There is a moment when a product stops being what it was marketed as and becomes something else entirely. For OpenAI's Codex, that moment was April 16, 2026. The update was framed as a feature expansion. What it actually did was reclassify the tool.
Codex started as a coding assistant -- brilliant at writing Python, reviewing pull requests, fixing bugs. More than 3 million developers were using it every week by early 2026. Since the desktop app launched in February 2026, overall Codex usage had doubled. OpenAI was not sitting on something niche. They were sitting on the foundation of something much larger.
The April 16 release added three things: background computer use, a built-in browser with memory, and more than 90 new plugins. Each sounds like a feature. Together, they are a completely different product category.
The Feature That Rewrites the Job Description
Background computer use is the one that matters. Not because it is technically impressive, though it is. Because of what it implies about how work is supposed to happen now.
Codex now has its own cursor. It sees your screen. It clicks buttons, types text, opens applications, and navigates between windows -- all while you continue working in the foreground. Multiple Codex agents can run in parallel without touching what you are doing. The computer is no longer yours alone. You share it with a process that operates on it the way you do.
"That is not a chatbot. That is an AI that works for you in the background, like a silent employee who never asked for a lunch break."
Currently available on Mac. Windows is coming. You do not need to understand APIs for Codex to interact with an app that has no API. It just uses the app the same way a person would: looking at the screen and clicking things.
Think about the practical version of that. You tell Codex to pull your unread emails, summarize them, check if any clients mentioned a specific project, and drop the summary into a doc. You do not wait. You do not watch it work. You are on a call. You come back and the doc is done.
There is also a cloud trigger layer. Tasks can run even when your computer is completely off. The agent is not waiting for you to be present. It is running on a schedule you set once.
Memory: The Feature Everyone Was Sleeping On
The memory update is the one most people skipped past in the changelog. They should not have.
Every AI tool has the same waste built into it: you spend the first five minutes of every session re-explaining yourself. Your project context. Your preferences. The same corrections you made last Tuesday. That overhead compounds across hundreds of sessions into dozens of wasted hours per year.
Codex now remembers. It knows what you are working on, how you like things structured, and what corrections you have made before. The next time you open it, it already knows you.
OpenAI's internal framing: the difference between working with a contractor you just hired versus one who has been on your team for a year. The experienced one is worth ten times more, not because they are smarter, but because you do not have to teach them the same things twice.
90+ Plugins and the "Morning Briefing" Demo
The plugin release was the most legible demonstration of what the new Codex is actually for. Slack. Gmail. Notion. Jira. GitLab. Google Calendar. The entire productivity stack, all wired in.
OpenAI demoed a single prompt during the April 16 launch:
"Check Slack, Gmail, Google Calendar, and Notion and tell me what needs my attention."
Codex ran it. Pulled everything. Synthesized it. Returned a prioritized list. That is a morning briefing. That is a task that, for years, has been the job description of executive assistants and virtual assistants charging $25 to $60 an hour. Codex ran it in under a minute. And you can set it to run automatically every morning.
The "For Almost Everything" Positioning Is Not Accidental
OpenAI's choice of subtitle for this update -- "Codex for almost everything" -- is worth slowing down on. They did not call it "Codex, now more powerful" or "Codex 2.0." They deliberately positioned it as a general-purpose tool.
The most telling line from the official framing: you don't need to know how to code at all.
This is not a small reframe. For four years, Codex has been a developer tool. Developer onboarding. Developer pain points. Developer pricing tiers. The April 16 update is OpenAI telling the other 99% of knowledge workers that the same tool is now theirs.
GPT 5.3 Codex Built Itself
There is a detail buried in OpenAI's documentation that most coverage missed. GPT 5.3 Codex -- the model powering all of this -- was used to help build itself. OpenAI's engineering team used early versions to debug its own training pipeline and manage its own deployment. The model participated in its own creation.
GPT 5.4 is now rolling out. It runs 25% faster than GPT 5.3. The rate limits on all paid plans doubled with the April 16 update.
On pricing: if you are already on ChatGPT Plus at $20 per month, you have Codex access. The $100 per month Pro tier includes five times the usage limits. OpenAI is temporarily giving free users access to try it.
What This Actually Means for Non-Developers
The instinct most people have when they hear "coding agent" is to zone out. That is the wrong call now.
The most creative use cases emerging from the April 16 update come from people with zero technical background. Communication workflow management. Competitive research. Document drafting and review. Repetitive administrative tasks that used to eat half a day.
The pattern is consistent: people are building Codex into systems rather than using it as a one-shot tool. You set up what to check every morning, what reports to generate each week, what to do when specific conditions are met. Then it runs. You designed the system once. The agent maintains it.
That is a different kind of productivity. Not faster task completion. Parallel execution while you work on something else entirely.
The Workspace Agents Expansion
Codex's pivot is part of a broader OpenAI push. Workspace agents -- fully customizable agents buildable through simple chat -- are now live for Business, Enterprise, and Education plans. They are exactly what they sound like: persistent agents with their own tool access, skill libraries, file memory, and cross-channel deployment including Slack.
OpenAI is explicitly positioning these as an evolution of their custom GPTs. The upgrade path is coming: existing GPTs will be convertible into workspace agents. The company is consolidating its agent surface. Everything is moving toward persistent, autonomous, cross-platform execution.
The Shift
For most of Codex's history, the mental model was: I open it, I give it a task, I wait for an answer, I close it. That mental model is now actively wrong.
The correct mental model as of April 2026: you are designing a system that runs without you. Background agents executing in parallel. Persistent memory that compounds context over time. Plugins that connect to the entire work stack. Cloud triggers that keep the system running when your computer is off.
Codex was always described as a coding tool because that was the most legible application of what it could do. What it could actually do was always more than that. OpenAI just stopped pretending otherwise.