Research

Published research on AI infrastructure economics and AI adoption in organizations.

Reflexive Demand in the AI Infrastructure Boom: Vendor Financing, Backlogs, & the CapEx-Cash Flow Imbalance (2022–2025)

Posted December 2025

SSRN, recognized as a Recent Top Paper (~1,461 reads) · 20 pages

Examines whether the post-2022 AI infrastructure expansion represents genuine demand or a reflexive, vendor-financed cycle. Documents three interlinked dynamics: a sharp rise in capital intensity across hyperscalers, exploding long-dated backlogs that remain largely unmonetized, and heavy reliance on vendor financing and abundant credit. Argues the resulting feedback loop (belief, financing, backlog, valuation, further financing) resembles the dot-com capex bubble with modern features like AI-specific vendor financing across Nvidia, Oracle, Microsoft, Amazon, and Google.

AI InfrastructureVendor FinancingCapital ExpenditureReflexivityFinancial Cycles
Read on SSRN →DOI

The Adoption and Human Systems Layer: AI Agents as Organizational Members and the Implications for Organizational Behavior

Posted April 2026

SSRN · New York University · 28 pages

Examines the organizational behavior conditions required for AI agents to function as genuine participants in workflows rather than underutilized technology. Presents three pillars of the adoption and human systems layer (change management addressing professional identity threat, training that develops collaborative competencies, and deliberate sociotechnical workflow redesign), plus a failure modes framework, an operational governance model, and a phased 90-day implementation roadmap. Central claim: AI agent adoption is not a technology problem with a behavioral component; it is a behavioral problem with a technology component.

AI AgentsOrganizational BehaviorChange ManagementHuman-AI CollaborationSociotechnical Systems
Read on SSRN →DOI

Active & Independent Research

A Structured Methodology for AI-Native Entrepreneurship

In Progress · 2026

Berkley Center for Entrepreneurship, NYU Stern

Original research defining the first ordered, structured, empirically grounded methodology for building AI-native startups: a codified method for the AI-native era, analogous to Disciplined Entrepreneurship or Lean Startup, which predate AI-native building. Built from deep case studies of ~12 AI-native startups (Cursor, Cognition, Perplexity, Replit, and others) drawn from a landscape catalog of ~50 companies, each profiled on a fixed template covering how they build, what they sell, validation approach, scaling levers, and failure modes. The methodology feeds an AI-powered founder-advising tool for NYU founders, gated on validation of the framework itself.

AI-Native StartupsEntrepreneurship MethodologyCase-Study ResearchFounder Tools

ADHD-Attention LLM: Can Loosening Attention Make Models More Creative?

Completed · 2026

Independent Study · Mechanistic Interpretability

Tested three attention interventions (attention dropout, pre-softmax temperature flattening, and head dropout) on Qwen 2.5 to probe whether loosening attention increases LLM creativity. The result is a clean negative finding with mechanistic failure analysis: an honest null result rather than a manufactured positive, written up publicly as "I Tried to Give an LLM ADHD."

Mechanistic InterpretabilityAttention MechanismsLLM InternalsNegative Results

Ongoing Research

  • •Energy constraints and grid bottlenecks shaping AI data center strategy and platform scalability