REDDIT · REDDIT// 31d agoNEWS

Anthropic coding forecast misses one-year mark

This Reddit post revisits Dario Amodei’s March 2025 claim that AI would write 90% of code within months and nearly all code within a year, arguing that reality still looks far messier. AI-generated code has surged, but usable production software still depends on humans for architecture, debugging, testing, and judgment.

// ANALYSIS

The hot take is less “AI failed” than “the timeline was hype”: code generation got dramatically better, but autonomous software engineering is still nowhere near hands-off.

–The prediction blurred together code volume and code quality; LLMs can generate plenty of code, but correctness and maintainability remain the real bottlenecks
–By early 2026, tools like Claude Code, Codex, and newer frontier models clearly boosted developer throughput, so the underlying trend was real even if the deadline missed
–The strongest evidence against the forecast is not that AI writes no code, but that teams still need humans to review outputs, run tests, handle edge cases, and make product decisions
–For AI developers, this is a reminder to separate “agentic coding assistant” from “fully autonomous engineer” — those are still very different products

// TAGS

anthropicai-codingagentllmdevtool

DISCOVERED

31d ago

2026-03-11

PUBLISHED

32d ago

2026-03-11

RELEVANCE

7/ 10

AUTHOR

puripy