Back to AI game news
AI game news

Ship or Cut makes unreliable AI agents a strategy game, tested across 49 team builds

By GameSlash

Released Nov 15 for Epic AI MegaJam, Ship or Cut is a 4-sprint tactical auto-battler. Its developer used Codex to simulate 49 valid team compositions.

Indie studio mohalkka released Ship or Cut for Windows on November 15, 2026, as an Epic AI MegaJam entry. Its Director's Cut theme is a management decision rather than movie editing: the player supervises five simulated AI team members who fight bugs over four development sprints. Those in-game agents are fictional combat units, not a claim that a live language model runs every battle.

Real Ship or Cut gameplay screenshot showing turn-based AI-team battle, agent malfunctions and build stability
Real gameplay screenshot from studio: mohalkka / itch.io · AI team members, work orders and build stability during combat.

A sprint-based strategy game about an unreliable AI team

Players recruit five agents from eight choices, covering Find, Contain, Fix and Check duties. They position them on three lanes with front and back slots; once the battle starts, members act automatically by speed. Bug types demand specific action chains, such as Find → Fix for a hidden bug or Find → Contain → Fix → Check for a major one. An agent may hallucinate a bug, fake task completion, or merely copy a teammate's last action, weakening a run.

Official Ship or Cut gameplay screenshot from Sprint 2 showing team formation, bug turn order and action log
Developer's official gameplay screenshot: mohalkka / itch.io · Sprint 2 combat with a bug sequence and action log.

Reboot the sprint or cut the agent

After a problematic sprint, the player can spend a limited Reboot credit to retry, or Cut a faulty team member and replace them with a reliable Mini of the same role. The goal is to preserve Build Stability through four sprints and ship a good game; poor results can cost the fictional director their job. The game has English and Korean UI, around 10–15 minutes per run and three possible endings, according to the developer.

The real AI production method involved 49 team-build simulations

In the production notes, mohalkka explains that Claude Code and Unreal MCP implemented its Unreal Engine 5.9 Blueprint systems. Separately, OpenAI Codex developed the Python simulation combat_v7.py, adjusted balance and surveyed 49 eligible team combinations; Codex did not operate the Unreal editor. The studio's human developer chose the mechanics, reviewed captures and playtested. Concept art was made with ChatGPT and turned into 3D models through Hyper3D Rodin Gen-2.5; these concept images are not the screenshots shown here.

The reusable lesson is to separate in-engine combat implementation from an external balance simulator, test many valid squad compositions before polishing and then compare those predictions with human playtests. The 49-combination count is the studio's report, not an independent GameSlash benchmark. This is a Windows download, not a browser game.

Download and authentic gameplay screenshots: Ship or Cut on itch.io

Production notes on Python simulations and Unreal MCP: Ship or Cut glass-box development log

View source