A jet crosses a pale sun above a valley of teal mountains. Click “Enter hangar” and Island Run asks you to name a pilot. The first level, Highlands, starts with an instruction to launch from the carrier and a counter for six river gates. It is an early preview of Alkemata’s experiments with AI-assisted game development, and it brings the hardest question into focus. Once the game runs, what makes someone want to fly the course again?

Open the first Island Run preview. The 3D flight scene requires WebGL.
The parts that got faster
The clearest result from our tests is how much AI simplifies and speeds up technical development. A working prototype puts a series of decisions within reach. Wiring up a throttle, storing pilot progress and trying another scoring rule can become short iterations. Cheaper revisions let a small team put more ideas in front of players while there is still time to change direction. That extends the kind of experimentation we described in an earlier Backstage note.
The design work begins with the promise of the game. Is Island Run about the thrill of flying low, the pleasure of learning a precise route, or the scramble to recover from a bad turn? Each answer suggests a different course. A winding river can teach anticipation. A wide valley can give the player room to experiment. A late gate after a blind bend can feel unfair even when the simulation is behaving exactly as designed.
Then come the thousands of small artistic decisions. How early does the next gate stand out against the terrain? Does a successful pass deserve a short chime or a change in the engine note? How quickly should the camera settle after a bank? Those choices reinforce one another. A gate that reads clearly, a plane that answers predictably and a sound that arrives at the right instant can make a single turn satisfying. Give the same turn a busy backdrop and ambiguous feedback and the player may struggle to understand what just happened.
One useful playtest would compare two versions of the same short course with identical controls. Change only when the next gate becomes visible. Alternate which version people try first, watch where they begin to turn, and ask what they expected after a miss. Then give them the option of another run. A clean completion tells us something about legibility. The desire to go again tells us something about the game.
What cheaper production could change
Smaller, stranger games. A two-person team could spend a month on one idea that would struggle to justify a larger production. Imagine a flight game built around the tension of one approach through fog. It could end as soon as the player has learned to read the terrain by sound. A game can earn its place in someone’s evening in twenty minutes.
Games with a shared core and many local versions. A club could build its own weekend course around a landscape its members know. Designers could publish short challenges with a fixed seed so friends can compare runs. Optional assists could change gate visibility or input sensitivity while leaving the published challenge available as a common reference. These variations would need playtesting of their own, especially where they change what a player can perceive.
There will also be pressure to ship more of the same. A tool that makes a novel idea cheaper to test can make a familiar formula cheaper to reproduce. Studios that spend the time they save on art direction and watching people play may make very different games from studios that spend it on a longer release list.
Finding a game worth your evening
SteamDB’s annual series puts Steam releases at roughly 14,000 in 2023 and more than 20,000 in 2025. That growth gives a sense of the scale of discovery even before assuming any further acceleration from AI. Steam Next Fest already combines playable demos, tags, early random exposure and later personalization. Here are three ways to take those ideas further. SteamDB release figures · Steam Next Fest documentation.
Search for the experience you want. “A short flight game with forgiving controls and quick restarts” is a better request than “show me flying games.” Developers could publish a small, portable record with session length, input options, accessibility features and a demo link. Curators could add observations from play. An assistant could compare listings across stores and say which details came from the developer and which someone actually tested.
Give players a representative minute. Use a demo or a short, uncut clip that shows the recurring action, including a mistake and recovery. Let people follow curators who explain those moments and disclose paid placements. Someone who likes the rhythm of a particular reviewer’s picks can follow that taste across genres.
Reserve space for unfamiliar games. A recommendation list could set aside one slot for a lightly reviewed game that meets the player’s stated preferences. Rotate that slot among eligible demos and let players choose how far it can stray from their usual picks. Check whether it leads to games they’re glad they tried, using a brief optional question after play. Clicks alone can reward a striking thumbnail.
For Island Run, we could start with one course, one uncut clip and a plain description of what the player does. Show that to a few people who enjoy different kinds of flight games. See who opts in, where they hesitate, and what they change on a second run. The answers would help us decide which part of the game deserves the next iteration.
What concrete example would confirm or challenge this article?
Choose a reaction, write a comment, or ask your AI assistant to send a proposal for review. Comments are reviewed before publication.