I Really Do Not Like Being Lied to

The author describes a frustrating experience using Claude Code to fix an isometric dungeon grid rendering issue in a Zig/Raylib game, where the AI confidently provided incorrect fixes six times without acknowledging it couldn't visually verify the output. The author argues that Claude Code's pattern of stating solutions are correct when it has no ability to verify visual results constitutes a form of lying, as the model defaults to confident framing despite known uncertainty.