Simon Willison’s Weblog

Subscribe

Introducing the Codex app. OpenAI just released a new macOS app for their Codex coding agent. I've had a few days of preview access - it's a solid app that provides a nice UI over the capabilities of the Codex CLI agent and adds some interesting new features, most notably first-class support for Skills, and Automations for running scheduled tasks.

Screenshot of a macOS desktop application with a dark sidebar and light main content area. Left sidebar shows navigation items "New thread", "Automations", "Skills", and a "Threads" section containing two project folders: "local-codex-scratch" with tasks "Reply to greeting task 2h" and "List Codex.app contents 3h", and "shot-scraper" with tasks "Validate tests on dependency... 3h", "Document shot-scraper CLI usage 18h" (highlighted), and "Run this: uv run pytest -x 1d". Bottom left shows "Personal". Top bar displays "Document shot-scraper CLI usage shot-scraper" with "Open" and "Commit" dropdown buttons. Main content shows a conversation: user message "Build a skill describing how to use the shot-scraper CLI tool" followed by assistant responses: "I'll use the skill-creator skill, scan the repo for existing shot-scraper docs/commands, then draft a concise skill in this workspace and sanity-check it." then "I've confirmed the skill-creator scripts exist. Next I'll inspect the CLI entrypoint and docs, then scaffold a skill under skills/ in this repo." then "I've scaffolded skills/shot-scraper-cli/. I'll now replace the template SKILL.md with a task-focused skill and add a compact CLI reference under references/." then "I'll run the skill packager/validator next to ensure the new skill structure is valid." then "Done — I created and packaged the skill. Key outputs: • Skill definition: SKILL.md • Compact reference: cli-reference.md • Packaged skill: shot-scraper-cli.skill". Bottom shows input field with placeholder "Ask for follow-up changes", "+ Custom Medium" dropdown, and "Local" and "main" branch indicators.

The app is built with Electron and Node.js. Automations track their state in a SQLite database - here's what that looks like if you explore it with uvx datasette ~/.codex/sqlite/codex-dev.db:

Database schema documentation on light gray background showing three tables: "automation_runs" (teal underlined link) with italic columns "thread_id, automation_id, status, read_at, thread_title, source_cwd, inbox_title, inbox_summary, created_at, updated_at, archived_user_message, archived_assistant_message, archived_reason" and "1 row"; "automations" (teal underlined link) with italic columns "id, name, prompt, status, next_run_at, last_run_at, cwds, rrule, created_at, updated_at" and "1 row"; "inbox_items" (teal underlined link) with italic columns "id, title, description, thread_id, read_at, created_at" and "0 rows".

Here’s an interactive copy of that database in Datasette Lite.

The announcement gives us a hint at some usage numbers for Codex overall - the holiday spike is notable:

Since the launch of GPT‑5.2-Codex in mid-December, overall Codex usage has doubled, and in the past month, more than a million developers have used Codex.

Automations are currently restricted in that they can only run when your laptop is powered on. OpenAI promise that cloud-based automations are coming soon, which will resolve this limitation.

They chose Electron so they could target other operating systems in the future, with Windows “coming very soon”. OpenAI’s Alexander Empiricos noted on the Hacker News thread that:

it's taking us some time to get really solid sandboxing working on Windows, where there are fewer OS-level primitives for it.

Like Claude Code, Codex is really a general agent harness disguised as a tool for programmers. OpenAI acknowledge that here:

Codex is built on a simple premise: everything is controlled by code. The better an agent is at reasoning about and producing code, the more capable it becomes across all forms of technical and knowledge work.

Claude Code had to rebrand to Cowork to better cover the general knowledge work case. OpenAI can probably get away with keeping the Codex name for both.

Monthly briefing

Sponsor me for $10/month and get a curated email digest of the month's most important LLM developments.

Pay me to send you less!

Sponsor & subscribe