AI coding tool adoption at 84% but trust declining — the "almost right" problem¶
Insight: A paradox is emerging in AI coding tools: 84% of developers use or plan to use them (up from 76%), with Claude Code reaching $400M in five months and Cursor crossing $1B annualized revenue. Yet nearly half of developers actively distrust AI-generated outputs (up from 31% in 2024), and only 3% "highly trust" the code. The core frustration: 45% report code that's "almost right" — looks correct but fails in production.
Detail: Builder.io frames this as the "eager intern" analogy — fast, enthusiastic, 80% of the way there, but the last 20% requires human supervision. This aligns with the industry moving past both "AI will replace us all" and "AI is just fancy autocomplete" toward a more nuanced understanding of AI as a powerful tool requiring skilled management.
Sources
- Builder.io — "The AI software engineer in 2026" (2025-12-17)
Related: existing entry "AI coding quality is a skill issue" in external/claude-code.md — COMPLEMENTS