Systems Thinking & Complexity Theory for AI Architects

Nimisha Asthagiri from Thoughtworks warns against churning out AI PoCs without a strategy for unintended consequences, using systems thinking (Cynefin, Iceberg) and complexity theory to tame runaway feedback loops that fuel issues like algorithmic addiction, burnout, or ethical drift. She breaks down causal flow diagrams, highlights new risks from frontier models and fake alignment, and walks through a meeting-scheduler agent example to show how these loops emerge—and how to spot and govern them early.

On the design side, she defines what makes an “agent” distinct from microservices, explores multi-agent patterns (RAG, chain-of-thought, reflection), and teases apart orchestration vs. decentralization trade-offs. Finally, she rounds out the talk with practical tips on explainability tools (LIME, SHAP), human-in-the-loop guardrails, governance agents, and the ethics of nudging behavior vs. genuinely solving problems.

Watch on YouTube