OpenAI’s developer-facing Codex app picked up two notable updates this week: custom themes for personalization and Automations moving to GA, with more knobs for controlling how those runs execute.
While neither change alters what Codex is at its core, both speak to the same direction: turning a coding agent into something that fits better into everyday developer workflows—both visually (themes) and operationally (repeatable automations).
Automations hit GA, with more control over execution
The other update is bigger for day-to-day “agentic” work: Automations are now GA. Along with that milestone, OpenAI lists several new controls:
- Set the model and reasoning level
- Choose whether runs happen in a worktree or an existing branch
- Reuse workflows with templates
This is the kind of configuration surface area that starts to make automations practical at scale. Being able to tune reasoning level and model selection gives teams a way to match cost and latency to task complexity—especially for routine “background” work that doesn’t always need the deepest analysis.
OpenAI also highlights recurring use cases including daily repo briefings, issue triage, and PR comment follow-up.
Themes arrive: import, share, and personalize
Codex now supports themes that can be used to personalize the app’s look and feel. OpenAI says themes can be imported and shared, suggesting a lightweight ecosystem of community-created styles in addition to whatever defaults ship in the app.
For teams, that kind of portability matters more than it sounds at first glance: consistent theming across machines can reduce “setup drift,” and shared themes can become part of internal tooling conventions alongside editor settings and dotfiles.
Early feedback: scheduling, UI friction, and workflow asks
Replies quickly gravitated toward a few themes of their own: requests for message editing and conversation branching, a desire for session-level model/reasoning controls rather than global settings, and interest in scheduled or cloud-run automations so jobs don’t depend on a local machine staying awake. There were also reports of potential macOS process cleanup issues and a bug where the automation tab “isn't scrollable” for users with many automations.
Those responses read less like feature creep and more like the predictable next layer of fit-and-finish once automations move from “neat” to “relied upon.”
