{
  "id": 5532515,
  "title": "AI Companies Should Stop Charging for Tokens and Start Charging for Outcomes",
  "url": "https://urgent.news/2026/09/04/ai-companies-should-stop-charging-for-tokens-and-start-charging-for",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-04T01:59:44.000Z",
  "source": {
    "name": "HackerNoon",
    "slug": "hackernoon",
    "url": "https://hackernoon.com/ai-companies-should-stop-charging-for-tokens-and-start-charging-for-outcomes?source=rss"
  },
  "original_language": "en",
  "account": "AI companies are reconsidering their pricing models, suggesting a shift from charging per token to charging based on outcomes. The reasoning is rooted in mathematical logic: a model that uses more tokens but delivers superior results may be more economically advantageous than a cheaper alternative. While charging per token is a common practice for foundation labs, it may not be the best approach for application developers who sell resolved tickets or shipped features. Uber's experience illustrates this point - their internal leaderboard rewarding teams for total AI usage led to unsustainable spending within a short period. Companies must instead focus on cost per successful outcome, not just cost per token. This includes considering factors like retries, supervision, and manual fixes. The real question for AI companies is no longer cost per token, but cost per successful outcome, such as cost per resolved ticket, per approved loan, or per hub fix. In essence, tokens are akin to raw materials and manufacturing costs, while outcomes are the products customers pay for.",
  "summary": "Customers don't buy tokens, they buy results. Why AI companies need to make outcomes the foundation of their unit economics.",
  "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."
}