Signal Blitz started as a simple idea: turn UX instincts into a fast, playable signal. From there, the goal was speed without sloppiness.
About 80 to 90 percent of the code was generated with AI, using Cursor as the primary build environment.
- Plan mode handled the big pieces like game structure, scoring logic, and routing.
- Agent mode handled smaller tweaks like copy changes, naming updates, and quick refactors.
Design & Build Process
- We started with rough code in Cursor, shaped the screens with visual design in Figma, then iterated back and forth until the visuals and behavior aligned.
- AI did most of the heavy lifting, but it was never left unsupervised. It handled the core functionality well: game flow, scoring calculations, callbacks, and performance optimizations.
- The scoring logic is a pure function that measures distance, hotspots, and timing. That math required tighter prompts and human guidance, but once aligned, AI supported it effectively.
Where AI fell short
Where AI fell short was visual judgment and guardrails. Styling, layout polish, and on-screen feel were handled manually. AI also tends to overuse React state and effects, which can introduce performance issues, so those areas were reviewed and corrected by hand. Security was another clear boundary. Admin access and password checks were moved server-side to avoid front-end exposure.
The result is a fully functioning game built in hours, not days. AI accelerated the build, but human judgment shaped the experience, enforced constraints, and owned the final decisions.
Lessons from building the game
AI is powerful at execution and scale, but quality still depends on knowing when to step in, refine, and take responsibility.