Marc Benioff on Agentforce, AI Workforce Plans and the Beginner’s Mind

Salesforce founder and CEO Marc Benioff describes AI agents as the company’s central priority in this December 2024 interview with Harry Stebbings. He discusses Agentforce, workforce planning, model competition and leadership. His product claims and forecasts reflect his commercial interest as Salesforce’s CEO.

Benioff says he has begun listening to more podcasts and wants to become a better guest. Stebbings describes effective interviews as a combination of storytelling that draws people in and clear lessons listeners can use.

A conversation with Nvidia’s Jensen Huang reminded Benioff of the phrase “no mud, no lotus,” which he attributes to his teacher Thich Nhat Hanh. He interprets it as a reminder that difficulty can precede growth and says meditation helps him return to a beginner’s mind.

He points to Salesforce’s difficult period two years earlier, including leadership departures and financial issues, followed by product and financial changes. He cites the company’s guidance at the time, including about $38 billion in revenue and $12.9 billion in cash flow; guidance is a forecast rather than a completed result.

Benioff describes the subsequent stock recovery and launch of Agentforce as evidence of that transformation. His description of the product as exceptionally successful is a company leader’s evaluation, not an independently verified ranking.

Asked whether AI agents help or threaten Salesforce, he argues that the company’s applications, data and workflows create a strong foundation. He cites enterprise transactions and customer scale, but the interview does not provide a comparable industry benchmark for his claim of leadership.

He describes a planned agent-based customer experience with Heathrow as an example. The deployment and its future effect on travelers are described as a project and expectation, not as a completed outcome for every passenger.

Benioff says Salesforce combines its sales, service, marketing, commerce, analytics and collaboration products with Data Cloud and an agent layer. He argues that access to data, metadata and workflows helps enterprise AI produce accurate results; no independent accuracy test is supplied.

He says Agentforce launched in late October and that Salesforce sold about 200 deals in the last days of the quarter. He expects thousands more in the following quarter. The latter is a forecast, and deal count alone does not establish revenue or usage.

Benioff describes replacing Salesforce’s support infrastructure with an agent-based system and says the company intends to rebalance human and digital work. His account is a company-reported implementation, not an independent measurement of outcomes.

For the 2025 plan discussed in December 2024, he says Salesforce was not planning to add software engineers, expected fewer support engineers and anticipated adding roughly 1,000–2,000 salespeople. He claims engineering productivity rose more than 30% with Agentforce and other AI tools; the interview does not define the measure or establish causation.

Agentforce uses consumption pricing, he says, starting at roughly $2 per conversation and negotiated by volume, while human users are generally priced per seat. These are the terms he describes at the time, not a quote for every customer.

Benioff argues that contemporary large language models are maturing and becoming commodities. He expects different specialized models to matter more. This is his forecast at the time, not an established limit on model capabilities.

He says Salesforce develops models and also uses models from other providers, choosing them for the task. He claims two Salesforce models were among a top five but gives no benchmark or context to verify the ranking.

He discloses that Salesforce has invested in model companies including Anthropic, Mistral and Cohere, and that he personally invests in AI companies. He also discusses his investment in You.com. These financial relationships are relevant to his comments on the model landscape.

Asked to choose among OpenAI, Anthropic and xAI at the valuations proposed by Stebbings, he says he would not necessarily invest in any at those prices. He believes providers will need distinctive products or specialties. This is a hypothetical opinion, not a recommendation or a verified valuation comparison.

His primary concern is explaining the potential value of Agentforce to customers. He says Salesforce will use it internally in support and sales to demonstrate what is possible, a strategy that also promotes the company’s own product.

He says Salesforce had about 9,000 support staff and expected fewer in support as agents took on more work. Asked whether people would be moved or removed, he says the vast majority could be moved into sales or other opportunities. These are plans and expectations, not a verified outcome for individual employees.

Reflecting on layoffs roughly two years earlier, Benioff says he disliked them but considered restructuring necessary. He argues layoffs can be a tool for leaders preparing a company for change. His view does not establish that a particular layoff was necessary or its effects on employees.

Benioff recalls Steve Jobs showing him iPad prototypes and saying Apple’s top team focused on one thing at a time. He uses that lesson to explain why he is concentrating on Agentforce, while acknowledging that leaders must keep other divisions engaged.

He asks each product organization how an agent layer could improve its offering. His ambition is to align Salesforce’s employees and customers around Agentforce; that is a management and product strategy, not an independent assessment of customer adoption.

Asked about mistakes, he returns to the idea of a beginner’s mind: leaders should be willing to reconsider assumptions as technology, politics and markets change. He says a major event can require a rapid change in focus.

Before Dreamforce, the team had planned a demonstration around Gucci and its point-of-sale work. Benioff says early customer feedback on Agentforce made him conclude the new product was further along than he had realized.

He recalls changing the conference message to focus on Agentforce within about two weeks, followed by production delivery in October. He cites the initial deals as evidence of momentum; future demand and value remain to be demonstrated.

Benioff says launches, earnings and travel can be exhausting, and he continually asks whether the company has the ideas, people and resources needed to execute. He describes sustained effort as part of leading Salesforce over 25 years.

Asked whether he would undertake the journey again, he instead tells a story about Nvidia’s GPU work and Huang’s decision to focus on deep learning. He praises the ability to recognize a technology shift and commit a company to it.

He says Salesforce adopted deep learning but did not pivot as fully as Nvidia did. His broader lesson is that leaders need room to slow down, question assumptions and recognize when circumstances warrant a change in direction.

Stebbings asks about Marc Andreessen’s claim that technology companies were coerced into censorship. Benioff says he was surprised by the discussion but had not seen such conduct at Salesforce and does not run a social-media platform. The interview does not establish the broader allegation.

He recalls Salesforce considering a purchase of Twitter and says he is glad the deal did not happen. His interest was in using its feed as a distribution platform for applications, inspired by Apple’s HyperCard.

Benioff connects that idea to his early experience writing games and building application-development tools. He sees Salesforce and agents as a continuation of his interest in making applications easier to build and deploy.

Asked about Musk and DOGE, he says he hopes for more fiscal responsibility and a balanced US budget, which he considers a bipartisan issue. This is a policy preference expressed before the administration took office, not a prediction of a particular result.

In quick fire, Benioff names Louis Vuitton as a favorite consumer brand and notes it is also a Salesforce customer. He says he changed his mind about how quickly AI could become useful after seeing Agentforce, reflecting his company’s commercial interest.

He says agents need to become more multimodal rather than primarily text-based. He also calls for education about both the benefits and risks of AI, arguing that what people do with technology determines its social effect.

He does not expect artificial general intelligence in 2025 and predicts that contemporary LLMs will mature and commoditize during that year. Both are dated forecasts, not conclusions established by the interview.

Benioff closes with Salesforce’s early commitment to put a portion of equity, profit and employee time toward philanthropy. He reports charitable giving, free services and volunteer hours over 25 years and argues that businesses can be a platform for positive change. These are company-reported figures in the conversation.