Salesforce: $300 Million for Anthropic Tokens

Marc Benioff, the CEO of Salesforce, recently praised AI-based coding agents, particularly those developed by Anthropic. According to him, these tools have significantly improved the efficiency and speed of product iterations within his company. During an appearance on the All-In podcast, aired last Friday, Benioff announced that Salesforce plans to invest $300 million in Anthropic tokens this year. These tokens, which are units of information processed by AI models, are used by companies to bill users and large organizations.

Benioff expressed his admiration for the capabilities of coding agents, stating, “These coding agents are incredible. Anthropic is incredible.” He added that this $300 million investment in Anthropic tokens is aimed at making coding more affordable and efficient. He also emphasized that AI agents have enabled Salesforce to achieve unprecedented levels of efficiency, particularly in service, support, distribution, and marketing.

Last August, Benioff had already announced that the use of AI agents allowed Salesforce to reduce its support staff from 9,000 to 5,000 employees. This significant reduction underscores the impact of AI technologies on the company's productivity and operational efficiency.

Furthermore, Benioff revealed that Salesforce is working on innovations aimed at facilitating coding within Slack, the productivity platform acquired by Salesforce in 2021. He stated, “We are even working on technology inside Slack to make coding easier for everyone.” Although he has not yet disclosed all the details, he promised interesting developments to come.

However, Benioff clarified that even though Salesforce plans to spend $300 million on Anthropic this year, he does not believe that every token used by an employee of the company necessarily needs to be directed toward a cutting-edge model like Claude from Anthropic. He suggested the creation of an “intermediate layer” that could determine which token inputs should go to Anthropic and which could be processed by smaller models. This approach could optimize resource use and maximize the efficiency of Salesforce's investments in AI technologies.