At the 2026 Artificial Intelligence Summit, Jerry Tworek, chief executive officer of Core Automation, pointed out that the best measure of artificial intelligence progress is not its most spectacular demonstrations, but rather the seemingly simple tasks that models still cannot complete.

The former OpenAI research lead stressed that rapid technological leaps can easily lead people to overlook the fact that models have actually made no progress in certain areas, and these shortcomings are precisely what determine whether artificial intelligence can reliably handle real-world work.