• LTSP26

Why AI Strategy Advice Keeps Pushing the Same Trendy Answers w/ Angelo Romasanta

Ask ChatGPT, Claude, and Gemini the same strategy question and 96% of the time you'll get the same answer, dressed up as if it were built just for you. Angelo Romasanta calls this "trendslop"—and it started with a study of AI playing Mafia. Testing seven LLMs across classic management tensions—explore versus exploit, centralize versus decentralize, automate versus augment—Romasanta's team found consistent convergence on the "sexier," socially desirable option: differentiation over commoditization, augmentation over automation. When given the freedom to hedge, models frequently chose both sides at once, a hybrid compromise that's strategically weak whenever the real decision requires saying no to one option. Small changes to wording or option order shifted results dramatically, while more context barely moved the needle—proof this is architecture, not a prompting problem. The fix isn't more technology. Use AI to expand options and stress-test assumptions across separate chats, then let human judgment make the final call. Timestamps: 00:00:54 Romasanta's research began by watching different LLMs play Mafia-style deception games, revealing distinct behavioral strengths and weaknesses 00:02:40 Seven LLMs tested on classic tensions consistently favored the sexier option: differentiation over commoditization, regardless of business context 00:06:52 Given freedom to hedge, models often recommended both sides of a strategic tension at once, a hybrid compromise avoiding real commitment 00:18:24 Testing whether more context or more frameworks fixed the bias showed the real solution isn't more technology, it's managerial judgment 00:24:41 The practical takeaway: use AI to expand options and stress-test assumptions, but keep human judgment responsible for the final choice Read More

• 26 min
• 31 views