September 9, 2026
Haste Has an Owner
Five model releases, four labs, three days. The cadence bills the teams that integrate: two to three engineer-weeks per migration.
Anthropic updated Fable and Mythos on Tuesday. The next day, Meta and Google responded; the day after, OpenAI released GPT-6 Astra. Five announcements, four labs, three days. It was no coincidence. According to Noah Faro of the startup Farsight, the companies compete for cloud capacity from the same suppliers and read availability as a clue to their rivals' plans.
According to Trace Cohen, an investor who follows these switches closely, his portfolio companies spent two to three engineer-weeks per migration.
Switching models at every release becomes recurring engineering debt with no proportional return. The team pins the version in production and only migrates when the measured gain exceeds the cost of the switch, never when the vendor announces new releases. That criterion needs to be documented before the updates arrive, because deciding on hype only generates more costs.
Sources: CNBC · Value Add Pulse (Trace Cohen)