Fable's quality makes you want to use it; its price forces you to do the math — Drew Breunig wrote last week that this marks the end of the AI industry's "free lunch." We note this is more than a tooling decision: it's the first time the AI industry has been pushed to run itself with the same operational discipline as cloud vendors.

What this is

In his post, Breunig walks through his own arc. Before Fable (a premium model from Anthropic) landed, his engineering logic was simple: make do with current tools, wait for the next generation, and trust that "a new model will arrive at the same price — or lower — and solve all the problems that were nagging you." That's what he calls the "free lunch" — generational gains in model capability automatically paper over the rough edges in your engineering.

Fable broke that loop. It's good enough to want it running every day, but expensive enough that the team has to do the math. Breunig's response was to tier the work: top-tier models like Opus and Fable for hard tasks; "good enough" models like GLM, K3, and 5.6 for everyday coding. The industry has a formal name for this approach — model routing — assigning different-priced models based on task difficulty.

Industry view

Backers see this as a sign AI is finally maturing. Tiering was the basic discipline of the cloud era — AWS doesn't bill S3 object storage and EC2 compute instances on the same logic, and AI shouldn't either. Breunig's setup, in essence, transplants cloud-architecture thinking into the era of large models.

But the pushback is sharp. Several engineers on Hacker News point out that Breunig's lessons are a "rich-person problem." For small teams and indie developers, the engineering cost of building a model routing system can easily exceed the API savings. "Are you going to hire a dedicated SRE just to decide which request goes to which model?" one commenter asked. The deeper worry: as premium-model prices stay high, the market will stratify faster — teams that can afford Fable and teams stuck on open-source will ship products with ever-widening quality gaps.

Impact on regular people

  • For enterprise IT: Model procurement is shifting from "which vendor" to "how to mix and match," and AI budget structures need to be redesigned.
  • For careers: The scarce skill for AI practitioners is shifting from "prompt tuning" to "workflow design and cost control."
  • For consumers: Consumer AI product pricing will stay flat in the near term, but enterprise AI service pricing will diverge further.