The latest AI boom is still centered on one question: how can companies turn models' raw intelligence into measurable business value? According to Sina Finance, OpenAI's unreleased new model recently broke through hundreds of major math and computer science problems, while reports said the company was discussing financing and preparing a $30 billion IPO.

At a conference in San Francisco's waterfront Midway venue, Modal gave the keynote to Scott Wu, who said the future would be filled with "virtual employees" focused on outcomes rather than individual tasks. He said the industry's progress would depend on enough data centers, memory, and sandbox environments.

Wu pointed to AIME as the turning point, saying advanced AI models had become strong enough to outperform human contestants in the high-level math competition he once entered. Diogo Almeida, a former OpenAI researcher and now head of TypeSafe AI, asked: "Where did automation go?"

Almeida said today's large language models are optimized with reinforcement learning from human feedback and work well as assistants, but cannot complete automation tasks without supervision. Anthropic's Claude Code product lead Cat Wu said the company is focused on identifying repetitive tasks, while other software leaders said many automation efforts still stop at partial completion and require humans to finish the last 20%.