Coding benchmarks are indirect measures of how automatable every white-collar job is, Redditor argues
simmol · reddit · 2026-09-06
Paying $400/month for Claude Code and Codex, a non-programmer computational researcher argues coding is the execution layer of most white-collar automation: early LLMs failed at law/analysis tasks often because scripts, context fetching, and data plumbing broke—not domain reasoning. As AI gets exceptional at coding, it can build the machinery to automate other professions itself, making coding progress everyone's job-prospect problem.
More from AGI Musings
- Most AI wet-lab 'data factories' are built for VCs, not markets, argues poster — kenbwork · 2026-09-06
- Dev argues 'alignment' has become a hollow term: just 'whatever leads to utopia, not doom' — zetalyrae · 2026-09-06
- Paul Graham says AI kills rambling writing; reply counters polish hides low idea density — omooretweets · 2026-09-06
- Companies are losing designers to engineering en masse as design and engineering merge into one role — yakuzeg · 2026-09-06
- Sam Altman on AI jobs: future work may look like playing games to us today — soumitrashukla9 · 2026-09-06
- Ben Todd: OpenAI Sits on the Most Important Misalignment Data Ever and Won't Share It — ben_j_todd · 2026-09-06