{
  "id": 9336399,
  "title": "McKinsey: Cheaper AI models, bigger AI bills",
  "url": "https://urgent.news/2026/09/23/mckinsey-cheaper-ai-models-bigger-ai-bills",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-23T12:12:23.000Z",
  "source": {
    "name": "Fortune",
    "slug": "fortune",
    "url": "https://fortune.com/2026/09/23/mckinsey-cheaper-ai-models-bigger-ai-bills-cfo/"
  },
  "original_language": "en",
  "account": "Good morning. Intelligence is becoming increasingly affordable, yet enterprises continue to spend more on AI systems. McKinsey senior partners Tanguy Catlin and Lari Hämäläinen discussed this paradox during a virtual session titled \"Improving the Economics of Agentic AI.\" According to Hämäläinen, the cost of GPT-4, launched in early 2023, has dropped drastically—from $60 per million output tokens to well below a dollar. However, as models become cheaper, businesses are demanding more from them, leading to soaring enterprise AI bills. These models are now being leveraged by autonomous agents that can perform extensive reasoning and work, generating more content than a human developer would typically handle. The economics of agentic AI remains unclear, as AI agents can take different paths to the same result, making costs highly variable. Some tasks can end up costing up to 30 times more on one run compared to another. The key to managing these costs lies in redesigning systems to avoid unnecessary token usage. Instead of measuring agents by cost per token, leaders should evaluate them based on task-level metrics: the cost of executing a task, the agent's success rate, and the human time required to verify its work. An agent can be economically viable if the time needed to verify its output is a small fraction of the time a human would take to complete the task from scratch. Optimizing workflows and managing AI spending effectively requires visibility into cost drivers, optimizing the match between model complexity and task requirements, and implementing greater sourcing discipline. Catlin emphasized that cost-cutting should not be indiscriminate, and companies should focus on areas where AI spending delivers the highest returns.",
  "summary": "McKinsey experts explain why per-token savings aren't reaching the bottom line, and what leaders should track instead of cost per token.",
  "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."
}