A version of this story was originally published in AI, Tech & Money, a free weekly newsletter packed with news and exclusive analysis on all things tech, platforms, and deals in advertising and media. Sign up for it here*.*
Welcome back to AI, Tech & Money. This is Trishla Ostwal, ADWEEK’s AI reporter, writing after two very long days in San Francisco, where Salesforce threw its biggest party: Dreamforce. Gwen Stefani, Katy Perry, and Usher were on the bill.
So were OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Nvidia CEO Jensen Huang, who took to the stage to debate a more consequential question: Does AI development need to slow down?
More on that in a minute.
First, there was a little friction between Microsoft and Anthropic. Microsoft AI chief Mustafa Suleyman is taking issue with Anthropic’s approach to training Claude on questions around consciousness and welfare interests. Suleyman called for removing speculation about AI consciousness from training documents, arguing that it could make future AI systems harder for humans to control.
That puts Anthropic in an awkward position: Amodei has been among the most vocal in calling for more caution around increasingly powerful AI. Now, one of his peers is arguing that some of Anthropic’s methods could create problems of its own.
Amodei has already given us plenty to chew on before he even spoke on stage at Dreamforce. Just days before the event, he published a 3,800-word essay arguing that AI is advancing too swiftly for safety research to keep pace, and calling for independent evaluators to get permanent access to frontier models to verify companies are following their own safety commitments. Altman agreed with Amodei on X .
“It’s very tempting to attack your competitor and say, ‘These guys are unsafe,’” Amodei said at Dreamforce. “The more responsible way to respond to it is to say, ‘Let’s look at our own record.’”
Nvidia’s Huang offered a contrarian take.
He argued that AI safety is fundamentally an engineering problem, not a legal one, and said new laws and regulations aren’t necessary. Companies, he said, should determine whether their products are safe before releasing them.
Nvidia is financing $105 billion for OpenAI’s data center and is reportedly investing $10 billion in Anthropic’s IPO. Both AI labs are among the big buyers of Nvidia’s chips.
So, you have AI’s biggest players effectively in opposing camps about AI safety.
The debate is also playing out in Washington. President Donald Trump has rejected calls from AI doomers for an AI slowdown, with concerns of losing the race to China.
“The only control or “guardrails” that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT,” Trump wrote on Truth Social.
Finally, a quick reminder that entries are now open for ADWEEK Tech Challengers to Watch, a new award dedicated to spotlighting up-and-coming tech companies making waves in adland. Learn more and enter here.
THE TOP LINE
- Sam Altman said AI companies should expect accidents as increasingly capable models are deployed.
- At our Brandweek event in Atlanta, Wikipedia dished on how it plans to survive the answer engine apocalypse.
- And LA28 is bringing a golden opportunity to brands with new offerings and massive viewership.
- Global ad revenue will reach $1.3 trillion by year’s end —an 11% increase year-over-year, per Madison & Wall.
THE QUANT
Comfort with using AI in the customer journey falls sharply as autonomy increases. According to research by DEPT’s survey of 2,606 consumers, 65% of people who’ve used AI for shopping are comfortable with AI through the recommendation stage. But only 15% will let it move into preparation or transaction, and just 7.5% are comfortable with fully autonomous actions. The same report found that 56% of people who recently used AI while shopping chose a different brand than they’d originally planned.
DEEP LEARNING: AI Agents Get a Reality Check: AT&T, Crocs Want Automation but Still Need Humans in the Loop
Marc Benioff wants AI agents to take on a lot more work that humans do.
But Salesforce’s adoption numbers suggest that vision is still a work in progress. Roughly 30,000 of its customers use Agentforce—a feature launched in 2024 that builds digital workers to handle business tasks without constant human direction—according to Benioff. That’s just around 20% of Salesforce’s roughly 150,000 customers.
Talking to brands on the ground, I kept hearing a more cautious version of Benioff’s vision: Sure, automate more. Just don’t automate everything.
I toured a section of the conference hall branded Agentic Enterprise City to look at some of those early adopters, and the AI agents Salesforce is pitching as the future of enterprise software.
I spoke with executives from AT&T, Southwest Airlines, CVS, and Crocs about how they are putting AI agents to work. The use cases themselves weren’t particularly novel. But the details of how these brands are deploying them—and perhaps, more importantly, where they aren’t—offer a more grounded picture of what companies actually want from AI agents.
AT&T, for example, is using AI to help cut the time it takes employees to upgrade a customer’s phone from about 30 minutes to roughly 10 minutes, according to John Miller, VP of consumer and business solutions.
When I asked Miller where AI vendors can be unrealistic about what brands actually want from the technology, he pointed to a familiar tension: Efficiency isn’t the same as eliminating human interaction.
“AI can make us much more efficient. But we really want to make sure that there’s still that human interaction,” he said.
That distinction shapes where AT&T is drawing the line. The telecommunications company is stopping short of using AI agents to handle cancellations, leaving its human employees to understand why a customer is leaving and potentially resolve the issue.
“We want to be able to understand from our customers’ perspective why they’re cancelling, and then be able to help,” Miller said.
Southwest Airlines is taking a similar approach, using agents to handle servicing tasks via chatbots, such as flight changes and completing transactions, while reserving human employees for what Megan Rauber, manager of technology at Southwest, described as “right here, right now” problems—situations where customers need immediate help and would benefit from the human hospitality that requires empathy and assurance.
Crocs is deploying AI agents both as a consumer-facing chat and voice bot called Rivet, and internally, where agents can help developers create promotions or categorize products on its websites.
But even as Crocs expands those use cases, it still has human hands on the tiller.
“There still needs to be a human in the loop as we’re still learning about capabilities and complexities of automation,” said Feliz Papich, SVP of digital technology and experience at Crocs.
Papich said some of that automation complexity gets lost in the rush to build increasingly sophisticated interfaces with AI vendors. The harder problem, she said, is making sure the data underneath those interfaces is good enough to deliver what people expect.
“Data is such an important conversation, respective to AI,” Papich said. “Sometimes that gets lost because you’re talking about the interface, and want to make the agent that delivers against expectations. The data foundation necessary to make sure that the quality is there—that complexity of that is not always communicated.”
What People Are Saying
DEALS, DEALS, DEALS
- OpenAI weighs new funding round at $1.5 trillion valuation before IPO.
- Health support app Headspace is being acquired by Sword Health for nearly $300 million .
- Bain Capital Ventures raises $1.6 billion to back startups focused on artificial general intelligence .
REVOLVING DOOR
- Mark Weinstein joins Target as CMO and guest experience officer.
- PebblePost names Magnite CPO Adam Soroca as its CEO .
- Maria Bartiromo is no longer with Fox News Media .