Rohit Krishnan is an independent researcher and writer who publishes on AI agent coordination, market mechanisms for AI agents, AI economics, and multi-agent systems. He writes the Strange Loop Canon Substack and co-authors academic working papers. Krishnan describes himself as an "independent researcher"; his formal academic affiliation appears limited, with influence coming through Substack and working papers rather than a tenured position.
Background
Krishnan's active venues are the Strange Loop Canon Substack and academic working papers. His work treats AI agents as market participants, combining a theoretical economic framework, an empirical benchmark, and an empirical demonstration across three publications produced in a four-month span.
Publications
- *The Coasean Singularity?* (NBER chapter, 2025) — co-authored with Shahidi, Rusak, Manning, Fradkin, and Horton. A formal economic framework for AI agents and the make-or-buy boundary they shift.
- *MarketBench* (working paper, April 2026) — co-authored with Andrey Fradkin. An empirical benchmark testing whether frontier LLMs can self-assess success probability and token cost well enough to participate in market-style coordination. It uses SWE-bench Lite as its task base.
- *Why Coase needs Hayek* (Strange Loop Canon, May 2026) — an empirical companion essay running solo, hub-spoke, and market coordination on 15 hand-written tasks; market beats hub-spoke on cost by roughly 4× and ties solo on quality.
Across these pieces Krishnan advances the position that markets, rather than hierarchical orchestrators, are the appropriate coordination primitive for multi-agent AI, conditional on the agents being able to self-assess.
Positions and arguments
Krishnan and his co-authors argue that hub-spoke coordination costs roughly 4× as much as market coordination with no quality advantage (Why Coase Needs Hayek, May 2026). They identify self-assessment as the binding constraint on multi-agent market coordination: frontier LLMs cannot calibrate their own success probability or token usage well enough to bid meaningfully (MarketBench, April 2026).
Krishnan extends Friedrich Hayek's argument to AI, contending that markets work for humans because of local, private knowledge that price signals can elicit, and that for AI agents this private knowledge emerges from divergent prompt, context, memory, and tool stacks. He projects that as models specialize, markets will become necessary as a coordination mechanism (Why Coase Needs Hayek conclusion).
With Fradkin, Krishnan advances a theoretical proposition that, under standard assumptions, market allocation weakly dominates every non-market alternative for the procurement of agent labor.
Affiliations and network
Krishnan co-authors with Andrey Fradkin (Boston University; MIT IDE; currently Amazon). The co-author network for The Coasean Singularity? also includes Peyman Shahidi (MIT), Gili Rusak (Harvard), Benjamin S. Manning (MIT), and John J. Horton (MIT and NBER).
Relationships
- supports: Agent Architecture Patterns — empirical evidence about hub-spoke vs. market topologies
- supports: Principal-Agent Problem Applied to AI
- related: Agentic AI
- related: Andrey Fradkin — primary collaborator
- related: AI Coding Agents — uses SWE-bench Lite as task base for MarketBench
Confidence
Confidence is high for the publications and their content. Krishnan describes himself as an independent researcher; formal academic affiliation appears limited, and his influence comes through Substack and working papers rather than a tenured position.