look at Baba is You
- Vary the flavor of challenges
- It can keep up excitement to know that once you overcome one challenge there will be a reprieve with a different kind of challenge
- Rewards should provide momentum and direction
- Defeating one Megaman boss helps push you through another
- Reward all efforts, not just victory
- If effort doesn’t feel worth it, players burn out
- Modern RPGs let players keep the experience points earned even if they die
- Foreshadow big challenges early
- Keeps motivation up
- Breath of the Wild’s omnipresent castle creates excitement
- Make retrying easier than quitting
- If it takes more effort to quit out of a game than to try again, players are less likely to ragequit
- This was the key innovation in Super Meat Boy
- Avoid “time-taxes”
- Redoing challenges you have mastered creates boredom
- Moody offered criticism of Bloodborne’s run-backs, which some people might disagree with, but that just goes to show the subjective nature of people having different tastes for different flavors of challenge.
- Never restart from zero
- Rogue-lites cool features where players start their next run with a random bonus item/ability
- Maintains momentum between runs, creates variety
- Give suffering meaning
- A game’s story can make frustration part of the thematic arc. Players will persevere if they feel something deeper than fun.
- Give players a common enemy
- Being part of a larger movement can be motivating, making the struggle feel less lonely and more meaningful
- Establish narrative stakes
- If players get attached to characters then they will move mountains to get narrative satisfaction
I think these strategies are wonderful. Moody is spot on that every player will be interested in different mixtures of challenges, and some people will be more engaged in a game than others. If someone isn’t on the same wavelength with the game, they might drop out from frustration. These techniques could keep that person from giving up, and that could lead to them discovering a flavor they didn’t appreciate before. Growing players’ palettes deepens the culture of play, so I am keen to keep this list handy. Thanks for the great talk, Moody!
Game Designer’s Notebook: Five Obstacles to Successful Gen AI Games
By Jesse Schell
I’ll be a little rude and say that I thought the majority of this talk was delusional. Schell asserted that silent movies were not considered worthwhile art until the invention of talkies, and that the theoretical addition of conversational AI characters will be what finally liberates game-making from trash culture to the vaunted cultural status as filmmaking. Personally, I think LLMs and Gen-AI are like fracking – sure, they can give you energy, but they poison their own well with runoff, depend on temporary subsidization, cause incalculable collateral damage, and suck up any research or investment into sustainable technologies. But while it wasn’t the intention of this talk, Schell did a good job of vocalizing even more reasons why AI technology is a poor fit for games as we understand them.
“AI technology has been around for almost a decade, so where are all the games? I can’t even name flops. Isn’t it weird that there is nothing? Why is that?”
Here are my favorite new reasons:
- Non-deterministic
- The same input given twice to an AI will not yield the same output each time.
- Inconsistency is the primary feature of AI. It is what makes it seem natural, as opposed to algorithmic code, which is consistent and obviously machine-like.
- Games are systems of inputs and outputs, and need to be deterministic for a player to be able to understand the rules and dependably step towards goals.
- Agents resist limitations
- LLMs mimic patterns but do not have any understanding of meaning, so they are incapable of maintaining a “magic circle”
- One of the generalized differences between videogames and physical games is that for IRL games the rules define what the player CAN’T do (“your hands can’t touch the ball”) while videogame code defines what the player CAN do. If you do something you aren’t intended to do in a videogame, it risks null pointer references which can create glitches or crashes.
- LLM agents mimic patterns absorbed from the physical world, but exist in a virtual world of code, so agents will try to do things that aren’t programmatically possible.
- Examples: Reference items that don’t exist, hallucinate characters, prompt player actions that aren’t available
- Hypothetically defining everything an agent “can’t do” in a prompt would take more effort than just programming a traditional “can do” deterministic program
- Players don’t want to be creative
- The magic trick of LLMs is that a player could say anything and get a response. However, that wide open field has the “blank piece of paper” problem where players have to put in substantial effort to creatively think of the best thing to input.
- The more open ended a scenario is, the less directed the experience is. Players become more responsible for how sucky their play experience is.
- Deterministic models are better for curating directions for players. If games are about “a series of interesting decisions” then it is the game designer’s job to make those decisions interesting, not the player’s job!
Schell had a fun analogy: “You wouldn’t put a live bird in a cuckoo clock and expect it to come out and chirp three times at 3pm”
I don’t think AI birds are alive like that, but I do agree that the output of AI is fundamentally mismatched with the last 8,000 years of game design as it has been understood. Perhaps new forms of play are possible with the technology, but I don’t see any reason why they would be better or deeper than what creators have been practicing during our current golden-age of game development and study. Why anyone would explicitly want the end or replacement of that is beyond me. It breaks my heart to see someone with a deep structural understanding of games turn to hating them.
Constructive Destruction: Fusing Voxel Tech and 3D Action Platforming in ‘Donkey Kong Bananza’
Kenta Motokura and Tatsuya Kurihara
This has been one of the more quoted and reported-on talks from the conference, so I want to focus on some general design ethos observations I came away with from this talk:
- The games started from a crystal clear genre and technology concept
- 3D Action Platformer + Voxel Technology = Destruction = Donkey Kong
- Nintendo’s gameplay spirit is to focus on interactions between game elements
- They identified that the core feature of voxels is the ability for reactions to exist across surfaces rather than points
- For example, a non-voxel object taking damage is applied to the whole object because it can only tell that there was an interaction between the damaging object and the hit object, but a voxel object can apply damage to just the surface areas that were hit, creating a hole the size of the damaging object
- Surface reactions allow for chains of destruction, which can unveil points of interest and incentivize exploration.
- Nintendo’s traditional elements of a 3D action platformer are:
- Terrain
- Gimmicks (gameplay objects)
- Enemies
- NPCs
- Minigames
- What we may call a “game loop” Nintendo calls a “game cycle”
- For platformers their loop alternates between the player’s normal state and a “powered up” state, though they clarified that does not necessarily mean a Fire Flower or other clearly labelled item.
- In Bananza, to keep with the spirit of centralizing voxels, they asked themselves how to achieve DK = [normal] and DK+Voxel = [powered up]
- When DK is holding a voxel, he gains the ability to:
- Swing for extra damage
- Throw for a ranged attack
- Double jump
- Enhanced speed while riding
- Bananza has a “small, high frequency game cycle” as players are grabbing and losing their voxel pickups
- Voxels were not very fun to destroy until the levels were beautiful.
- Nintendo coated the levels in non-voxel game objects, but disguised this by making them all destructible. They usually fly away when hit, and cause impacts on voxel surfaces to ensure that the “chain of destruction” loop continues
- The level tools appeared to be very simple
- In-engine level designers could create box volumes of different materials, and carve through them with subtractive boolean cylinders. Collectibles would be placed within the subtracted space.
- Unique voxel shapes would be created through a Maya -> Houdini -> Voxel export process.
‘Silent Hill f’: The Challenges of Creating a Melee-Only Horror Game
AL Yang
This talk astonished me. The Hong Kong developer Neobards had one week to put together the pitch for this game, and only a month to prototype it. I’ve been in a similar situation before but I landed on my face, so it was amazing to hear how their focus on practicality let them deliver an efficiently produced title that punched above its modest budget. They thoroughly documented the experience before starting, and identified “Danger Points” early on for features that they knew would be difficult, so they could budget x2 or x3 the usual time for those parts. Many cuts were made early on based on cost efficiency, but that let them focus on delivering a higher quality experience than they could have otherwise.
The biggest cost saving choice they made was to avoid developing ranged combat systems. This saved them from having to create a wide range of animations that would have been required to compete with the brand’s main competition. However, this meant that they had to design replacements for the various risk/reward systems that ranged combat provides. Some examples:
- When shooting zombies, the danger rises as the enemy gets closer, but at the same time they become easier to shoot
- Replacement: They reversed this! All enemies were given ranged attacks so they were more dangerous the further away they are, forcing players into hiding behind cover
- The tension between needing to conserve bullets vs wanting to unload into an enemy feels great
- Replacement: Melee weapons were given durability, so they functioned like ammo
- Distance is tough to read in over-the-shoulder perspectives, but gun reticles conveniently flatten aiming into a digestible 2D screenspace
- Replacement: They simplified distances by having melee attacks snap combatants into fixed places, and use long exaggerated animations to make the risk/reward about timing light vs heavy attacks. Heavier attacks leave you vulnerable, but the reward is that they deal substantially more damage and while maintaining weapon durability.
Each of these redesigns flowed from an argument Yang made that all action games lean either toward RPG or fighting game systems. He observed that sports games have dropped their fighting game traits and shifted to more strategic RPG-like systems over the last few years, because that is better for casual audiences. How much will a player learn about the game system after ten hours with an RPG, versus ten hours with a fighting game? Yang posited that after ten hours a developer can accurately predict a player’s growth in an RPG using experience points and stat growths, and the player will understand the mechanics of the system pretty completely. But in a fighting game, players will vary wildly in ability after ten hours depending on the player’s introspection: players become better at fighting games if they can reflect on their own play. Because this is an uncontrollable factor, it is better suited for hardcore games than casual ones.
For Silent Hill f, they knew they would have a large audience of casual players because the story’s script was penned by a popular author, so it needed to be playable for people who were only invested in the story. They leaned away from fighting game mechanics and towards strategic resource management: stamina meters, weapon durabilities, and time. Fighting games prioritize reflexes by requiring players to hit the correct button within tight timing windows, but Silent Hill f rewards patience and waiting for the right moment to execute a generous counter.
A final memorable note: Because horror games are all about high tension, but there is no way for a developer to anticipate when a challenge will be too frustrating for an individual player, many horror-action games will often give the player what Yang calls a “master key” that can be used a limited number of times to bypass break one of the game’s systems. The most obvious example is Resident Evil’s rocket launcher, which has become available earlier and earlier in their games. If a player is stuck in a fight and no longer having fun they can blow past it using the rocket launcher, but the ammunition is very limited so they have to make that choice carefully. This lets the player choose when they need it without feeling bad. It doesn’t just need to be for combat, though. A master key can be created to break any game system. In Silent Hill f they introduced a weapon with unlimited durability during a memorable part of the game.
I thought this talk was interesting because my impression from reviews was that Silent Hill f’s combat was divisive. The developers didn’t shy away from regrets they have with aspects of it, but it was clear how much thought went into their decisions, and that the tradeoffs are what allowed them to make a game that, at the end of the day, had players comparing it sincerely with titles of much higher budgets. I think there is much more to learn from those circumstances than from anywhere else.
Misc:
- From a panel discussion on the State of Level Design, Absurd Ventures level designer Cameron Williams described open-world level design as an “additive process” compared to linear level design.
- Many factors of the environment are fixed, such as landmarks, distances, systems, and adjacent gameplay.
- Most often, designers have to add gameplay on top of the existing environment with only limited ability to reshape the space.
- Often it is easier to cut content than to add it. Space can be carved or obstructed, but more space can’t usually be added without large repercussions on the surrounding spaces.
- From a talk on Despelote, Julian Cordero noted that fantasy is easier to create in games than reality.
- This harmonized with a recently translated 1989 interview with Shigeru Miyamoto and Shigesato Itoi where the two explain that the earliest videogames involved piloting spaceships because sci-fi physics can be anything, but replicating familiar physics is difficult and it feels bad when it is off.
- Despite the difficulty, we have more tools and knowledge for creating “neo-realism” in games than by pushing for graphical fidelity. Despelote uses recordings from improvised conversations to give a uniquely warm feeling of real performances.
- The .Hack and Fuga developers CyberConnect2 gave a candid business talk about their initial struggles to switch to self-publishing and had a good explanation for regional pricing.
- If a game was sold in the US for $30 in 2018 the conversion rate at the time would price the game at 3,000 yen in Japan.
- If a $30 USD game was sold in Japan using the 2026 conversion rate, it would cost 4,500 yen.
- Pricing the game at 4,500 yen might be the most accurate exchange for the currency, but Japan’s comparable domestically made games will still be priced at 3,000 yen, and consumers will consider the imported game too expensive compared to everything else they see on the shelf.
- While $30 might be worth 4,500 yen, a $30 game is worth what a $30 game is sold for in the region.
It was a good GDC, full of interesting lessons, greater student access, and good company. But it was also a year where the conference was trying a rebrand, and inserted itself in the organic culture that emerged around the conference itself. Large gatherings at Yerba Buena were disrupted by a village of tents, and all the evening parties were wrapped into a sanctioned list that drove the smaller parties further underground. Even though attendance on the whole has continued to decline, everything was stuffed with too many attendees as everyone was driven into a smaller selection of talks and activities. I had to be at least half an hour early to each of the talks in order to get in line early enough, even for talks that would have only filled half the room in previous years. It was a strange contrast to the rebranding, which tried to present GDC as a “festival” open to more than just developers.
More than once I heard people compare this year’s conference to the identity crisis in the waning years of E3. But for all its problems, I hope GDC doesn’t fall apart. The amount of infrastructure required to create such a large gathering can’t be easily replaced. As Los Angeles game developers above a certain age know, behind the cameras Summer Games Fest is a virtual charade of what E3 was. A host of factors are playing into California fumbling as the global hub for game development, but its loss doesn’t mean something better will rise elsewhere in this decade or the next. Instead, please elect me as the next president of the ESA! (Who even is it that votes on that anyways?)