The Morning Download: Agents Rewrite the AI Stack

Good morning. The AI computing boom was supposed to be all about GPUs, but agents are rewriting the rules across the entire AI stack from chips to software.

Agents also promise to transform the workforce, from the CIOs deploying autonomous AI to a new class of executives now managing humans and AI agents side by side. More on that below.

But first, Arm Holdings. The British chip designer has long made its money licensing its CPU architecture to bigger companies, Now it's launching its own chip. The reason? AI, of course.  

The WSJ reports that Arm this week unveiled its first purpose-built AI chip, an "agentic CPU" it claims is twice as efficient as rivals. Meta Platforms has signed up as the lead partner and first big customer for the chips. 

Here's the WSJ's Dan Gallagher on how CPUs (and Arm) are suddenly relevant again in the age of AI agents:

AI computing needs are shifting to the types of central processing chips—or CPUs—that are needed to power inferencing, where AI models produce answers to user queries, rather than model training.

The adoption of AI agents is expected to push the need for inference even further. But AI data centers still demand a lot of energy, so Arm’s roots in designing CPUs that have to run on batteries is a big advantage. Mark Lipacis of Evercore ISI said Arm’s designs make it perfectly matched for a market needing more power-efficient chips to run agents. “Agentic to Arm is like AI to Nvidia,” Lipacis wrote in a report Wednesday.

Arm is just one piece of the puzzle. Here's what else is happening across the agent stack and what it means for the CIOs who have to build on top of it.

OpenAI backs new AI startup seeking bot army breakthroughs. Founded by two 23-year-old researchers, Isara's technology can coordinate thousands of AI agents simultaneously; an early demo used roughly 2,000 agents to forecast gold prices. The San Francisco startup, valued at $650 million, plans to initially target investment firms with predictive modeling tools.

"Deep Seek of the West" eyes $25 billion valuation. Nvidia-backed Reflection AI is raising $2.5 billion in a deal that would add firepower to efforts building open-source U.S. AI models that can rival China's offerings, WSJ reports. Reflection is part of a new consortium of startups, called the Nemotron Coalition, that Nvidia has assembled to develop open AI models. 

China grounds Manus. China has told two co-founders of artificial-intelligence startup Manus not to leave the country while authorities review the company’s $2.5 billion sale to Meta. The AI agent maker last year relocated operations and staff to Singapore, a move that angered Chinese regulators, the WSJ reports.

Legal AI startup Harvey raised $200 million in a funding round that values the company at $11 billion, Bloomberg reports. Sequoia and GIC led the round.

A new survey from Ernst & Young finds that consumer adoption of AI is moving faster than most enterprises realize, and for CIOs, that gap represents both a challenge and an opportunity.

EY's 2026 AI Sentiment Survey, based on responses from more than 18,000 people across 23 countries, found that 84% used AI in the past six months and 16% have already used AI that acted without human intervention (think AI agents managing investments or hailing a self-driving taxi without a human in the loop).

In markets like India, China and the UAE, that figure for autonomous AI climbs to 24%.

Even respondents who have not yet used autonomous AI are ready to do so. All told, 36% said they would prefer AI to automatically apply discounts at checkout.

At the same time, risk awareness is high: Respondents indicated concern over AI systems being compromised and their own ability to tell real content from AI-generated content.

But that anxiety isn't slowing anyone down. "Instead, they are shaping expectations for how autonomy should be introduced," according to EY.

"When you talk to consumers, day-to-day people looking for efficiency and knowledge, that adoption is pretty high," Raj Sharma, a global managing partner for growth and innovation at EY, tells the WSJ Leadership Institute.

For CIOs, waiting isn't a strategy. "As you are rethinking your new products, as you're thinking of what you could offer to them, know that your consumer sentiment is: if it's an AI-powered product, the adoption will be there."

Trust, he says, built through the actual experience of interacting with AI, is crucial.

His advice to CIOs: commit to reimagining, not just optimizing, the way work gets done. "Whenever you design something, trust has to be an important part of it."

Social media may have met its 'Big Tobacco' moment. Back-to-back jury verdicts holding Meta Platforms liable for harms to minors could unleash a wave of litigation forcing social media companies to redesign their products, the Journal reports.

• On Wednesday. A 20-year-old woman prevailed in a Los Angeles trial against Meta Platforms and Google’s YouTube in which the companies were accused of designing their apps to be addictive and harmful to adolescents.
• Just 24 hours earlier, a New Mexico jury found that Meta exposed minors to harmful content, including online solicitation, sexually explicit content and human trafficking under consumer-protection laws.

The Journal reports that LA verdict carries major implications for Meta and other platforms, which have long sheltered behind Section 230 of the 1996 Communications Decency Act. Plaintiffs' lawyers bypassed that shield by arguing that the way the platforms themselves were designed was harmful and intentionally so.

Trump names Silicon Valley chiefs to tech panel. President Trump installed some of the biggest names in business—including Meta Chief Executive Mark Zuckerberg, Oracle Executive Chairman Larry Ellison and Nvidia CEO Jensen Huang—to a technology council to weigh in on AI policy and other issues, the Journal reports. Several of the council members lead companies that have helped fund the president’s pet projects, the WSJ reports. Meta donated to Trump’s ballroom as did Huang.

Chief People Officers from some of America’s largest companies gathered Wednesday in Menlo Park, Calif. to kick off the WSJ Leadership Institute’s inaugural CPO Council.

CPOs have quickly become the first generation of executives to manage a joint workforce of both humans and AI agents and shepherd their companies through that transition.

Agents are already working side by side with humans at companies like Microsoft, said Chief People Officer Amy Coleman, “But we are in that messy time of how we’re going to figure it all out,” she added.

Coleman said she frequently has conversations with CEO Satya Nadella about navigating this tension between betting on the promise of AI, but also betting on the people building that promise. The reality is, “Shifting our workforce will be something we always do,” she said.