System 1 Runs Your Stack by Default
Daniel Kahneman's Thinking, Fast and Slow describes two modes of cognition: a fast, automatic system that pattern-matches without effort, and a slow, deliberate system that audits those answers at real cost. Agents behave like pure System 1 — they produce fluent, confident outputs at speed. The danger is not that they are stupid; it is that they are convincing. A founder reading a polished agent-drafted proposal activates their own System 1 and nods along, because the output looks like good reasoning even when it is not.
Anchoring Happens Inside the Workflow Too
Kahneman shows that the first number or framing a person encounters pulls every subsequent estimate toward it, often invisibly. In an agentic workflow, the first draft, the first cost estimate, or the first diagnosis an agent produces functions as an anchor for every human review that follows. Founders who design approval layers need to account for this. Asking a human to check an agent's output is not the same as asking them to form an independent judgment — it is mostly asking them to confirm or slightly adjust what the agent already said.
The Planning Fallacy Is a Workflow Design Problem
The planning fallacy, another of Kahneman's core findings, describes the human tendency to build schedules and budgets around best-case scenarios while ignoring the base rate of how similar work actually goes. Agentic companies are acutely exposed here because the speed at which agents produce outputs creates an illusion of compounding progress. Founders mistake throughput for completion. The antidote is the same one Kahneman recommends for human planners: look at the outside view, build evals that surface failure rates honestly, and size approval layers to the actual stakes rather than the hoped-for stakes.
Slow Thinking Is the Human's Competitive Job
The canon position on human judgment is precise: humans should own goals, values, tradeoffs, and any decision they would need to defend in a room with a customer. That is exactly the domain where System 2 deliberation earns its cost. Founders building AI-native companies should treat slow, effortful review not as a bottleneck but as the work itself at the narrow, consequential tip of the pyramid. Kahneman's research makes the case plainly — overconfidence is the default, anchoring is automatic, and the planning fallacy is invisible from the inside. Designing for human judgment means designing against all three.
