Building AI solutions at companies requires the buy-in of the organization’s domain experts, who will need to commit time and energy to figuring out how this technology could be used to reconfigure their workflows. The added work of this collective gen AI experimentation can come with a hidden cost that, if left unchecked, can break ambitious AI initiatives. A two-year field study of gen AI innovation at two organizations showed one important factor: scaffolds, or structures to support collective gen AI experimentation, which made it feasible for domain experts to stick with the innovation process. The study further identified three types of collective gen AI experimentation and the scaffolds needed to support them.