What's up: Three questions for BofA's tech chief; Meta restructures AI teams; AI concens rile markets; Is American capitalism in retreat?

Good morning. After the billions poured into backing artificial intelligence efforts–from data center builds to payrolls for 20-something AI wunderkinds–it can sound almost sacrilegious to voice caution around AI spending.

But here’s Bank of America’s newly appointed Chief Technology and Information Officer Hari Gopalkrishnan telling WSJ Leadership Institute’s Isabelle Bousquette that he aims not to overspend or overengineer with unnecessary features.

“This isn’t cheap stuff. You’re using GPUs that are not commodity,” he said, referring to Nvidia’s pricey AI chips. “You’re using money and resources in the cloud that are not cheap,” he said.

To be sure, Gopalkrishnan can tap the bank’s $13 billion tech budget for 2025, with $4 billion set aside for developing new technologies like AI. But the point remains that when it comes to AI, as with any technology, he is trying to build prudently. “It’s very easy to let the hype get away from you,” he said.

Below, Gopalkrishnan recaps an episode involving the bank’s popular Erica AI chatbot where he did not let hype stand in the way of delivering the right user experience.

As it turned out, customers did not want to talk out loud to their banking app about money issues.

Gopalkrishnan sat down with the Wall Street Journal Leadership Institute’s Isabelle Bousquette to talk about how to track this feedback early on and why it’s important to pivot early.

WSJLI: What was your thinking with the initial voice integration into Erica?

Gopalkrishnan: When we first built Erica, we thought: hey, voice is going to be just as important as text. People are going to speak into the machine as much as they will type and text into it. We realized pretty quickly that 90% of the people were just typing into it. And it kind of makes sense because people probably aren't comfortable yelling out like, ‘hey, how much is my balance?’ So why overengineer that? Let's worry about the next set of integrations we can do.

WSJLI: What do you mean by overengineering it?  

Gopalkrishnan: You'd be amazed how much time we were spending thinking about how we should process voice. We were like, whose voice should it be? And how do you tune for voice? Should someone stop speaking and should we detect pauses in their speech before we start processing it? Should we have them press a button to say I have stopped speaking?

So we stopped focusing as much of our efforts on trying to make the voice experience that much better because people just weren't using it.

WSJLI: How did you catch onto that so early on? What’s your process for rollouts and feedback?

Gopalkrishnan: We build it, then we would first roll it out to our associates, and we would get feedback from them. Then we would do a launch in a single state. And we would pick a small state because we want to minimize the blast radius. And we would look at feedback from the first state. Then we'd say, let's launch to the next five states.

What happens each step of the way is we're processing feedback. In this case, actually, interestingly enough, it was after the second state rollout, I believe.

No one told us they didn't like voice. We looked at the usage and said: you know what, people just aren't using voice (because we can measure all that). Let's not overinvest here. People were still happy and we could focus our attention on other things.

A drop in megacap technology stocks pushed major indexes into the red Tuesday. The eight companies valued at more than $1 trillion lost a combined $385 billion in market capitalization. Palantir, which trades at more than 200 times projected earnings, led S&P 500 decliners with a 9.4% drop. Oracle dropped 5.8%.

Amid the declines, the FT detects rising concerns with the state of the AI boom, pointing to remarks by OpenAI CEO Sam Altman last week about the possibility of an AI bubble.

Closer to home and more concerning was the release this week of a Massachusetts Institute of Technology report finding that the vast majority of AI pilot programs launched to drive revenue growth fall flat, according to Fortune. 

One trader tells the FT that the report "is spooking people."

On Tuesday, Meta announced internally that it is splitting its AI division — which is known as Meta Superintelligence Labs — into four groups, two people with knowledge of the situation tell the New York Times. One group will focus on A.I. research; one on a potentially powerful A.I. called “superintelligence”; another on products; and one on infrastructure such as data centers and other A.I. hardware, they said.

An Oregon man is accused of operating one of the most powerful attack ‘botnets’ ever seen. Federal prosecutors have charged a 22-year-old Oregon man with operating a vast network of hacked devices that has been blamed for knocking Elon Musk’s X social-media site offline earlier this year. The network, known as Rapper Bot, was operated by Ethan Foltz of Eugene, Ore., the prosecutors said Tuesday.

Rapper Bot was made up of tens of thousands of hacked devices. It was capable of flooding victims’ websites with enough junk internet traffic to knock them offline, an attack known as a distributed denial of service, or DDoS.

Apple is expanding iPhone production in India for US-bound new models. The expansion covers five factories, including a pair of recently opened plants, as it seeks to lessen its reliance on China for US-bound models, Bloomberg reports. WSJ recently reported that Apple got the jump on Trump’s recent tariffs by moving some of its iPhone production to India years ago.

OpenAI is rolling out its cheapest ChatGPT plan at roughly $4.60 per month in India to chase growth. ChatGPT Go is designed for Indians who want greater access to ChatGPT’s advanced capabilities at a more affordable price, the ChatGPT maker said in a statement, according to Reuters.

Nvidia is working on a new AI chip for China that outperforms the H20. The chip giant is developing a new AI chip for China based on its latest Blackwell architecture that will be more powerful than the H20 model it is currently allowed to sell there, two people briefed on the matter tell Reuters.

The latest mattress cover favored by tech elites costs $5,000 and that’s not including required app subscription. “Sleep fitness” startup Eight Sleep said it had raised $100 million in a Series D funding round, partly to develop an AI agent designed to interpret users’ biometric data and make temperature, elevation, sound and other environmental tweaks while users are sleeping.

President Trump signaled on Tuesday that the U.S. is prepared to use air power to support a European security force in Ukraine but ruled out deploying American ground troops. (WSJ)

McDonald’s is lowering the cost of its combo meals, after consumers were left sticker-shocked by Big Mac meals that climbed to $18 in some places. (WSJ)

In return for billions of dollars of investment pledges and promises to buy more American goods, U.S. allies in Asia and Europe say President Trump agreed to lower tariffs on key exports such as cars and steel. Weeks later, they are still waiting. (WSJ)

This September, The WSJ Technology Council will convene senior technology leaders to examine how AI is reshaping organizations—redefining teams and workflows, elevating technology’s role in business strategy, and transforming products and services. Conversations will also delve into AI’s evolving security risks and other fast-emerging trends shaping the future.

Select speakers include:

Carolina Dybeck Happe, COO, Microsoft
Severin Hacker, CTO, Duolingo
Yang Lu, CIO, Tapestry
Tilak Mandadi, CTO, CVS Health
Helen Riley, CFO, X, Google's Moonshot Factory

September 15-16 | New York, NY
Request an invitation | Participants and program

The WSJ CIO Journal Team is Steven Rosenbush, Isabelle Bousquette and Belle Lin.

The editor, Tom Loftus, can be reached at thomas.loftus@wsj.com.