Fable Killed the Free Lunch, and That's the Point
Drew Breunig wrote the developer-side companion to the FT's Fable story, and it is the sharper of the two because it names what actually changed. His frame is Herb Sutter's old free lunch. For years, single-threaded CPUs just got faster, so you never had to think about how your code was structured. Then that stopped, and suddenly everyone had to care about parallelism and architecture. Fable, he argues, is that moment for AI coding.
Before, it felt silly to spend too long tuning your harness or your context strategy, because the next model would just be smarter and paper over your sloppiness. Now the best model costs enough that defaulting to it is wasteful. GLM-5.2, he notes, runs about a ninth of Fable's price and a fifth of Opus 5's, and for whole classes of tasks it is close enough. So the question every serious builder now asks is the optimizer's question: what work went where.
That is a real shift in how the craft feels. The skill is no longer just picking the smartest model. It is knowing which slice of a job needs the smart one and giving everything else strong context and a cheap model. Breunig's quiet observation is that future price cuts will hit every tier at once, so this cost-driven specialization does not go away when things get cheaper, it just moves down the curve. The free lunch is over. What replaces it is engineering, which is arguably where it should have been all along. Read it at dbreunig.com.
Related on clauday: the numbers behind it, Anthropic's Best Model Is 11% of Its Own Bill β https://clauday.com/article/22ee7969-a690-43c9-b546-14f194d19b9d
← Back to all articles
Before, it felt silly to spend too long tuning your harness or your context strategy, because the next model would just be smarter and paper over your sloppiness. Now the best model costs enough that defaulting to it is wasteful. GLM-5.2, he notes, runs about a ninth of Fable's price and a fifth of Opus 5's, and for whole classes of tasks it is close enough. So the question every serious builder now asks is the optimizer's question: what work went where.
That is a real shift in how the craft feels. The skill is no longer just picking the smartest model. It is knowing which slice of a job needs the smart one and giving everything else strong context and a cheap model. Breunig's quiet observation is that future price cuts will hit every tier at once, so this cost-driven specialization does not go away when things get cheaper, it just moves down the curve. The free lunch is over. What replaces it is engineering, which is arguably where it should have been all along. Read it at dbreunig.com.
Related on clauday: the numbers behind it, Anthropic's Best Model Is 11% of Its Own Bill β https://clauday.com/article/22ee7969-a690-43c9-b546-14f194d19b9d
Comments