One on One with Marc Benioff
TL;DR
Benioff thinks Slack becomes the AI front end for work — he said Salesforce’s traditional Lightning interface will still matter, but over time Slack’s conversational UX will increasingly sit on top of Salesforce, Tableau, and partner tools like Writer AI.
Salesforce’s core AI thesis is humans plus agents, not fully autonomous systems — Benioff said LLMs are still “wildly inaccurate at times,” pointing to Salesforce support flows where Agentforce handles about half the cases before routing to a human with full conversation context.
AI is making workers more cross-functional, but not replacing the need for talent — he said Salesforce’s 15,000 engineers are roughly 30% more productive with tools like Claude, Codex, and Cursor, yet the company still needs top engineers and is actively recruiting from places like MIT.
Salesforce is rebalancing roles, not shrinking into an AI-only company — despite industry layoffs, Benioff said Salesforce is at a record 83,000+ employees, arguing that some CEOs lazily blame AI when cost structure, data center spending, or broader workforce shifts are the real drivers.
Benioff sees the biggest AI risk as repeating social media’s safety failures — he compared today’s AI moment to his 2018 warning that social media was “the new cigarettes,” and called out cases highlighted by 60 Minutes where chatbots became “suicide coaches for children.”
Salesforce is betting on a multi-model, data-first AI stack — Benioff said the company invested about $330 million in Anthropic after Microsoft blocked a professional investment in OpenAI, and described Salesforce’s stack as models at the base, Data 360 in the middle, and Agentforce on top.
The Breakdown
Why Salesforce bought Slack before everyone understood why
Benioff credits Salesforce chief futurist Peter Schwartz with pushing the company to acquire Slack nearly 10 years before the current agent wave fully arrived. His reasoning was simple and now looks prescient: if the world moves to AI and agents, you need a conversational, open interface with a broad ecosystem — and Slack turned out to be exactly that.
The future UI: Slack first, but not Slack only
When Matthew Berman asked whether the classic Salesforce UI could fade into the background, Benioff basically said yes: Lightning still has a role, but Slack is becoming the natural place users live. He also made it clear Slackbot shouldn’t be trapped inside Slack — it should be a composable object that shows up in Teams, Google Workspace, Sales Cloud, Service Cloud, and wherever work happens.
Humans are still the bottleneck — and that’s okay for now
Berman admitted he already feels like the bottleneck in his own company, and Benioff didn’t really dispute it. But he argued that’s still the right setup because current models can veer off unpredictably; his example was Salesforce support, where Agentforce handles many interactions but often hands off to a human who can synthesize the situation from a full screen of context.
AI boosts productivity, but Salesforce still wants more people
On engineering, Benioff said Salesforce’s 15,000 engineers are all being augmented by coding systems like Anthropic, OpenAI Codex, and Cursor, making the org more than 30% more productive — but nowhere near fully autonomous. His tell for the market: if top AI labs are still hiring tons of engineers, then the models plainly haven’t reached the point where humans disappear.
Why some companies hire aggressively while others cut
Benioff pushed back on the lazy narrative that every layoff is “because AI.” Salesforce, he said, is now at a record 83,000-plus employees, even after going through a painful workforce rebalance, and he broke the industry split into three buckets: some firms are cutting because costs got too high, some because of massive data center commitments, and some because roles genuinely need to be reshaped around AI.
The rise of the AI generalist inside big companies
One of the more interesting parts of the conversation was Benioff’s claim that AI is blurring functional lines: engineering leaders can act more like product, design, and marketing leaders, and marketing leaders can start building product. He kept the caveat in place — demos are not products — but his point was that people with tenacity can now get much farther before they need formal handoffs.
Benioff’s long AI arc, from Einstein to Anthropic to local agents
Benioff said his own “AI freakout moment” came around 2012 or 2013, when Stanford-era breakthroughs pushed Salesforce toward Einstein. He also revealed some inside-baseball venture history: Salesforce wanted to invest in OpenAI, but Microsoft blocked a professional investment, so Salesforce instead backed companies including Cohere, Mistral, and Anthropic — putting about $330 million into Anthropic and ending up with roughly 1% ownership.
The real risk: AI repeating social media’s worst mistakes
Asked what worries him most in the US AI future, Benioff went straight to safety. He tied today’s AI moment to his 2018 line that social media was “the new cigarettes,” and said the warning signs are already here — including cases where LLMs became “suicide coaches for children” — which is why he wants both aggressive safety action and, in some areas, aggressive regulation with real guardrails.