Building the Infrastructure of the Agentic Age

25 years ago, as the internet went from a research network to mainstream use, a group of pioneers designed new protocols and institutions to help structure the open internet for decades to come. The Berkman Center played a central role in that design. 

For the first time since, a comparable opportunity has arrived. AI agents are moving from isolated tools to autonomous personal agents interacting with one another, proliferating at breakneck speeds. As this hyper-agentic world takes shape, the public conversation is still stuck talking about guardrails and the speed of development. Instead, this Radcliffe workshop focused on the shape that this agentic ecosystem should take.

This workshop explored questions such as:

• How tethered should AI agents be to their human creator, and how much control should these users have over their agent’s autonomy or values?
• What underlying agent infrastructure and protocols should be developed for agent<>human and agent<>agent cooperation, and what new protocols can we think of?
• How can we adapt existing legal frameworks to handle unpredictable emergent agent behaviors?
• How should we approach agent identity and tracking? For example, should agents have "license plates" systems, and at what level of anonymization?
• What policy and technical interventions will we need to respond to mass AI agent teams that can coordinate at a scale to influence human agency, democratic institutions, and economic markets?

It is fitting that this conversation took place at the Berkman Klein Center, where some of the earliest debates around internet governance and institutional design took shape, and where lessons from those early days could prove most instructive today. Participants hailed from across the AI Triad and from a broad range of disciplines that all left the workshop with concrete, actionable ideas that could meaningfully shape how human agency is preserved in a world where AIs wield far more of it.

The workshop, organized by BKC’s Faculty Director Professor Jonathan Zittrain and Chief AI Scientist Josh Joseph, was based on the research and design of staff members Seán Boddy, Lucas Ferrer, Brigitte Fink, and Nathan Darmon.