Tibo
@thsottiaux
With some small tweaks, Codex can work for days on hard tasks. We will release some changes to make this easier to use for everyone. What’s the hardest task you’ve seen GPT-5.5 succeed at?
79retweets3.9klikes
View on XOpenAI Says Codex Can Already Work for Days, Easier Controls Coming
· Updated
OpenAI signaled that Codex, its agentic coding platform, can already sustain multi-day work on hard tasks when configured with small tweaks. The team will release changes to make these long-running workflows easier for everyone, marking a next step toward unattended engineering agents.
Long-running agentic workflows have been bottlenecked by context loss and orchestration drift over multi-hour sessions. The directional shift mirrors the unattended-project pattern targeted by Google's Gemini Enterprise Agent Platform, which launched with explicit multi-day support. Two large vendors converging on the same horizon points to where agentic coding is heading.
Simplified controls aren't broadly available yet — OpenAI plans to ship them soon. The push toward multi-day work follows OpenAI's Auto-review mode for unattended Codex runs, which already lets the agent execute long-running builds without constant human approval. Watch for the simplification rollout before planning multi-day workflows.
Still wondering? A few quick answers below.
Tibo Sottiaux, OpenAI's Codex engineering lead, said Codex can work for days on hard tasks when configured with small tweaks. He confirmed that OpenAI plans to release changes that make these long-running workflows easier for all users to set up, signaling near-term simplification of multi-day agentic coding sessions.
Multi-day Codex work is technically possible today with manual configuration tweaks, but the simplified controls Tibo Sottiaux described are not broadly available yet. The team plans to release these changes soon, but no specific availability date has been shared. Most users should wait for the official rollout announcement before planning multi-day workflows.
OpenAI Codex uses GPT-5.5, the company's flagship model for agentic coding work. Tibo Sottiaux's tweet referenced GPT-5.5 directly when asking the developer community for examples of hard tasks the model has succeeded at, indicating GPT-5.5 powers these long-running Codex sessions handling multi-day agentic work.
This aligns with industry direction toward multi-day autonomous agents. Google launched its Gemini Enterprise Agent Platform with explicit multi-day support for enterprise workloads. Two large vendors converging on similar horizons suggests unattended multi-day engineering work is the next plateau for agentic coding tools, beyond the single-session paradigm.
