Dev argues the real AI agent problem is unmeasurable requirements, not capability
algo_diver · x · 2026-09-23
Quoting a discussion that AI with open-source engines can already build things (slower now, 15 minutes within months), this developer argues the core issue is that critical errors can't even be verified. Delegating work to AI requires measuring whether requirements are met — and not knowing what you're asking for makes the whole premise contradictory.
More from AGI Musings
- NYT: OpenAI's Claimed Navier-Stokes Proof Shows How AI Endangers Science — joannashields · 2026-09-23
- Fields Medalist Martin Hairer: Private-Site Math Releases Undermine Scientific Standards — littmath · 2026-09-23
- Ben Todd: Blaming effective altruism is usually a sign someone doesn't want you thinking — ben_j_todd · 2026-09-23
- AI Agents Aren't People: Anthropic Blamed for Personhood Narrative Shifting Liability — gerardsans · 2026-09-23
- Gergely Orosz: one driven engineer with AI beats hundreds of AI agents — intellectronica · 2026-09-23
- Rob Leclerc: AI doomsday talk repeats the Malthusian error — tech will solve its own problems — robleclerc · 2026-09-23