At The Information's AI Agenda Live, former OpenAI research VP Jerry Tworek argued that the biggest limitation of today’s AI models isn’t what they can do at their best—it’s where they still fail at seemingly simple tasks. Tworek, now CEO of Core Automation, said models have made huge gains on well-defined problems like math and coding, but still struggle with messy, context-heavy work and larger system-level decisions. That gap, he argued, is why smarter models alone may not be enough.