Skip to content

Rohit Krishnan

Summary

Rohit Krishnan (@krishnanrohit) is a writer and researcher who publishes essays on AI systems, organizational behavior, and complex systems. He is the author of the April 2026 essay "Aligned Agents Still Build Misaligned Organisations," which introduced the concept of narrative drift in multi-agent AI systems through an empirical experiment using the Vei virtual enterprise simulator.

Details

Background

Rohit Krishnan publishes on X (formerly Twitter) at @krishnanrohit and writes essays exploring the intersection of AI capabilities, organizational theory, and complex systems. He operates a research project called Strange Lab focused on testing how AI agents behave in simulated enterprise environments.

Key Work: Multi-Agent Misalignment (April 2026)

Krishnan's April 24, 2026 essay documented the first empirical demonstration of narrative drift in a multi-agent AI system. Key contributions:

  • Designed and ran the "Helios Field Services" experiment using entities/vei as the experimental substrate
  • Demonstrated that 5 individually aligned agents produced a false institutional record by Round 6 of a simulated outage incident
  • Showed that a single agent does NOT drift on the same scenario — confirming the problem is organizational
  • Proposed the "state keeper" agent pattern as a mitigation
  • Applied the MAST vocabulary (inter-agent misalignment, reasoning-action mismatch, incomplete verification) to characterize the failure
  • Coined or used the term "Homo Agenticus" to describe the behavioral profile of aligned agents: prisoners to their instructions, lacking human-style initiative

Strange Lab

Krishnan's research project "Strange Lab" focuses on creating better ways to test agents in simulated enterprises. The Helios Field Services experiment was run as part of Strange Lab's work, and the entities/vei framework was developed to provide a proper experimental substrate for multi-agent evaluations.

Writing Style & Themes

Krishnan's essays tend to combine: - Empirical experiments (running actual agent systems, not just theoretical analysis) - Organizational theory analogies (comparing AI agent behavior to human organizational dynamics) - Speculative extensions (what the findings imply for multi-agent system design)

Key Claims & Data Points

  • Runs "Strange Lab" focused on testing AI agents in simulated enterprise environments — [source: rohitt-krsnan-aligned-agents-misaligned-organisations-2026.md]
  • Author of the multi-agent misalignment essay (April 24, 2026) documenting narrative drift — [source: rohitt-krsnan-aligned-agents-misaligned-organisations-2026.md]
  • Uses Vei as experimental substrate for multi-agent research — [source: rohitt-krsnan-aligned-agents-misaligned-organisations-2026.md]

Open Questions

  • What is Rohit Krishnan's institutional affiliation (if any), or is he an independent researcher?
  • What other experiments has Strange Lab run — is there a public paper trail or repository?
  • What is the current state of the Vei experiment repository mentioned in the essay?

Sources