ChatGPT’s new Codex upgrade changes the app from a coding tool into something much bigger. This is not just about writing code anymore. With skills, plugins, app connections, computer use, and automations, Codex is starting to look like a true AI super app that can help run real work across your tools.
If you already have a ChatGPT subscription and you care about productivity, AI agents, or automating repetitive tasks, this is one of the most important updates to pay attention to. The big shift is simple: Codex can now do far more than code. It can connect to apps, follow repeatable processes, operate on a schedule, and even interact with your computer in ways that feel much more agentic than the old version.
That opens up a lot of possibilities, from checking Gmail and Calendar to creating reports, drafting content, testing software, and managing recurring workflows.
What changed in Codex?
The core change is that Codex is no longer limited to development use cases. Previously, it was mainly built around coding. Now it includes a much broader layer of automation features that make it useful even for people who are not trying to build software full time.
The new experience introduces a few major building blocks:
- Plugins for tools and integrations
- Apps you can connect directly
- MCPs for expanded capability access
- Skills for repeatable SOP-style behaviour
- Automations for scheduled and agentic task execution
- Computer use so Codex can control your Mac from inside the app
Put all of that together, and the result is something much closer to an AI operating layer than a basic assistant.
Why this matters: Codex is becoming a super app
The easiest way to understand this upgrade is to stop thinking about Codex as a coding assistant and start thinking about it as a work orchestration app powered by AI.
It can still help you code. In fact, Rob points out that you should still use Codex-specific models for coding tasks to get the best results. But now it can also:
- Check your email
- Review your calendar
- Look through Slack or Notion
- Draft updates and recaps
- Create assets for pages
- Test software and identify bugs
- Run recurring automations on a schedule
That is a massive expansion in practical use.
The new plugin and integration system
One of the biggest upgrades is the new plugin management flow. Inside Codex, you can now manage all the different extensions and connected tools from one place.
That includes:
- Plugins for new functionality
- Apps such as Gmail and Calendar
- MCPs for additional connected services
- Skills that define how Codex should perform specific tasks
There is also a Create option that lets you make your own plugin or skill simply by describing what you want it to do.
That is where this starts getting really interesting. Instead of manually building every workflow from scratch, you can ask Codex to help create the exact tool or procedure you need.
Computer use is a huge deal
One standout addition is computer use. This gives Codex the ability to control your Mac from inside the app, similar to other AI coworker-style systems.
That means AI is not just generating instructions anymore. It can actually navigate apps and perform actions for you.
That is a very different category of capability compared with a normal chatbot.
Skills: the SOP layer that makes AI more reliable
One of the smartest parts of this upgrade is the ability to create skills. Rob explains skills in a very practical way: they are basically SOPs for AI.
That matters because one of the biggest problems with AI tools is inconsistency. You can ask for the same task over and over, and the output can vary a little each time. Sometimes that creativity is helpful. Sometimes it is a problem.
Skills help solve that by giving Codex a defined process to follow.
For example, you can create a skill that says:
- Use a specific tone
- Follow a certain structure
- Prioritize certain elements
- Format the output in a repeatable way
Rob demonstrates this with a skill for generating YouTube scripts in his own tone and format. Codex then scaffolds that skill and turns it into a reusable system.
The resulting skill includes things like:
- An overview of what the skill does
- Workflow instructions
- Priority rules
- Voice matching guidelines
- Output structure
This is important because it makes AI more operational. Instead of prompting from scratch every time, you are teaching it how to work like your process.
Why skills matter for everyday work
If you create content, manage projects, or repeat the same workflows every week, skills can save a lot of time.
They are especially useful when:
- You want consistent output
- You have a repeatable process
- You want to reduce prompt rewriting
- You need AI to act more like a trained assistant than a general chatbot
That is a big leap from casual prompting.
Automations: where Codex becomes agentic
The real power shows up when you combine skills + plugins + automations.
This is where Codex starts to behave less like a tool you use manually and more like a system that can handle work on its own.
You can build automations for things like:
- Status reports
- Release prep
- Incidents and triage
- Code quality checks
- Maintenance workflows
- Growth and exploration tasks
Creating an automation is straightforward. You can choose:
- The project to work in
- The work tree or context
- Whether it runs in chat, local, or another environment
- The schedule
- The model
- The reasoning level
Then you just provide the prompt, name the automation, or use one of the built-in templates.
That means recurring workflows do not have to start from zero every time. Codex can keep them running.
A practical example
One example Rob shows is asking Codex to check Slack, Gmail, Google Calendar, and Notion and identify what needs attention. That is already useful as a one-time task. But the more powerful move is turning that into an ongoing automation that runs every hour and sends notifications inside Codex.
That is the difference between “AI helps me” and “AI handles this for me unless I need to step in.”
Permissions, models, and reasoning levels
There are also some important controls built into the new system.
Permissions
You can choose between:
- Full access, which gives Codex broader control but comes with more risk
- Sandboxed execution, which is safer and helps prevent unintended damage to files or systems
This matters because once AI can interact with your computer and connected apps, permission settings are no longer a minor detail. They are a core part of using the system responsibly.
Model selection
You can also switch between different models depending on the task.
Rob’s guidance is simple and useful:
- Use Codex models for coding-related work
- Use regular GPT models for non-coding tasks
That helps you get stronger output based on the actual job you are trying to complete.
Reasoning level
You can adjust the reasoning depth as well. Higher reasoning can be powerful, but it may use credits faster. So if the task does not require deep analysis, it makes sense to keep that setting under control.
That balance matters if you want to use Codex heavily without burning through usage unnecessarily.
Real examples of what the new Codex can do
To understand why this update is such a big deal, it helps to look at the examples Rob walks through.
1. Testing a Mac app automatically
Using the Xcode plugin and computer use, Codex can launch an app, interact with it, test it like a user, identify bugs, try to reproduce them, and generate a report.
That is a serious leap in capability.
Instead of only helping write code, Codex can help operate software and inspect behaviour. That makes it useful for QA-style workflows, debugging, and iterative improvement.
2. Creating a landing page image and inserting it into HTML
Rob also shows a workflow where a landing page is missing a hero image. Inside Codex, using GPT-5.4, he can simply ask it to generate an image for the hero section and add it to the HTML.
Codex can access the page files, generate the asset, and place it into the markup.
That is exactly the kind of cross-functional workflow that makes the “super app” label feel justified. It is no longer just writing code or generating text. It is handling multiple steps across creation and implementation in one place.
3. Monitoring your work stack
Another strong use case is AI-assisted triage across your work tools. By checking Gmail, Calendar, Slack, and Notion together, Codex can surface what needs attention right now.
That can turn into:
- A daily review assistant
- An hourly ops monitor
- A personal chief-of-staff style workflow
And because it can notify you inside Codex, it starts to act less like a passive tool and more like a working system.
How to set Codex up the right way
If you want to get actual value from this update, the setup order matters.
1. Connect your apps and plugins first
Start by linking the tools you already use. The examples shown include things like:
- Gmail
- Calendar
- Google Drive
- Xcode
- Productivity tools
- Various coding and design integrations
The more relevant tools you connect, the more useful your automations can become.
2. Create skills for your repeatable tasks
Once the tools are connected, build skills for anything you do repeatedly.
Examples include:
- Writing scripts
- Drafting weekly recaps
- Preparing reports
- Formatting summaries
- Running standard quality checks
The point is to convert your repeated manual process into a reusable AI workflow.
3. Automate what is still manual
After your apps and skills are in place, look at what you are still doing by hand.
Rob’s position is pretty clear here: if a task can be handled by the skills, MCPs, or plugins available inside Codex, and you are still doing it manually, you are wasting time.
That may sound aggressive, but the logic is solid. If the system can do the work consistently, on schedule, and with access to the right tools, then your role should shift toward review and decision-making rather than repetitive execution.
Where this is headed
The broader takeaway is that OpenAI appears to be building Codex into a direct competitor in the emerging AI coworker space.
This upgrade shows the strategy pretty clearly:
- Connect to tools
- Define process with skills
- Execute with automations
- Use computer control where needed
- Keep everything inside one AI workspace
That is how you build a super app.
And it is not hard to imagine where this keeps going. More integrations, more templates, better automations, and more ways for AI to coordinate work across apps without constant human prompting.
Best practices for using the new Codex without creating chaos
Just because Codex can do more does not mean you should hand it everything at once. A few smart habits go a long way.
- Start with low-risk workflows like summaries, recaps, or reminders
- Use sandbox mode when testing new behaviours
- Create skills before building big automations so your outputs stay consistent
- Match the model to the task instead of defaulting to one model for everything
- Keep an eye on reasoning settings so you do not waste credits
- Review high-impact actions before giving broader permissions
FAQ
What is the new Codex super app?
It is an upgraded version of ChatGPT’s Codex that now supports much more than coding. It includes plugins, app connections, skills, automations, and computer use, allowing it to complete a wider range of productivity and workflow tasks.
Can Codex be used for non-coding tasks now?
Yes. That is one of the biggest changes. Codex can now help with tasks like checking Gmail, reviewing Calendar, looking through Notion or Slack, drafting recaps, creating assets, and running recurring automations.
What are skills in Codex?
Skills are structured instructions that act like SOPs for AI. They help Codex complete tasks in a consistent way by defining tone, structure, workflow, and output format.
What are Codex automations?
Automations are scheduled or repeatable AI tasks that can run using your connected apps, plugins, and skills. They are useful for things like status reports, triage, monitoring, and maintenance workflows.
Should I use Codex models or GPT models?
Use Codex models for coding work. For non-coding tasks, use regular GPT models. Choosing the right model helps improve output quality and efficiency.
Is computer use in Codex risky?
It can be, depending on the permissions you allow. Full access gives more control but also increases the chance of unintended actions. Sandboxed execution is safer when testing or working on lower-risk tasks.
Final takeaway
The new Codex upgrade is not just a feature refresh. It is a pretty clear signal that ChatGPT is moving toward becoming a full AI productivity platform.
The combination of skills, plugins, automations, and computer use makes Codex useful in a way that goes far beyond prompting for answers. It can now behave more like an AI operator that works across your apps, follows your processes, and handles recurring tasks with less manual effort.
If you already have access, the smartest move is simple:
- Connect your tools
- Create a few key skills
- Automate one or two recurring tasks first
- Expand from there
That is how you turn this from a cool demo into a real productivity advantage.
If you are publishing this on your site, a good next step is to add related internal links, encourage comments about real-world use cases, and continue building around AI automation topics while this category is moving fast.