Randy's technique for evaluating AI-generated output: apply the same critical-thinking scrutiny you'd apply to a claim from a human teammate — 'I have to have enough context to be able to call BS on them.' The catch is that this check only works if you already have enough domain context to catch a plausible-sounding error; without it, AI's mistakes are harder to detect than a colleague's, precisely because the output reads fluently. This pairs with AI for Synthesis, Done With the Team as a second guardrail against over-trusting AI in discovery and synthesis work, and echoes the caution in Caution Against Abdicating Product Sense to AI Tools.