As we near ChatGPT’s second anniversary, we’re expecting some sort of new feature announcement from OpenAI. It can’t be Advanced Voice Mode, which already launched several weeks ago after being unveiled in mid-May. It also can’t be ChatGPT Search, which became available a few days ago.
GPT-5 was always a possibility, considering all the rumored work OpenAI has done on the next-gen ChatGPT model. While some reports said the next-gen version of ChatGPT might be available to select enterprise users in December, that might no longer be the case.
GPT-5, if it’s even called that, might see delays. More recent reports said that all AI labs are facing issues training more sophisticated models. The list includes OpenAI, so GPT-5 could be postponed to next year. At no point did OpenAI commit to a late 2024 release for the GPT-4 successor, however.
The same reports said that AI companies might look at offering other features to consumers rather than next-gen models. They pointed to AI agents being the next big thing in genAI products. This applies to OpenAI, of course, since CEO Sam Altman hinted at AI agents recently.
A new story says that the first ChatGPT AI agent might drop early next year. Codenamed Operator, the AI might be able to control your computer on your behalf, similar to what Anthropic and Google want to do with their AI agents.
The Operator codename, which triggered Matrix connections immediately even though it has nothing to do with those fictitious operators, comes from Bloomberg. The report notes that an Operator AI agent will be able to write code or book travel via a web browser on your computer.
For some reason, the same travel-booking example comes up every time companies try to explain what AI agents might be able to do. Are they already running low on ideas?
As I told you recently, I don’t want to use genAI products like ChatGPT to only ask questions. I want to tell the AI what to do on my computers (smartphone included) on my behalf. It won’t be just about booking travel. I’ll want the AI to browse the web, save files, send emails, and adjust settings when I tell it to. I also want all of that action to happen on-device with no data sent to a company’s servers.
Before we get to that, we’ll need to use first-gen agent tools like ChatGPT’s Operator. According to the report, OpenAI informed staff in a meeting on Wednesday that the tool will be available in January 2025. It’ll launch as a research preview for developers rather than regular users. It’s unclear when end-users will get access to it.
People familiar with OpenAI’s work on AI agents said the company has been developing several agents. The one nearest to completion will be a “general-purpose” one that can browse the web for you.
The closest thing to a confirmation that OpenAI will release AI agents for ChatGPT soon came from Sam Altman a few days ago during a Reddit Ask-Me-Anything (AMA) chat with users. “We will have better and better models,” Altman said. “But I think the thing that will feel like the next giant breakthrough will be agents.”
Altman never mentioned the Operator codename. He didn’t offer launch details for the ChatGPT AI agents either.
Anthropic has already demoed a similar AI agent concept for Claude, which can use the computer in real-time to browse the web and code. Microsoft has similar AI agents in place that can send emails and manage user records. Google’s Jarvis will also get internet-browsing powers.
Let’s not forget that Siri will also be able to manage some apps and execute more complex actions for you on the iPhone next year. It’s all part of the Apple Intelligence vision that Apple unveiled at WWDC. In hindsight, that’s agentic functionality from Siri.
