1️⃣ Make priorities survive a direct comparison Three questions cut through a crowded list: defend doing A instead of B, name what cannot fail, and state the cost of delay. Priority labels alone avoid the trade-off.
💡 Why it matters Force every new request to displace something visible. Otherwise the team gets a longer list, not a decision.
☕ Coffee talk Which current priority would you struggle to defend against the one it pushed aside?
2️⃣ Buying a company still beats searching forever Stanford’s 2026 study puts the historical search-fund acquisition rate at 58%. Recent cohorts fell to about 48%, while the average successful search took 20 months.
💡 Why it matters Back the operator and the search process, not just the target. Deal discipline matters more when competition extends the hunt.
☕ Coffee talk At month 18, is patience still discipline or has the searcher started needing a deal?
3️⃣ Replace the AI spec with a test Anthropic’s product team treats evals as the main artifact for model work. Product managers inspect failures, turn them into test cases and use the scores to guide iteration.
💡 Why it matters A PRD describes intended behaviour. An eval exposes whether the system can deliver it, and catches regressions when models change.
☕ Coffee talk How many AI requirements would survive being rewritten as pass-or-fail examples?