OpenAI's Joseph Kim: agents ready for trading by 2027

Kim's keynote framed autonomy in finance as a product design question rather than a question of how many restrictions to impose. He said the industry's safe use of agents depends on where controls and human judgment sit, and that a model's growing capability is what pushes it past the "threshold of usefulness." He added that rising capability also brings better traceability, which matters because a number's origin and context must be checkable.

His 2027 prediction followed from that view: as models clear the threshold, analysts can hand over more of the work. For OpenAI, the interface itself is the lever — Kim said deciding what a user must control, review or approve is design work that can start now, which is why he described review as a natural part of the ChatGPT for Financial Services experience.

The wider conference context was that frontier labs and financial firms are separately working out where human sign-off belongs. Kim's specific claim sits at the more aggressive end of that discussion, since he named trading, not just drafting, as something agents will be good enough to do.