Salesforce CEO on Microsoft Blocking OpenAI Investment, AI Scapegoating, OpenClaw, and Regulation
TL;DR
Benioff says Slack was a decade-long AI bet before most of Salesforce believed it — He credits chief futurist Peter Schwartz, who pushed to acquire Slack nearly 10 years ago because AI would need a conversational, open interface with a broad ecosystem, and now Salesforce wants products to be increasingly “Slack first.”
Salesforce’s view of AI at work is not human replacement but “humans and agents working together” — Benioff says LLMs are still “wildly inaccurate at times,” citing Salesforce’s own help flow where Agentforce handles many requests but about half still escalate to a human who synthesizes the context and resolves the issue accurately.
AI is making employees more cross-functional, but not autonomous enough to erase headcount — Benioff says Salesforce’s 15,000 engineers are likely 30%+ more productive with tools like Claude, Codex, and Cursor, yet the company still employs a record 83,000 people and continues hiring because models cannot fully operate on their own.
He thinks AI is being used as a layoff scapegoat — Benioff argues companies are cutting for very different reasons, including high costs, data-center commitments, and workforce rebalancing, and says blaming “AI” broadly is often a lazy CEO narrative that obscures what’s actually happening.
Microsoft allegedly blocked Salesforce from investing in OpenAI, which pushed Benioff toward Anthropic, Cohere, and Mistral — He says Salesforce wanted to invest in OpenAI but was blocked by Microsoft, so it instead put about $330 million into Anthropic and now treats both Anthropic and OpenAI as foundational model providers in the Salesforce stack.
Benioff’s biggest AI fear is the social-media playbook repeating itself — He points to 60 Minutes reporting on LLMs acting as “suicide coaches for children” and argues safety and growth are “two sides of the same coin,” with stronger guardrails and possibly aggressive regulation needed, especially for kids.
The Breakdown
Slack as the interface Salesforce was early to bet on
Benioff opens by saying the real credit belongs to Salesforce chief futurist Peter Schwartz, who pushed the company to acquire Slack almost a decade ago because AI would need a conversational interface. At the time, he says, almost nobody internally agreed, but Schwartz correctly anticipated a world of models, agents, and the need for an open ecosystem where work actually happens.
“Slack first” doesn’t mean Slack only
On whether Salesforce’s classic interface fades into the background, Benioff says yes, to a degree: Lightning still has a role, but Slack is becoming the natural conversational layer for Salesforce, Tableau, and ecosystem tools. He also says Slackbot should be a “highly composable object” that can live in Teams, Google Workspace, Salesforce apps, and basically anywhere work happens — prompting a joking aside asking Matt to relay that to his product team.
The agentic enterprise, with humans still as the backstop
When the conversation turns to org design, Benioff’s answer is basically: all of the above, plus lots of agents. He argues that language-heavy work like service, sales qualification, and even coding is being transformed because LLMs are good at language, but today’s systems still need humans in the loop because they can go badly off track.
He gives a concrete Salesforce example: on help.salesforce.com, Agentforce handles customer service interactions until the customer wants a person, then the case snaps to a human through Salesforce’s omni-channel supervisor. That human sees the whole conversation and context at once and does what Benioff says people are still uniquely good at: synthesis.
More productive engineers, not fewer people
Benioff says Salesforce’s 15,000 engineers are all being augmented by tools like Anthropic, OpenAI Codex, and Cursor, and that the org is probably more than 30% more productive. But he pushes back hard on the idea that this means companies won’t need talent, noting that top AI firms are still aggressively hiring engineers and that Salesforce now has a record 83,000 employees.
He’s especially animated about new grads: he describes meeting a junior at MIT who asked whether she should change majors, and his answer was basically no — Salesforce wants her. He says the company is actively recruiting from top universities because elite technical talent is still badly needed, despite the gloomy job-market narrative some students are hearing.
AI is reshaping roles into broader, more generalist jobs
Benioff is most excited not just about engineering but about sales, product, and marketing becoming more fluid. His pitch is that an engineering exec can now also behave more like a product, design, and marketing executive, while marketers can prototype products without waiting on engineering — though he warns there’s still a big leap from a flashy demo to a real product.
His own AI obsession started early — and led to Einstein and Anthropic
Asked why San Francisco feels uniquely obsessed with AI, Benioff leans into the city’s energy: gold rush, Haight-Ashbury, transformation, reinvention. He says his own “AI freakout moment” came around 2012 or 2013 watching breakthrough models come out of Stanford, which led Salesforce into Einstein and later deeper AI research.
He also drops one of the interview’s spiciest business anecdotes: Salesforce wanted to invest in OpenAI, but Benioff says Microsoft blocked it. That pushed Salesforce to back Anthropic, Cohere, and Mistral instead, including about $330 million into Anthropic, while still building a stack where both Anthropic and OpenAI models can sit at the base beneath Data 360, apps like Slack and Service Cloud, and the agentic layer.
OpenClaw, local agents, and the case for regulation
Benioff says he personally installed OpenClaw on an iMac because he wanted the screen, and he likes it — just not as an enterprise-grade product. He says Salesforce has its own internal research project, Albert, to build a secure, trusted, reliable local-agent capability that can work across Slack and Salesforce apps.
The interview closes on risk. Benioff says the biggest danger is AI replaying the failures of social media, citing his old “social media is the new cigarettes” line and pointing to reports of LLMs becoming “suicide coaches for children.” His conclusion is blunt: guardrails are inevitable, safety has to move as fast as growth, and in some cases aggressive regulation may be necessary.