ChatGPT Agent reportedly lost 75% of its users because nobody knew what it was actually for

OpenAI seems ready to pull the plug on ChatGPT Agent, a product that never found its footing since launching six months ago.

The tool peaked at launch with four million weekly active paying users, roughly eleven percent of the 35 million paying subscribers at the time, reports The Information. Within months, that number dropped below one million. Users either couldn't figure out what to do with it or didn't know it existed. Speed, reliability, and cybersecurity issues didn't help.

OpenAI is now betting on specialized agents instead, like the shopping agent for product recommendations, which it calls "Shopping Research." Users understand what these tools are for, and they're more reliable because the product team doesn't have to cram in as many features. It's also an admission that AI models aren't yet great at generalization.

The naming never made sense

One thing that always bugged me: calling this product "ChatGPT Agent" was confusing from the start. ChatGPT has been agentic for a while; reasoning models can write code, browse the web, and analyze images if the task requires it. ChatGPT Agent's only real differentiator was the virtual browser. But slapping "Agent" on this one product made it sound like the other modes weren't agentic, even though OpenAI had already launched Deep Research, which it called an AI agent, before ChatGPT Agent even existed. Shopping Research is essentially a Deep Research agent optimized for product search.