ChatGPT Agent Will See You Now

OpenAI’s latest launch turns ChatGPT into a junior analyst with a browser, a calendar, and a to-do list.

OpenAI is back in the news. This time, for something it actually launched. Over the past few months, the storylines have been mostly external: Meta hiring away senior OpenAI researchers.Whispers of e-commerce tools. Rumors of a secret browser.

From OpenAI itself? Nothing. Until today.

This morning, OpenAI hosted a good old-fashioned livestream and introduced ChatGPT Agent - its most significant step yet from assistant that answers to agent that acts. The pitch is simple: describe a task in plain English, and it does the thing.

Not just answering questions - actually doing. The agent clicks through websites, fills out forms, runs code, checks your calendar, generates editable PowerPoints and spreadsheets - and circles back only when it needs approval.

You prompt. It produces a rough draft. You polish.The agent now handles the tedious 80%.

It fuses OpenAI’s past work - Operator (browser automation) and Deep Research (synthesis engine) - into one persistent, goal-completing system. Think of it as ChatGPT with a mouse, memory, and a working knowledge of Excel.

OpenAI says it’s especially good at “first-year analyst” work: compiling competitive landscapes, summarizing earnings calls, maybe even drafting that 20-slide board update you were going to pull an all-nighter on. If this feels like a shot across the bow at Microsoft… well, it is. But OpenAI insists it’s not.

“This is just how people communicate,” said OpenAI’s Chief Product Officer. “They create spreadsheets, they create PowerPoints.”

Sure. And now, so does ChatGPT.

So what’s actually new?

• It runs on a virtual computer. So it can interact with real software UIs, not just APIs

• It outputs fully editable files. Presentations and spreadsheets you can open in Microsoft Office (or Notion, if you’re feeling chaotic)

• It handles research-to-output workflows. Ask it to look something up → analyze it → summarize it → generate a slide deck

• It integrates with your apps. Gmail, GitHub, calendar, etc. (no financial actions… yet)

Latency is still an issue (some tasks take 15–30 minutes), but OpenAI’s not optimizing for speed. It’s optimizing for replacement labor.

With this launch, distribution shifts from API to behavior. This isn’t something a dev team integrates. It’s something you use directly. A personal software layer that lives inside your browser, listens to your prompts, and quietly takes work off your plate.

Let’s talk about safety. There are safeguards. Watch modes. Refusal training. Red teams. It prompts for permission before doing irreversible things, like bookings or purchases - for now

Still, we’re now in a world where you can say:"Plan my birthday, compare cake prices, and email me the top three vendors." - and it’ll do all of that. Autonomously. While you scroll Reddit or write Substack posts about it.

The dream of AI agents has been around forever. But today’s launch is productized, permissioned, and rolling out to Plus, Pro, and Team users immediately. OpenAI has officially graduated from offering inference as a service to execution as a product. If GPT was the brain, this is the body - clicking, typing, formatting.

Like every AI launches, the product will get battle-tested in public now. Let the games begin.

---