Familiarity is the enemy: On why Enterprise systems have failed for 60 years

· ai · Source ↗

TLDR

  • Enterprise knowledge management has failed for 60 years because buyers select on vendor familiarity, not correctness, producing hundreds of billions in write-offs.

Key Takeaways

  • The HP/Autonomy $11.1B acquisition – with only 6 hours of due diligence – is the canonical proof that familiarity beats product evaluation in enterprise deals.
  • Rich Hickey’s Simple vs. Easy distinction is the core framework: enterprise procurement rewards “easy” (familiar) over “simple” (correct), compounding accidental complexity for decades.
  • Consulting firms pitch six-figure engagements to build working systems while independent builders demo working systems – buyers still choose the familiar firms for risk insurance.
  • The familiar-language argument (Java, .NET, Oracle) is structurally weakening under AI: LLMs optimize for token efficiency and semantic stability, axes where Clojure-style languages outperform Java-era defaults.
  • Enterprise software avoids outcome-based pricing because vendors know implementation failure rates are high – Akerlof’s market-for-lemons dynamic rewards familiar signals over actual results.

Hacker News Comment Review

  • Strong pushback that the author misreads enterprise procurement: multiple commenters argue IT optimizes for uncertainty reduction and minimal capability sets, not technical quality, making “familiarity” rational under those constraints rather than a pathology.
  • The LLM-favors-unfamiliar-languages claim drew direct skepticism: models are trained on large corpora, and Java/Python have vastly more training data than Clojure, making the author’s argument potentially backwards.
  • Commenters surfaced a missed incentive layer: MBA-network purchasing dynamics and misaligned personal incentives (title, compensation, promotion risk) explain enterprise buying behavior better than simple familiarity bias alone.

Notable Comments

  • @somat: Directly challenges the AI/language claim – LLMs perform best on languages with the largest training corpus, which is the opposite of the author’s argument.
  • @thelastgallon: Points to MBA-network purchasing as the missing root cause: high-dollar contracts signal status and drive compensation, independent of product quality.
  • @louiereederson: Argues enterprise IT is a dynamic system optimizing for abstraction and uncertainty – decisions that look irrational technically are often rational under real organizational constraints.

Original | Discuss on HN