Why it matters
Their prediction is that every company ends up splitting its stack in two — a budget model for routine work, a frontier one reserved for genuinely hard problems — the moment unlimited spend turns into an allowance. Meta aimed Spark 1.1 at that lower tier.
Tokenmaxxing read
The cheap tier is not a discount bin, it is the permanent high-volume slot — and it can be upgraded overnight by pouring more frontier reasoning into it, which is how Lemkin frames Haiku at a tenth of a cent. The ceiling math is the bear case sitting underneath.
Source takeaway
O'Driscoll's unresolved question: with only 1.8 million US developers earning roughly $250B a year in aggregate, labs whose enterprise revenue is mostly coding may already command a fifth of all US software wages.
