Skip to content
anythingai.space
Design Patternsbeginner15 min · 10 steps

AI Anti-Patterns

The common ways AI systems go wrong, from god prompts and single-provider bets to runaway agents and vibes-based evals, and the fix for each.

CONTEXT WITH A GOD PROMPTproduct Q&A rulessupport rulesrefund rulesscheduling rulestranslation, code, reports…user's questionIntent routerSupportfocused promptRefundsfocused promptSchedulingfocused prompt
requestresponsedatatool callsignal
1 / 10
ProblemStep 1 of 10

The god prompt

Anti-pattern: one huge system prompt that tries to do everything: product Q&A, support, refunds, scheduling, translation, code, reports, running to about 5,000 tokens. Instructions crowd out user content, conflict with each other, can't be tuned for every case, and any edit affects everything.

Fix: classify intent and route to specialized handlers, each with a short, focused prompt. The router can be keyword-based or an LLM for harder cases.

InsightThe general rule: break complexity into specialized components instead of one monolith.
What changed
  • +Intent router
  • +Three focused handlers
  • −One 5,000-token prompt

Lesson 2 of 2 in Design Patterns