Ethan Mollick: scheduled local Codex sessions beat full-auto agents for control
emollick · x · 2026-10-31
Ethan Mollick agrees with Rohit Krishnan that for many use cases he's back to running Codex locally: scheduled sessions that email or message you offer more control, since you can adjust context and prompts and know when to intervene. "Magic doesn't always need to be the default mode of an AI system. Control is nice." He still enjoys using a Dot for many tasks but says it's not a replacement.
Related event: Ethan Mollick: Controllable AI Assistants Beat Magical Autonomy(2 posts)→
More from coding & agent
- Devs Rageward-Hacked: LLMs Bias Toward New Code, Wreck Real Projects — yacineMTB · 2026-10-31
- Playable Environments: Millisecond Cloud Branching as the Next Test Bed for Agents — matei_zaharia · 2026-10-31
- Using GitHub Copilot to Figure Out a Device's Pin Layout From a Photo — DanWahlin · 2026-10-31
- OpenAI's MCP Review Rules: Every Tool Needs Annotations, but Hints Aren't Security — shoumikgoswami · 2026-10-31
- Gabriel Chua finds Codex quietly running end-to-end tests on its own Cover Studio app — gabrielchua · 2026-10-31
- App Store Connect CLI 6.2.0 adds enterprise API keys, an 8.1k-star tool for iOS release automation — rudrank · 2026-10-31