During the PC boom of the 1990s, Microsoft introduced Clippy as a friendly face to help users hone their word-processing skills. An animated paper clip with round cartoon eyes and expressive eyebrows floating over a sheet of yellow legal paper, Clippy would frequently and spontaneously pop out of the corner of the screen to provide advice and tips on saving files and using good grammar. Clippy was also almost universally hated, his shifting eyes and unsolicited feedback providing more irritation than comfort.
Salesforce CEO Marc Benioff believes Microsoft is on its way to a repeat act of creating a loathsome program meant to assist users in Copilot, the tech company’s foray into AI-powered assistance.
“We all know now that Microsoft Copilot is basically the new Microsoft Clippy, that customers have not gotten value from it,” Benioff said in a Bloomberg interview Thursday during Salesforce’s Dreamforce conference in San Francisco.
Microsoft has invested over $13 billion into OpenAI, the large language model powering Copilot. Launched in November 2023, Copilot promised to help expedite menial office tasks like summarizing Teams meetings and draft emails. Chemical company Dow plans to introduce the assistant to its 35,900 employees by the end of the year after a successful pilot test, but others aren’t sold on the program’s $30-per-month costs—particularly after Microsoft has had to iron out Copilot Designer’s issue of generating inappropriate images and its Recall feature that took excessive screenshots, jeopardizing user privacy.
Of course, Benioff’s criticisms of Copilot are in Salesforce’s best interest. The software giant announced a “hard pivot” to Agentforce, its own AI assistant program. The program, introduced last week, promises to integrate with hundreds of other applications to streamline business operations. It’s part of a growing movement of implementing AI agents over copilots to take tech assistance one step further.
“This is like we’re selling science projects to companies, and they’re tired of it. They have not gotten the value, which is why you see these customers so excited they’re coming here,” Benioff said. “And they’re getting immediate value, and they’re able to use this next generation platform to do everything that they’ve heard that’s possible.”
Benioff is banking on Agentforce’s success, which is already being piloted by a few dozen companies like Wiley, Open Table, and Fossil. It hasn’t been all smooth sailing for the program so far. Andreessen Horowitz challenged Salesforce’s ability to create a compelling AI assistant, saying in a July blog post, “We believe AI will so fundamentally reimagine the core system of record and the sales workflows that no incumbent is safe.”
Salesforce and Microsoft did not immediately respond to Fortune’s request for comment.
AI skepticism
AI skeptics are bracing themselves for the future of tech assistants. With upwards of $200 billion expected to be invested into AI by 2025, according to Goldman Sachs, some in the industry fear AI will not be able to keep its promise of revolutionizing how people engage with tech, threatening the lofty valuations of many tech startups.
Beyond a potentially muted moment in the AI hype cycle, firms also have to contend with mounting concerns over privacy and safety.
“Tech companies are putting this deeply flawed tech in the hands of millions of people and allowing AI models access to sensitive information such as their emails, calendars, and private messages,” Melissa Heikkilä wrote in the MIT Technology Review last October.
After introducing updates and AI agent features to Copilot last week, Microsoft doubled down on its promise of privacy when using Copilot’s chatbot feature. Jamie Teevan, Microsoft’s chief scientist, said Copilot customers will see greater value in the product once they optimize how to use it. She added the company is keen to avoid the pitfall of a nagging assistant like Clippy when continuing to develop its AI agents. Benioff isn’t convinced.
“Microsoft has deceived customers with their AI strategy, they don’t need to DIY it,” he said. “We build it into our platform, customers shouldn’t be forced to train and retrain their models.”