Only a few years ago, using a computer meant typing a question and reading the reply. Today, AI agents can open a browser, move through a website and finish an errand while you do something else. That is a lot of change in a short time, so it helps to look at how we got here and what it means for everyday life in India.
The short version
- November 2022: ChatGPT puts a chat box in everyone’s hands.
- October 2024: Anthropic lets Claude see a screen and click, in public beta.
- October 2025: OpenAI launches Atlas, a browser with an agent mode.
- July 2026: OpenAI announces Atlas will be discontinued, as agents move into apps people already use.
The chat box that started it all
ChatGPT launched on 30 November 2022. A UBS analysis, as reported by TechSpot, estimated that it reached 100 million users by January 2023. That made it the fastest growing consumer app up to that point.
For many people, it was the first time a computer answered in plain sentences. Students used it to understand a topic. Writers used it to find a first line. Office workers used it to tidy an email. However, it only talked. It could draft the message, but you still had to copy it, paste it and press send.
From answers to errands: how AI agents began
The next step was to let the software use the screen itself. On 22 October 2024, Anthropic announced “computer use” as a public beta, as Engadget reported. In simple terms, Claude could look at a screen, move a cursor, click buttons and type text, much like a person would.
The early version was honest about its limits. Anthropic called it early and buggy. In one test, Claude stopped a coding task and began browsing photos of Yellowstone National Park instead. Still, the idea was clear. The chatbot was no longer only a place to ask questions. It was becoming a pair of hands.
The browser became the battleground for AI agents
Browsers were the obvious next stop, because so much of daily life now happens in one. On 21 October 2025, OpenAI launched ChatGPT Atlas, a browser with an agent mode, according to Built In. Paid users could ask it to handle multi step tasks while they browsed, such as booking an appointment or planning an event. OpenAI kept tight limits at the start. The agent could not run code, download files, install extensions or act without oversight.
Then the story took a turn. On 9 July 2026, OpenAI announced it would discontinue Atlas, with a shutdown date targeted for 9 August, as 9to5Mac reported. A new ChatGPT desktop app, along with a plugin for Chrome, would carry the same ideas forward.
Our reading is simple. People did not seem to want a brand new browser. They wanted help inside the tools they already use. So AI agents are now moving into existing apps, desktops and extensions, instead of arriving as one more icon to learn.
What AI agents look like in India
India has its own version of this story. From 4 November 2025, OpenAI offered ChatGPT Go free for a year to users in India, with one claim per account, according to Business Today. Anyone who claimed it on the first day will see that year end in early November, so it is worth checking your plan.
Payments are the other signal. On 9 October 2025, NPCI, Razorpay and OpenAI announced a pilot that lets people pay with UPI inside ChatGPT, using the reserve pay feature, as Onmanorama reported. Bigbasket was among the first platforms. It is a pilot, not a finished product. Still, it shows the direction: an assistant that does not only suggest a purchase but completes it.
What has not changed with AI agents
Mistakes still happen. An agent that clicks the wrong button costs more than a chatbot that writes a wrong sentence. So a few habits matter. Watch the first few runs. Keep payments behind your own approval. Never hand over passwords or one time codes. If anyone or anything asks you for an OTP, stop.
Scams are part of the same story, because fraudsters adapt to every new tool. Our guide to reporting suspected fraud calls explains what to do when something feels off.
Where this leaves us
In under four years, AI moved from answering to acting. The next question is not whether AI agents get smarter. It is how much control we are comfortable giving them, and how clearly they ask before they act. That is a choice for each of us, and it is worth making on purpose.
FAQ
What are AI agents?
AI agents are programs that carry out a task in several steps, not just answer a question. For example, one can open a website, fill in a form and report back. You stay in charge by approving the steps that matter.
Is it safe to let an AI agent handle payments in India?
It depends on the tool and the controls it offers. Payments inside ChatGPT with UPI are still a pilot. Keep final approval with you, and check the amount and the merchant before anything is confirmed.
Is an AI browser different from a normal browser?
An AI browser adds an assistant that can read a page and act on it. However, OpenAI is winding down Atlas, and similar features now come as desktop apps and plugins for browsers you already use. So the idea has outlived the product.