Meta AI research vice-president Dawn Song says the next phase of artificial intelligence will focus on “AI agents” that can carry out work with “economic value” in real-world settings. Speaking to the South China Morning Post during the World Economic Forum in Dalian, also known as “Summer Davos,” Song frames the goal as complementing human effort rather than replacing people. She says the priority is making agents more effective at tasks that matter to businesses and society, across a range of practical domains.
Both outlets report that Song links this direction to broader concerns about how well current AI benchmarks translate into real, measurable outcomes. She argues that many existing evaluations may not reflect what companies need to demonstrate return on investment (ROI). By emphasizing economically valuable performance, Song positions Meta’s research focus toward agents that can be deployed in operational environments and assessed by tangible impact.
Her comments come ahead of her joining Meta, and they outline a shift from model-centric progress toward agent-based systems designed to produce usable results.
