On the widening gap between what the labs ship and what enterprises put into production. Spring 2026.
A frontier model drops roughly every six weeks now. Coverage focuses on benchmarks, capability deltas, and the question of which lab is briefly ahead. In the enterprises I work with most closely, the conversation is on a different schedule. The model that just shipped is not the model that will be in production this year. It may not be the model that is in production three years from now.
01 /
Enterprise procurement runs on cycles measured in quarters. Security review is its own track. Compliance against frameworks such as SOC 2 Type II, HIPAA, FedRAMP, or DORA can take longer than the operational life of the model being reviewed. By the time a frontier model has been approved for a regulated workflow, two generations of successors have been released. This is not failure. It is what regulated software has always done. The frontier did not used to move this fast.
02 /
Two clocks are now running. The labs work on a release rhythm closer to consumer hardware. Enterprises work on a deployment rhythm closer to building codes. Both rhythms are correct for what they are doing. Neither is wrong about the other. The result is that the headlines and the actual deployed footprint of AI in regulated industries describe almost different worlds.
03 /
Most of the work in enterprise AI for the rest of this decade will not be at the frontier. It will be at the floor: the version that passed legal review and survived audit, the integration the security team finally approved. This is less interesting to write about and more interesting to do. The companies that figure out how to operate close to the frontier while shipping at the floor will define the next category, and the category does not yet have a name.
The race the press is covering is the race for the next state of the art. The race the operators are running is the race to move what the lab released a year ago into the room where the work actually happens. Both races matter. They are not the same race, and the prize at the end is not the same prize.