You know that moment when you realize you’ve been using a Ferrari to go to the grocery store? That’s the feeling you’ll get the second you understand what OpenAI has quietly done to ChatGPT.
Most people still think of it as a smarter search engine. A glorified text generator. Ask a question, get an answer. But over the past few weeks, OpenAI has been merging ChatGPT and Codex into something that looks nothing like a chatbot. It looks like a cloud-based operating system — and your browser is the screen.
ChatGPT isn’t just answering questions anymore. It’s executing workflows. And if you’re still using it the old way, you’re already falling behind.
Let me show you exactly what changed.
Previously, there were three modes: Chat, Work, and Codex. Now there are only two: ChatGPT and Codex. But inside ChatGPT, the ‘Work’ mode has become a full-blown remote execution environment. You can upload files, install Skills, set up scheduled tasks, and run code in a Linux container that lives in the cloud.
Here’s the part that stopped me: I uploaded my custom z-web-pack Skill to the ChatGPT web interface — a skill I originally built to run locally in Codex. The web version executed it on a remote sandbox: 4 CPU cores, 4 GB RAM, 60 GB temporary disk, no GPU. The output was a compressed file that I downloaded directly from the browser.
OpenAI isn’t building a better chatbot. They’re building the operating system for a world without local computers.
Think about what that means. You don’t need to install anything. You don’t need to keep your laptop running. You don’t need to manage dependencies or environments. You just upload a Skill, tell ChatGPT what to do, and the task runs in the cloud. When it’s done, you get the result.
This is the exact same architecture as Codex, but without the heavy client. The web version has integrated file libraries, projects, scheduled tasks, plugins, Skills, and a sandboxed execution environment. The boundaries between a lightweight web chat and a heavy local IDE are dissolving.
For power users, Codex still wins for heavy programming and local development. But for the other 99% of tasks — generating reports, processing data, building websites, automating workflows — the web version is becoming a ‘Cloud Codex’ that anyone can use.
This also explains why OpenAI acquired a company that builds persistent background agents. They’re not adding another mode. They’re merging chat, office work, coding, and automation into a single, unified entry point.
Previously, we opened ChatGPT to ask a question. Soon, we’ll open it to hand off a job and wait for the results.
The direction is unmistakable: ChatGPT is becoming a persistent, always-on AI workstation that runs in your browser. It installs Skills instead of software. It executes tasks instead of generating replies. It works while you sleep.
The next time you open ChatGPT, don’t ask it a question. Give it a job. That’s the only way to keep up with what’s coming.
FAQ
Q: Isn't this just a fancy chatbot with plugins?
A: No. The underlying architecture is fundamentally different. You're not just chatting; you're delegating tasks to a remote virtual machine that runs persistently. Plugins were add-ons; Skills are executable programs that run in a sandboxed Linux environment.
Q: How should I change my workflow today?
A: Start thinking in terms of jobs, not queries. Instead of asking 'What is the weather?', create a Skill that fetches the weather every morning and emails you a PDF. Use the scheduled tasks to run that Skill daily. The goal is to shift from reactive asking to proactive delegation.
Q: Isn't this just a step toward vendor lock-in?
A: Yes, that's the risk. OpenAI is building a proprietary platform where your Skills, files, and automation live in their cloud. The counter is that this is the most accessible compute environment ever created. The trade-off is convenience vs. control. But ignoring it means missing the biggest shift in how we use computers since the browser itself.