GPT-6 in Codex still needs human steering on hard problems, researcher reports
DimitrisPapail · x · 2026-09-05
Dimitris Papail reports that GPT-6, now available inside Codex, still requires human steering on hard problems — a hands-on data point on the new flagship model's autonomy limits in real coding workflows.
More from coding & agent
- Shanghai AI Lab's Harness-of-Harness runs 70+ iterations to build an FPS game autonomously — aigclink · 2026-09-05
- pdf-inspector retest: 947 pages in 6.9 min, but 66.6% of numeric values missing from markdown — Ubunta · 2026-09-05
- Stanford Launches CS329Z, a Full Course on Engineering AI Agents — jyangballin · 2026-09-05
- Opus 5 autonomously chains ZImage and Minimax Music via MCP to make art on a 3090 — dkackman11 · 2026-09-05
- Open-source options workbench builds a local AI brain that learns your trading discipline — garrytan · 2026-09-05
- Why doesn't Codex CLI on Linux have Claude Code-style remote control? Users want it — Lanky-Storm7 · 2026-09-05