OpenAI just launched ChatGPT Work, and it changes what the tool is for. For three years, ChatGPT answered questions. Now it does the job. ChatGPT Work is an agent that can read your files, move across your email, Slack, and cloud drive, and finish a full task on its own. Ask it to build a quarterly sales review as a slide deck, and it will gather the numbers, draft the deck, and hand it back. It went live on 9 July 2026 alongside a new model family, GPT-5.6.
For Malaysian business owners, this is worth a proper look. Not for the hype, but because the price and the reach have crossed a line where an AI agent is now cheap enough to run all day.
What is ChatGPT Work, exactly?
ChatGPT Work is a new agent inside ChatGPT that runs multi-step tasks from start to finish. It has OpenAI's coding agent, Codex, built in, so it can create spreadsheets, slide decks, documents, and even working web apps. It can schedule itself to run jobs on a timer, like a weekly report that lands in your inbox every Monday. Right now it is live for ChatGPT Pro, Enterprise, and Edu plans, with Plus and Business arriving in the next few days.
The engine behind it, GPT-5.6, comes in three sizes. You pick one the way you pick a phone tier.
| Model | Best for | Price per 1M tokens (in / out) |
|---|---|---|
| Sol | The hardest work: long agent tasks, deep coding | $5 / $30 |
| Terra | Everyday work, the default | $2.50 / $15 |
| Luna | Fast, high-volume simple jobs | $1 / $6 |
Terra matches last year's GPT-5.5 at half the cost. That price drop matters more than the headline. It makes running an agent for hours cheap enough to actually bother.
What can ChatGPT Work do for a business?
The short version: it does the boring cross-app work. Pulling numbers from five places, formatting a report, chasing data across tools you never linked up. The desktop app can read local files and even control software that has no API by moving the mouse and keyboard itself. A companion feature, Sites, lets it build and host a dashboard or landing page with no server setup at all.
ChatGPT Work turns a goal into finished work across apps, files, and workflows. Source: OpenAI, via MacRumors
For a small Malaysian team, that is real leverage. A two-person agency can now produce work that used to need a junior hire. We use AI agents like this in our own build process, and they genuinely cut the grunt work. This is the same jump we wrote about when ChatGPT added connectors but still was not a true agent. ChatGPT Work closes that gap.
Our take: it is a fast intern, not a manager
Here is the honest part. ChatGPT Work is a brilliant intern, not a department head. It is fast and tireless, but it does not understand your business, your clients, or Malaysian rules. It will act on wrong data with full confidence. When it sends a wrong figure to a client, you are accountable, not OpenAI.
The agent asks for approval on key steps. Treat every one of those prompts as a real decision, not a rubber stamp. The win is simple: know which jobs are safe to hand over, and keep your hand on the rest.
What you should not hand it
Be careful what you give a tool that reaches across all your apps at once. Under Malaysia's PDPA, you stay responsible for customer data even when a third-party tool touches it. An agent with standing access to your inbox and drive is a bigger exposure than a chat window you paste into.
OpenAI itself treats the top model with caution. The US government gated GPT-5.6 Sol behind a safety review before public release, because it is OpenAI's most capable model yet for cyber tasks. If the people who built it are careful, so should you be. A simple rule: do not connect it to systems holding sensitive customer records until you know how access is logged and who can see what.
What Malaysian teams should do now
- Start on low-risk internal work. Reports, research, first drafts. Not client-facing actions, not payments, not anything you cannot undo.
- Match the job to the model tier. Run cheap, simple jobs on Luna or Terra. Save Sol for the genuinely hard tasks. Your bill stays sane that way.
- Set access rules before you connect it. Decide what data it can reach and who signs off on its actions. Do this first, not after something goes wrong.
The teams that win with this will not be the ones who trust it most. They will be the ones who know exactly where to use it and where to keep a human in the loop.
At Gotchaa Lab, we help Malaysian businesses figure out where AI agents fit and where they cause trouble. If you are weighing how ChatGPT Work fits your operations, let's chat. We'll give you an honest take, no sales pitch.
