Closed-source pathfinding, open-source harvesting: the division of labor in enterprise AI is becoming increasingly clear.

CoinWorld News reports that on-chain analyst “AI Auntie” pointed out that the open-source share of total enterprise large-model spending has fallen to 11%. This decline is attributed to most enterprises still being in the early stage of exploring AI use cases that haven’t yet taken shape, so they default to closed-source models. However, he emphasized that once application scenarios mature, open-source models will take over production environments with extremely low latency and advantages in deep fine-tuning. In Decagon’s own production environment, 90% of call volume has already been switched to open-source weight models. In the future, enterprise AI will follow a division-of-labor pattern: top closed-source labs will continue to lead exploration and discovery in new fields, while open-source weight models will increasingly take over the actual production of mature businesses.