Codex Is Learning to Work Continuously
💡Codex may evolve from a coding assistant into an always-on autonomous project worker.
⚡ 30-Second TL;DR
What Changed
Persistent mode is currently being tested in the Codex command-line tool and is not widely available.
Why It Matters
If released, persistent Codex could shift coding agents from request-response tools toward autonomous project collaborators. Developers will need stronger task boundaries, approval workflows, monitoring, and cost controls for long-running agent processes.
What To Do Next
Prototype a long-running Codex workflow in a sandbox with explicit approval gates, session persistence, activity logs, and per-task token budgets.
Key Points
- •Persistent mode is currently being tested in the Codex command-line tool and is not widely available.
- •Codex can continue working until it is put to sleep, potentially consuming more compute, tokens, and time than other modes.
- •An initiative capability lets the agent schedule follow-up work using prior conversations and its understanding of the user.
- •The mode preserves existing permission boundaries and still requires user approval for actions affecting external systems.
🧠 Deep Insight
Background and context from public sources — not the original article. 11 sources cited.
🔑 Enhanced Key Takeaways
- •Codex has evolved from a code-specific assistant into a comprehensive desktop workstation capable of screen interaction, web browsing, and image generation.
- •The platform now supports integration with over 90 external enterprise tools, including Slack and Notion, expanding its utility beyond software engineering.
- •OpenAI open-sourced the core 'Harness' framework in August 2026, enabling developers to embed AI agent loops directly into their own proprietary software.
- •Codex has seen significant adoption in non-technical sectors, with rapid growth in legal, business operations, and human resources departments.
- •The persistent mode is categorized within the Codex 'Inference Intensity' menu, allowing users to manually calibrate the trade-off between compute resources and task autonomy.
📊 Competitor Analysis▸ Show
| Feature | Codex (Persistent Mode) | Claude Code (Anthropic) |
|---|---|---|
| Core Focus | Desktop Workstation/Agentic Loop | Multi-session/Developer Routines |
| Integration | 90+ External Tools | Developer-centric ecosystem |
| Market Position | Enterprise/General Business | Developer/Engineering focused |
| Availability | Testing (Closed) | Available (2026 H1) |
🛠️ Technical Deep Dive
- Persistent mode utilizes an agentic loop architecture that maintains state across sessions by leveraging the newly open-sourced Harness framework.
- The system implements a permission-gating layer that intercepts calls to external APIs, requiring explicit user authorization for non-local system modifications.
- Inference intensity settings allow for dynamic adjustment of token budget and compute allocation, directly impacting the agent's planning depth and proactive task scheduling.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (11)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: IT之家 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.