Meta's New VP of AI Research Dawn Song: AI Agents Are the Next Real-World Milestone
Taylor Wilson
UC Berkeley professor Dawn Song joins Meta's Superintelligence Lab as VP of AI research, calling AI agents that perform economically valuable tasks the industry's next major milestone — a hire that will shape how Meta balances AI safety with commercial deployment.
Who is Dawn Song, and why does Meta want her?
Song is a computer science professor at UC Berkeley and co-director of the university's Center for Responsible Decentralized Intelligence (RDI), with a global reputation in AI safety and security.
She co-founded enterprise AI safety startup Virtue AI and announced she will bring "many members" of that team to Meta.
This means → Meta is not just hiring one academic star — it is importing an entire team, aiming to land AI safety capability fast.
What is the "next milestone" she is talking about?
At the Dalian Summer Davos forum, Song stated that AI's next major real-world milestone is AI agents capable of performing "economically valuable" tasks across a wide range of real scenarios — autonomous programs that do actual work, not just chat.
She emphasized: "The goal is not to replace humans, but to make AI agents more efficient in important real-world domains, helping people do their jobs better."
In plain terms = today's AI mostly answers when you ask; the next step is handing it a task and having it handle the whole thing end-to-end — and the task has to generate real economic return.
Does her academic work back this up?
Song's Berkeley RDI center launched a new benchmark this month called "Agents' Last Exam (ALE)", evaluating AI agents across 55 industries and over 1,500 real-world tasks.
This reflects that she is not just pointing a direction — she is already building quantifiable standards to measure how close AI agents are to doing real work.
This means → what she brings to Meta is not just a safety philosophy but a ready-made evaluation framework that can directly test how well Meta's own agents perform.
What does this mean for Meta and the industry?
Meta placed Song inside its Superintelligence Lab, signaling that AI safety is being elevated from a compliance sidecar to a core R&D function.
The key question: whether academic AI safety research can translate effectively into Meta's large-scale commercial deployment — that is the real test of this hire.
In plain terms = academic safety research can afford to move slowly, but Meta needs to ship agents to billions of users, which means safety has to run as fast as the product — whether Song can hit both accelerators at once is the biggest unknown.
Content is for reference only, not financial advice.