LLM failure modes mirror ADHD cognitive science

Viewing AI behavior through the lens of neurodivergence offers a surprisingly pragmatic framework for designing agentic systems.

• –Both systems exhibit associative processing and confabulation, prioritizing creative pattern completion over precise sequential logic.
• –An LLM's limited context window functions identically to working memory deficits, making RAG and system prompts analogous to behavioral planners.
• –Structured environments and clear constraints act as force multipliers, dramatically reducing unfocused, rambling output in both.
• –Sustained focus on a single thread compounds quality, whereas context switching or interruptions destroy continuity entirely.