OpenAI's Jason Liu Reveals Advanced Codex Workflows for Continuous Automation

Persistent Threads as Long-Running Work Systems

Jason Liu's core insight is treating Codex not as a chat tool but as a continuous work system. Instead of single-session chats, he maintains cross-month "giant threads" for distinct workflows: one for calendar management, one for open-source projects, one for social-media monitoring. He switches between them with Command-1 through Command-9 shortcuts. Each thread accumulates months of conversation history, preferences, and decisions, so the agent can resume context without re-briefing.

Voice Input and Steering for Frictionless Delegation

Liu dictates tasks rather than typing. He argues voice preserves raw, unstructured thought — including vague, jumping, or traceability-requiring ideas — without prompt engineering overhead. Combined with Codex's Steering feature, he can interject new instructions mid-execution and walk away, eliminating wait time.

Heartbeats + @computer: From Tool to Autonomous Employee

The Heartbeats feature acts as a cron-like scheduler. Liu's "Chief of Staff" thread runs every 30 minutes: it scans Slack and Gmail, prioritizes messages, drafts replies (but does not send), and leaves final approval to him. A more complex example: during an animation project, he posts a video to a Slack review thread and sets Codex to poll every 15 minutes. When feedback arrives, Codex re-renders a new version and replies in-thread. Because the Slack MCP server lacks file-upload support, the agent uses @computer to click "Add file" and upload the rendered asset.

Another anecdote: before a shower, Liu asked Codex to monitor an Amazon customer-service queue; by the time he finished, the refund had been processed. Similar loops now extend to Google Docs comments, GitHub PR reviews, and any scenario with a feedback loop.

Verification Mechanisms: The Gatekeeper of Autonomy

Liu stresses that without a verification mechanism, ambition is just a wish . He tasked Codex with porting the Python Rich library to Rust, mandating that all original Python unit tests must pass . Test success determines task completion; failures trigger automatic repair cycles. This principle is now baked into Codex's Goal mode (recently graduated from experimental): users define a final goal and acceptance criteria, and Codex drives toward it for hours or days, with progress checks and course corrections allowed.

Local-First Memory: Obsidian Vault Over Platform Lock-In

Liu rejects platform-native memory (like Codex's Chronicle screen-capture feature, which he considers immature due to permission, rate-limit, and privacy constraints). Instead, he anchors all long-term context in a personal Obsidian vault structured as TODO, people, projects, agent, notes. A top-level AGENTS.md defines sync rules: personnel changes, project updates, task completions must update the vault. This gives him full control, version history, rollback capability, and portability across tools.

Evolving Workbench: Sidebar Rendering, Browser Control, Reusable Workflows

Codex's sidebar now renders Markdown, filters tables, displays PDFs and PPTs. The built-in browser lets agents control pages via JavaScript; Liu often keeps Storybook, Remotion Studio, and Slidev open side-by-side. His preferred delivery artifact is a single index.html with embedded JS/CSS — no deploy, no server. He also packages successful flows as Connectors and Skills for one-click reuse.

Remote Operation After Lock Screen

A recent update enables Codex to continue working after the host machine locks. Mobile apps provide real-time monitoring, approval, and task takeover. As Liu quips: "You clock out, it works overtime; you lock the screen, it keeps grinding."