Spiraling

The AI gets stuck in a loop, flip-flopping between wrong answers or digging deeper into the wrong direction.

Symptoms

• AI flip-flops on diagnoses (“the issue is X… wait, it’s Y… no, it’s actually X again”)
• Spiraling deeper into wrong directions
• Amplifying its own previous noise instead of course-correcting
• Spawns multiple processes (frontend, backend, tests), loses track, tries to kill them, spawns duplicates, repeats

Theory

Each AI response becomes part of the context for the next response. When the AI makes a wrong guess, that guess influences subsequent thinking. Instead of stepping back, it builds on the mistake - or oscillates between competing wrong hypotheses.

Mitigation

Interrupt the loop. Either:

• Start a fresh context
• Provide a clear correction that breaks the pattern
• Ask the AI to step back and reconsider from scratch