本文由 AI 分析生成
建立時間： 2026-03-28 來源： https://blog.cashwu.com/blog/2025/kent-beck-software-estimation
Summary
Cash Wu summarizes Kent Beck’s interview on software estimation, arguing that the fundamental problem with estimates is a communication problem, not a technical one. Beck distinguishes three different things people mean by “estimate” (target, commitment, probability distribution), explains why software has inherent fractal measurement properties that make precise estimates impossible, and proposes asking “who wants to know, and for what decision?” before providing any number.
Cash Wu 整理 Kent Beck 關於軟體估算的訪談，論點是估算問題根本上是溝通問題而非技術問題。Beck 區分人們說「估算」時的三種不同含義（目標、承諾、機率分佈），解釋為何軟體具有固有的碎形測量特性使精確估算不可能，並建議在提供任何數字前先問「誰想知道，為了什麼決策？」
Key Points
- Asymmetric risk: missing an estimated deadline causes team pain; finishing early brings no reward — this distorts incentives
- Three types of estimates: Target (fixed deadline, adjust scope), Commitment (triggers accountability planning), Probability distribution (long-tailed, not symmetric)
- Fractal measurement problem: the closer you get to “done,” the more the true scope expands — like measuring coastline length
- False precision: “52.5% complete” gives illusion of accuracy in inherently uncertain work
- Three task categories: trivially small (just do it), known scope (can estimate reliably), unknown scope (spike first)
- Story point estimation’s real value: the team discussion around a story, not the number itself
Insights
The asymmetric risk framing is underappreciated: it explains why developers chronically underestimate (optimism bias + fear of seeming slow) and why managers chronically interpret estimates as commitments. Beck’s “who wants to know, for what decision?” reframe is practically powerful — it often reveals that the requester actually wants a target (can we ship by Christmas?) and not a commitment, which changes what kind of answer is useful. The fractal measurement insight explains why 90%-done projects routinely take as long as the first 90%.
Connections
Raw Excerpt
估點數的真正價值並不在於數字本身，而在於團隊成員圍繞 Story (需求) 進行的討論。