A year is a long time in AI.
At last year’s Central Park AI Forum, participants considered how AI could reshape society, the economy, law, and policy. Twelve months later, AI agents are embedded in business workflows, demand for compute is surging, and lawmakers are contemplating an agentic economy.
Against this backdrop, Norm convened the annual Central Park AI Forum on September 24 at The Plaza in New York. More than 200 leaders from business, technology, law, and government gathered to discuss the practical decisions they already face.
Five themes emerged.
AI Should Expand Individual Autonomy
Forum participants returned throughout the day to a shared premise. AI capabilities will continue to advance as resources flow toward developing more powerful models, infrastructure, and applications.
Norm Founder and CEO John Nay shared how American law can help guide that progress. The country’s legal tradition balances protections to individual autonomy with limits on conduct that harms others. It reflects values honed through democratic institutions and tested through courts.
Vinod Khosla described what greater autonomy could make possible. He envisions AI operating as a great equalizer, making medical, educational, financial, and legal expertise broadly available at very low costs. People who cannot afford a tutor, lawyer, or financial adviser today could gain access to comparable, tailored forms of support.
Khosla also addressed the economic disruption that will accompany this shift. He argued for policies that protect individuals through periods of change and help them move into more fulfilling work, including care, craft, and creative pursuits.
Together, these discussions offered a practical test for AI progress. Greater capability should give people more power to shape their own lives.
AI Demand Is Running Into Physical Constraints
The rapid growth of AI is creating a shortage of compute. Mike Forman described a data center market setting record levels of demand year after year, with agentic development expected to increase compute requirements further.
Building enough capacity will be difficult. New data centers need power connections, transmission capacity, suitable land, equipment, permits, and skilled workers. A shortage in any one of these areas can delay a project, even as demand continues to rise.
Public support has also become a practical constraint. Community debates frequently center on water, electricity, employment, and the effect of new facilities on local infrastructure. The answers vary by project, which places more responsibility on developers to explain their plans and show tangible local benefits.
New York State Assembly Member and NY-12 Democratic Nominee for U.S. Congress Micah Lasher argued that government will need to bring developers and communities together. Agreements covering power, construction, taxes, and community investment will shape which projects move forward and where they are built.
America’s AI ambitions ultimately depend on its ability to turn investment into usable compute. That will require more energy and infrastructure, along with public support from the communities hosting it.
The Agentic Enterprise Requires New Ways of Working
For many enterprise use cases, current models are already capable enough to produce value. The bigger challenge involves redesigning the organization around them, connecting those models to the information, systems, and processes that run the business.
Enterprise adoption tends to progress in stages. Companies begin with personal productivity tools, then they apply AI to services such as software development and legal work, and eventually introduce agents that operate across more complex business processes.
That final stage requires process mapping. Companies need to identify where information enters a workflow, where judgment is applied, which decisions require approval, and where mistakes create risk. They can then determine which tasks an agent should perform and when a person should intervene.
Context is another constraint. Agents need secure access to documents, communications, policies, and software systems. Large companies have accumulated this information across disconnected platforms with different permissions and security requirements. For many enterprises, information architecture will determine how quickly agents can move from isolated tools into core operations.
Legal departments offer an early test of this operating model:
Salesforce President and Chief Legal Officer Sabastian Niles described the company’s work to build an AI-native and trust-first legal, compliance, and risk function. Agents can help manage large volumes of internal requests, reduce delays for employees, and give experienced lawyers more time for difficult decisions.
Morgan Stanley Chief Legal Officer and Chief Administrative Officer Eric Grossman described how AI can improve standardized legal processes and create greater continuity by preserving the knowledge of experienced lawyers when they retire or leave.
The enterprise advantage will come from embedding agents deeply in real workflows, supported by secure data, clear escalation points, and accountable professionals.
Judgment Becomes More Valuable as Intelligence Becomes Abundant
At last year’s Forum, participants discussed why wisdom would matter more than raw intelligence. This year, the conversation focused on judgment.
In a discussion with Anthropic Chief Economist Peter McCrory, Coatue Founder Philippe Laffont described judgment as distinct from expertise. AI systems can demonstrate extraordinary technical skill and still produce a poor recommendation.
McCrory pointed to Anthropic data showing that domain experts who know how to guide Claude Code are systematically more successful. As tasks become more complex, performance declines more sharply in fully automated workflows than when humans remain involved. Domain expertise helps users supply context, spot errors, and steer the model through difficult work.
Laffont argued that this perspective is developed by employees through experience and mistakes. This creates a challenge for companies considering how AI will change entry-level work. Junior employees build judgment by participating in real matters, observing experienced colleagues, making decisions, and seeing the consequences.
If organizations automate every developmental task, they may weaken the pipeline that produces future senior decision-makers. They will need new forms of apprenticeship that preserve learning while using AI to improve productivity.
Stephen Wolfram added technical rationale for why judgment becomes more important as AI systems grow more capable. Sophisticated computational systems can behave in ways that cannot be fully predicted, even when their underlying rules are known. Rules and safeguards will never anticipate every outcome. People must therefore decide when to trust a system, when to intervene, and how to respond when it behaves unexpectedly. Judgment is what makes responsible action possible under that uncertainty.
Wolfram also emphasized that law begins with human choices about the outcomes society wants. AI may translate legal language into precise symbolic representations and make rules easier to analyze, but people still have to define the values those rules express and decide how they should be applied.
Organizations will continue to need people who can frame problems, assess tradeoffs, challenge persuasive but logically weak answers, and accept responsibility for decisions.
Law Is Becoming Infrastructure for an Agentic Economy
AI agents are beginning to perform work, make decisions, and interact with institutions. Law will provide the structure that determines how those activities fit within the economy.
Delaware Secretary of State Charuni Patibanda-Sanchez and former Thomson Reuters CEO Tom Glocer discussed Delaware’s proposed framework for Agent-Managed Companies.
Under the proposal, an AI agent could manage an entity owned by a human member. The entity would operate inside a supervised regulatory sandbox with registration, capitalization, disclosure, and oversight requirements. Governance bodies could monitor participating entities, courts could resolve disputes, and officials could stop an agent’s activities when necessary.
The sandbox would allow Delaware to observe how these entities behave before establishing permanent rules. The approach draws on the state’s experience defining ownership, responsibility, liability, and recourse for new forms of business.
Norm is contributing technical expertise and helping the Delaware Department of State build the infrastructure for the proposed sandbox.
Legal services are undergoing a related shift. AI can perform a growing share of work historically completed by junior lawyers, allowing legal services providers to review more information, complete matters faster, and offer more comprehensive support.
Capturing those benefits will require business models aligned with more AI deployment. AI-native legal services can price work around outcomes, transactions, and completed matters, allowing clients to benefit as the technology improves.
Last year, Forum participants suggested that AI could usher in a “golden era” for legal and compliance. This year, we’ve started to implement that vision. AI agents are entering real workflows, legal services are being redesigned around them, and governments are considering new forms of legal identity and accountability.
In Closing
The Forum surfaced a practical agenda for the next stage of AI. America must build enough compute to support growing demand. Enterprises need to redesign workflows. Employers need to preserve the experiences through which people develop judgment. Governments and businesses need legal and economic structures suited to AI agents.
The standard running through each discussion was how AI can expand opportunity.
Norm is honored to convene the leaders working through these questions. We look forward to continuing the conversation.