By Belle Lin

OpenAI is giving businesses the ability to build their own artificial intelligence agents, which are technologies that can independently perform tasks on behalf of humans.

The San Francisco-based AI lab on Tuesday unveiled an agent-building platform that lets companies create their own bots for work such as financial analysis and customer service. OpenAI now has two million paying business users of its ChatGPT Team, Enterprise and Edu products, it said, compared with the one million it had last fall.

The company's announcement comes as both competition and hype around agents grow. More recently, a Chinese upstart called Manus AI generated waves on social media when it made a debut of its own "general" AI agent, which it said can autonomously perform tasks such as data analysis.

Agent technology, which hasn't yet seen wide adoption among companies, promises to usher in the next wave of corporate productivity after initial excitement over the release of OpenAI's ChatGPT in late 2022.

But that wave of productivity is still far off, experts say. AI agents can perform simple tasks, such as ordering office supplies, but they aren't yet trusted by enterprises for high-stakes tasks like financial transactions or hiring new workers.

OpenAI hopes that will start to change as its AI -- especially its so-called reasoning models -- improve. Like some of its peers similarly selling agent software, the AI startup has proclaimed that 2025 is "the year of agents," where the capabilities of the technology finally match up to its usefulness for businesses.

"Now, people can engage with agents," OpenAI Chief Operating Officer Brad Lightcap said in an interview. "Those agents can go off and actually reference files, they can search the web, they can use computers."

In January, OpenAI announced an AI agent called Operator, which can independently browse the internet to do things like buy groceries and file expense reports. And in February, the company announced "deep research," an agent that can perform more complicated research tasks on the internet -- promising to accomplish in tens of minutes what takes humans many hours.

Both those agents don't offer as much customizability, which OpenAI is betting businesses will want. The startup said it would charge customers based on the number of search queries, actions and data storage that the agents actually end up using, as well as for general AI model use. For instance, a company building a legal assistant agent will be charged $2.50 for every thousand queries a user makes to a knowledge base of past legal cases, according to OpenAI.

To use its AI agent building platform, enterprise developers still need to have a comprehensive technical background, OpenAI said. Developers can use any of OpenAI's models to power the agents they build, including the model that powers Operator, and can incorporate its file and web search tools, the company said.

Fintech giant Stripe tested OpenAI's agent-building platform and built a prototype of an agent that can read a sales-tracking spreadsheet a small-business owner might create. The AI agent can then create invoices and send them out to customers using Stripe's own agentic platform, said Jeff Weinstein, a Stripe product lead.

Aaron Levie, chief executive of Box, said the cloud-storage company is making it easier for its own business customers to build custom agents, using OpenAI's platform, that link up to their data stored in Box.

"Every single enterprise conversation I'm on, they're now talking about agents," Levie said. "Twelve months ago, only about three to five percent of our customers could even process what we were talking about."

News Corp, owner of The Wall Street Journal, has a content-licensing partnership with OpenAI.

Write to Belle Lin at belle.lin@wsj.com

(END) Dow Jones Newswires

March 11, 2025 13:00 ET (17:00 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.