Multiplayer agentic engineering — Arjun Singh, Superconductor
Teams should stay model-agnostic because the best AI model can change weekly
“the incentives of the people selling you tokens aren't really aligned”
Arjun Singh of Superconductor (ex-GradeScope team) argues that engineering teams integrating agents should be model- and harness-agnostic to avoid workflow disruption as the model landscape shifts rapidly. He notes open-weight models like GLM 5.2 are now competitive and cheaper, and flags a misalignment between token-seller incentives and team interests. The talk focuses on human-in-the-loop collaboration patterns rather than agent-centric design, a framing that pushes back on the conference's dominant narrative.