Most people use Codex for exactly one thing: writing code. But Jason (@jxnlco), a member of the official Codex team, shared a set of deeper techniques — combining durable conversation threads, voice input, task intervention, automation, and other capabilities to turn Codex from a coding assistant into an all-purpose tool that handles all kinds of computer work for you. Here are the 7 core techniques.

1. Durable Threads

Don't start from zero every time. "Pin" the conversation threads you use often (Pinned threads), and Codex will remember the decisions, preferences, and progress from before.

Recommended pinning setup:

• A "chief of staff" thread that handles daily odds and ends
• A thread dedicated to product launches
• A thread responsible for reviewing documents
• A monitoring thread that keeps an eye on external data

The shortcuts Command-1 through Command-9 switch you to the corresponding pinned thread instantly.

2. Capture Raw Ideas with Voice Input

The value of voice input is this: capture the rawest thought in your head before you start agonizing over the wording.

"I remember someone named Ben mentioned this on Slack. I forget the details. Go find it for me."

This kind of vague but directional instruction is entirely enough for an Agent like Codex that can search and gather context on its own.

Tip: Spending 2-3 minutes rambling at Codex to dump your thoughts is worth more than writing a polished paragraph. Rough notes preserve the hesitant tone, the points you stressed, and the half-finished sparks of ideas.

3. Steering vs. Queuing

These are two easily confused but completely different capabilities:

Steering: interrupt Codex mid-task and change what it's doing. Best for emergency course corrections when you notice the AI going off track.

• "Make this a bit smaller"
• "The spacing between these two elements is wrong"

Queuing: don't interrupt the current task; line up new tasks behind it. Best for arranging upcoming work.

• "When this job is done, send the preview link to Slack for the reviewers"

The key difference: steering changes the present, queuing arranges the future. Combining the two keeps you in a constant "human and machine as one" sense of control.

4. Extend Codex's Reach

Codex's reach can extend outward layer by layer:

5. Thread Automations

This is the key capability that lets Codex keep working while you're away.

Thread automation works like a scheduled "heartbeat" mechanism: on a set timetable, it returns to the same thread and continues working.

Example: run the "chief of staff" thread every 30 minutes:

Every 30 minutes, check my Slack and Gmail for messages that need handling but haven't been answered. Prioritize them for me. If someone asked a question, research it thoroughly and draft a reply, but don't send anything directly.

By the time you're back at your computer, the most time-consuming part — gathering background material — is already done, and all you have to do is make the final call.

Feedback loops: thread automations can also watch for comments on pull requests, Google Docs, or Slack, and automatically push the follow-up changes forward while you're away.

6. Goal Setting (Goals)

Goal setting = continuous execution + a verifier. You define the desired outcome, the stop conditions, and the signals to judge by.

A bad goal:

Implement the plan in this Markdown file.

A good goal must have measurable success criteria:

Development of this new version isn't complete until every unit test passes.

Good verifiers include: complete test suites, benchmark performance tests, reliably reproducible bugs, and end-to-end workflows.

Ambition without a verification mechanism is just wish-making.

7. Shared Memory with a Knowledge Base

"Anchor" your durable threads to a knowledge base (an Obsidian vault, which is essentially just a folder of plain text), and let context break free of the boundaries of a single chat.

Recommended directory structure:

vault/
├── AGENTS.md     # Ground rules for Codex
├── TODO.md
├── people/
├── projects/
├── agent/
└── notes/

State the rules in AGENTS.md:

• Treat ~/vault as the long-term working memory area
• Keep your notes organized; don't scatter fragments everywhere
• Categorize todos, people, projects, and daily summaries accurately
• Don't modify files without substantive new progress

This knowledge base isn't for storing code — it stores the ever-rolling context: who is involved, what changed, where things are stuck, and who follows up next.

Summary: Codex's real power doesn't lie in any single feature, but in combining them — durable threads provide continuity, voice input lowers the barrier to expression, steering and queuing keep you in control, automations let the system run itself, goal setting draws a clear finish line, and the knowledge base settles into long-term memory.