Meta AI research chief: economic-value agents are AI's next frontier
Synopsis
Meta's new AI research VP Dawn Song, a UC Berkeley professor and Virtue AI co-founder, says AI agents that deliver measurable economic value — not human replacement — are the field's next frontier, as her team joins Meta's Superintelligence Labs.
Key Takeaways
Dawn Song joined Meta Platforms as vice-president of AI research at Meta's Superintelligence Labs , announced on Friday, 27 June 2026 .
She brought 'many members of the Virtue AI team' with her, according to her post on X .
Song stated the goal of AI agents is to provide 'economic value' and help humans work more effectively — not to replace them.
UC Berkeley 's RDI centre launched the Agents' Last Exam (ALE) benchmark this month, covering more than 1,500 tasks across 55 industries.
Song is a co-director of UC Berkeley 's Centre for Responsible, Decentralised Intelligence (RDI) and co-founder of enterprise AI safety start-up Virtue AI .
Dawn Song, Meta Platforms' newly appointed vice-president of AI research, says the next frontier of artificial intelligence is agents capable of performing 'economically valuable' work across a wide range of real-world domains. She made the remarks on the sidelines of the World Economic Forum in Dalian — also known as Summer Davos — days before formally joining the company.
The vision: augmentation, not replacement
Song was clear that the objective is not to displace human workers. 'The goal is not to replace humans,' she said. 'But we want these AI agents to be more effective in these important real-world domains and help humans do this work better and provide more economic value,' she added. The framing positions Meta's AI research agenda squarely around measurable productivity gains rather than speculative general intelligence.Who is Dawn Song?
Song is a professor of computer science at the University of California, Berkeley and co-director of its Centre for Responsible, Decentralised Intelligence (RDI). She is a widely recognised Chinese-American scientist specialising in AI security, and is also co-founder of enterprise AI safety start-up Virtue AI. On Friday, 27 June 2026, she announced on X that she, along with 'many members of the Virtue AI team', was joining Meta's Superintelligence Labs to help shape the company's AI safety and security efforts.New benchmark targets real-world AI performance
UC Berkeley's RDI centre introduced a new evaluation tool this month called Agents' Last Exam (ALE), designed to test how well AI agents handle more than 1,500 real-world, economically valuable tasks spanning 55 industries. The benchmark signals a broader shift in how the research community is measuring AI progress — moving away from abstract reasoning tests toward tasks with tangible commercial relevance.Why it matters
The hire represents a significant talent acquisition for Meta at a moment when the race to build capable, deployable AI agents is intensifying across the industry. Rivals including OpenAI, Anthropic, ByteDance, and Alibaba Group Holding are all accelerating their own agent-focused research and product pipelines. Bringing in a leading voice on AI safety and security — alongside her team from Virtue AI — suggests Meta is betting that trustworthy, verifiable agents will be a key differentiator in enterprise adoption.What's next
Song's dual mandate — advancing AI capability while reinforcing safety and security — will be closely watched as Meta pushes deeper into agentic AI. The ALE benchmark's industry-wide adoption could become a standard measure for evaluating agent readiness, and Song's influence on both the academic and commercial sides of the field makes her a pivotal figure in shaping that standard.Point of View
Mirroring the acqui-hire playbook that defined the cloud era. What mainstream coverage underplays is the strategic pairing: Song's safety and security credentials give Meta political cover as regulators in the EU and US tighten scrutiny of agentic AI systems. The 'economic value' framing is also deliberate — it shifts the conversation from existential AI risk to measurable ROI, the language that enterprise buyers and sceptical policymakers both respond to. With rivals like Anthropic and OpenAI racing to place agents inside corporate workflows, Meta's bet is that trustworthy, auditable agents will win the enterprise market even if they trail on raw capability benchmarks.
NationPress
29 Jun 2026
Frequently Asked Questions
Who is Dawn Song and why is she joining Meta?
Dawn Song is a professor of computer science at UC Berkeley, co-director of its Centre for Responsible, Decentralised Intelligence, and co-founder of enterprise AI safety start-up Virtue AI. She joined Meta Platforms as vice-president of AI research at Meta's Superintelligence Labs on 27 June 2026, bringing many members of the Virtue AI team with her to focus on AI safety and security.
What did Dawn Song say about AI agents at the World Economic Forum in Dalian?
Speaking on the sidelines of the World Economic Forum in Dalian (Summer Davos), Song said the next frontier of AI is agents that perform 'economically valuable' work across real-world domains. She stressed the goal is not to replace humans but to make AI agents 'more effective in these important real-world domains and help humans do this work better.'
What is the Agents' Last Exam (ALE) benchmark?
Agents' Last Exam (ALE) is a new benchmark introduced by UC Berkeley's RDI centre in June 2026. It tests how well AI agents perform across more than 1,500 real-world, economically valuable tasks spanning 55 industries, shifting evaluation away from abstract reasoning toward commercially relevant performance.
How does Meta's AI agent strategy compare to rivals like OpenAI and Anthropic?
Meta is competing directly with OpenAI, Anthropic, ByteDance, and Alibaba Group Holding in the race to build deployable AI agents. By hiring a leading AI safety researcher alongside her team, Meta is signalling that it views trustworthy, verifiable agents — not just raw capability — as the key to enterprise adoption.
What is Meta's Superintelligence Labs?
Meta's Superintelligence Labs is the division within Meta Platforms dedicated to advanced AI research, including safety and security. Dawn Song and members of the Virtue AI team joined the lab to help shape Meta's approach to building and governing next-generation AI systems.
