Kahneman's core finding is that fast, confident judgment is the default — and it's wrong in predictable ways. If your agents are making dozens of decisions a minute, you've just industrialized the failure modes he spent a career cataloging.
Daniel Kahneman spent decades mapping the gap between how we think we reason and how we actually do it. The central finding in Thinking, Fast and Slow is that intuitive judgment — pattern-matching, low-effort, immediate — dominates human cognition far more than deliberate analysis does. That wouldn't be so bad if fast thinking were merely imprecise. The problem is that it's biased in consistent, exploitable directions: anchored to irrelevant first numbers, overconfident about thin evidence, and systematically optimistic about timelines and costs. Slow thinking corrects these errors, but it's lazy and gets skipped.
Your agents are pure System 1 at scale. They pattern-match, they generate confident outputs, and they never feel the friction that occasionally nudges a human toward double-checking. Anchoring means an early bad retrieval poisons a whole reasoning chain. The planning fallacy means your agent's time and cost estimates will be cheerfully wrong. Overconfidence means it won't flag what it doesn't know. Kahneman's insight translates directly into a design obligation: System 2 doesn't emerge on its own, so founders have to build it in — structured challenges, explicit uncertainty thresholds, human review gates — or they're just shipping bias at throughput.
- Treat agent confidence scores as anchors to interrogate, not verdicts to trust
- design explicit slow-down triggers for high-stakes decisions rather than hoping the model self-corrects
- your planning estimates from any agentic pipeline need a systematic upward adjustment before they touch a deadline or a budget.
