'Prompts aren't real': users say frontier models have outgrown the problems they can pose

ctjlewis · x · 2026-09-05

ctjlewis amplifies @miguelfserna's observation and declares: "Prompts are not real."

The quoted post notes that people getting access to frontier models like Astra increasingly struggle to find "worthy" prompts, defaulting to 3D modeling as a test. The author argues many users have genuinely hit the useful intelligence cap: day-to-day technical problems are crackable by models like Fable given enough time, and the bottlenecks have moved to agency, emulating human behavior, and softer skills. Code quality remains an issue, but not due to raw intelligence.

His favorite agency test: ask the model to make itself useful. A notable signal that model capability may be outpacing the questions users can pose.

Related event: Developer Claims 'Prompts Don't Exist,' Sparking AI Community Debate(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →