OpenAI Frontier Review

Frontier is the first serious attempt to make agents an operating layer rather than a feature. It connects the systems a company already runs, the ticketing tool, the warehouse, the internal applications nobody has opened in years, and gives agents a shared picture of the business to work from. The management model is the interesting part. Agents are onboarded, scoped, given permissions and judged on their output, which is closer to hiring than to installing, and the platform deliberately supervises agents built on rival models rather than locking the company into one supplier. Fortune read the launch as a threat to the software industry itself, on the theory that if agents do the work then the applications underneath become plumbing, and seven months of customer stories have not disproved it.

Buying it is another matter entirely. Availability is still a limited set of enterprises arranged through a sales conversation, seven months after the announcement, so most companies weighing agent platforms this year cannot even get a trial. There is no published price at all, and OpenAI would not discuss what a contract costs when it launched, which makes comparison impossible and leaves everyone below the enterprise tier guessing. Deployment is a program rather than a signup, with data connections, permissions, evaluation harnesses and a team to own them. The idea earns the top grade on the first axis. Everything about getting your hands on it earns what follows.