Skip to content

Ai rules

Code: ai

Total rules: 20

Rules

ECO-AIM-AGENT-001 — Unbounded agent tool-call loop

An agent can repeatedly call tools without a bounded budget, convergence check, or escalation path.

  • Category: AI/ML
  • Family: AI Agents
  • Layer: ai

ECO-AIM-AI-001 — Oversized model selection

Using larger models than needed increases inference cost and emissions.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-002 — No inference batching

No batching increases per-request overhead and lowers throughput.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-003 — Re-embedding unchanged data

Recomputing embeddings for unchanged inputs wastes compute.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-004 — No prompt caching

Repeated prompts without caching waste tokens and compute.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-005 — Always-on inference endpoints

Always-on endpoints waste baseline compute when idle.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-006 — Unbounded context window usage

Excessive context increases token cost and latency.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-007 — No model quantization

Failure to quantize when appropriate wastes inference compute.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-008 — Re-training without drift detection

Training without drift checks wastes compute and introduces risk.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-009 — No evaluation before scaling model

Scaling without evaluation wastes resources and can degrade outcomes.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-010 — Overly frequent fine-tuning cycles

Frequent tuning without clear value wastes compute.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-011 — Storing all embeddings indefinitely

Embedding stores without retention grow unbounded and expensive.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-012 — Large model in low-SLA workload

Using high-cost models where latency/quality needs are modest wastes resources.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-013 — No GPU utilization monitoring

Without GPU utilization metrics, accelerator waste stays invisible.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-014 — Inefficient feature preprocessing pipelines

Preprocessing waste increases training and inference cost.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-AI-015 — No batching of vector search queries

Unbatched vector queries increase overhead and reduce throughput.

  • Category: AI/ML
  • Family: AI
  • Layer: ai

ECO-AIM-PROMPT-001 — Repeated static prompt context

Large static instructions or reference material are injected into every prompt instead of being cached, retrieved, or shortened.

  • Category: AI/ML
  • Family: Prompt Construction
  • Layer: ai

ECO-AIM-RAG-001 — Embedding regeneration without change detection

Embeddings are regenerated for unchanged content, wasting compute and increasing pipeline latency.

  • Category: AI/ML
  • Family: Retrieval-Augmented Generation
  • Layer: ai

ECO-AIM-RAG-002 — Excessive retrieval fan-out

RAG retrieval queries too many sources, chunks, or indexes before ranking, increasing latency and inference context.

  • Category: AI/ML
  • Family: Retrieval-Augmented Generation
  • Layer: ai

ECO-SUS-WATER-001 — Water-stress-blind AI inference placement

AI inference workloads are placed without considering regional water stress or cooling impact.

  • Category: Sustainability & Environmental Impact
  • Family: Water-Aware Computing
  • Layer: ai