AiD10t Games

We let AI cook. Someone still directs the kitchen.

AiD10t Games is my experiment in running an AI-driven game studio under human direction. Specialized agents contribute design, programming, testing, and documentation, while I guide the work toward a coherent game.

The studio’s public home presents its current build, an agent-log area, and project status. Cogworks, a Godot puzzle game, is the current project connecting that studio idea to playable work.

Visit the AiD10t studio · Explore the Cogworks case study

Give each discipline a clear responsibility

Cogworks’ project documentation separates direction, gameplay, Godot architecture, puzzle design, UI/UX, and QA into defined responsibilities. That makes the development process more concrete than a single open-ended request to produce a game.

Puzzle design owns teaching order, difficulty, and solution quality. QA owns acceptance criteria, regression checks, and reproducible failures. Direction owns the product vision, priorities, milestone approval, and the final decision when disciplines disagree.

The boundaries give review a purpose. A technically impressive addition still has to improve the puzzle. A level still has to be understandable and solvable. A feature still has to fit the current milestone.

Keep human judgment in the loop

My role centers on choosing what the game should become and evaluating whether the work supports it. The director’s brief for Cogworks emphasizes playable puzzle value, readable decisions, and controlled scope.

The discipline briefs turn those goals into practical checks. Puzzle work records intended and alternate solutions. QA checks placement, route failures, output order, and whether play works again after a reset. Bug reports distinguish expected behavior from observed behavior and include the steps needed to reproduce the problem.

These are familiar production habits applied to an agent-assisted process. They provide a way to judge the result beyond whether generated code runs.

A game gives the workflow something to prove

Cogworks currently has a playable 20-level campaign in its source project, with authored puzzle data and automated reference-solution validation. Its reset rules, ordered outputs, and shared simulation clock create concrete behavior to test as the game evolves.

That makes it a useful companion to the studio experiment. The Cogworks case study describes the game’s systems; AiD10t describes the organization of the work around them.

An experiment in production

AiD10t remains an experimental studio. Its public status page is a window into the project, not evidence that every agent is continuously running or that the process operates without human oversight.

The ongoing question is how well defined responsibilities, explicit review, and repeatable validation can turn agent contributions into a game worth playing. Cogworks provides a concrete place to keep testing that question.