I’ve reviewed 20+ enterprise AI builds this year—they all skip this conversation + 7 prompts that force it before anyone writes code

Source: Nate’s Newsletter Date: 2025-12-16 URL: https://natesnewsletter.substack.com/p/ive-reviewed-20-enterprise-ai-builds

Summary

Across 20+ enterprise AI builds reviewed in 2025, Nate identifies a consistent failure mode: teams skip defining “what does good look like?” before writing code, leaving correctness implicit and unmeasurable. The root cause is vagueness as avoidance — deferring hard trade-offs (boldness vs. caution, completeness vs. speed) until production exposes them as hallucinations or shifting quality standards. Seven prompts are provided to force the correctness conversation before any build begins.

Implications

• Feeds the enterprise AI adoption and eval infrastructure threads: the pattern of skipping correctness definition is the same gap that causes agentic systems to fail silently at scale — this is the practitioner-level evidence base.
• The seven prompts (correctness discovery, tradeoff forcer, pre-mortem, eval design, production monitoring) form an informal pre-build checklist that mirrors what mature agentic pipelines need as formal gates.
• “Vagueness as avoidance” is a cultural diagnosis as much as a technical one — suggests that adoption stalls are often organizational (deferred decisions) rather than capability gaps, with implications for how to structure AI rollout.