{
  "id": 11167585,
  "title": "Don’t follow the herd on AI cost optimization – control compute this way",
  "url": "https://urgent.news/2026/10/01/dont-follow-the-herd-on-ai-cost-optimization-control-compute-this-way",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-01T10:12:55.000Z",
  "source": {
    "name": "TechRadar",
    "slug": "techradar",
    "url": "https://www.techradar.com/pro/dont-follow-the-herd-on-ai-cost-optimization-control-compute-this-way"
  },
  "original_language": "en",
  "account": "As AI agents move from experimentation to scale, companies are facing higher AI bills. Uber's 2026 AI budget burned through in just four months, prompting interest in model routing and open-source models to cut costs. While switching models can optimize spending, a narrow focus on token usage alone won't be enough. Organizations need to also adopt disciplined methods to determine when to use LLMs and when other data processing techniques are more efficient. Running LLMs as an analytics engine can lead to unnecessary token consumption since they constantly rebuild context to answer questions. Instead, LLMs should leverage existing organizational knowledge rather than continuously reinventing the wheel. Many problems are repetitive, so models shouldn't need to start from scratch each time. The most efficient route to AI cost optimization involves integrating LLMs with a business logic layer that contains pre-built workflows, analytics, and compliance rules. This layer can serve as a connective tissue between cloud data platforms and LLMs, avoiding redundant compute costs. By combining trusted workflows, business logic, and targeted use cases, organizations can maximize the value of AI investments and avoid costly mistakes.",
  "summary": "AI cost optimization starts by reducing unnecessary compute, not chasing lower-cost models.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}