Case study / Sweeper Sinclair

Sweeper Sinclair app icon

Built with AI, End to End

AI made shipping fast. I make what ships land. How a three-person studio shipped Sweeper Sinclair with custom agent skills, MCP tooling, and human judgement steering every decision.

App Store Google Play

42K

lines of GDScript

13

custom agent skills

12

MCP command modules in the editor

1,900+

commits, three-person team

The system

A production line, not a prompt box

Reusable agent capability, governed by house rules.

  • Engineered a production system that lets a three-person studio ship a 42,000-line Godot codebase: 407 scripts, 193 scenes, and 1,900+ commits.
  • Wrote 13 custom agent skills covering combat, economy, enemy AI, UI components, performance, and persistence, turning domain knowledge into reusable agent capability.
  • Ran an MCP server inside the Godot editor with 12 command modules, so agents create scenes, edit nodes, and wire UI directly rather than generating blind code.
  • Enforced house engineering standards the agents must follow: typed GDScript, config-driven balance, and component architecture, with human design judgement steering every decision that ships.
Claude Cursor Godot MCP
Agentic workflow diagram
(in progress)

Positioning, UX, and taste for teams that build at speed, with the same agentic stack. I'm not the designer who needs the tools explained; I run them daily in production. What I add is what the stack can't generate: the judgement that makes fast products land.