Good morning. Networking-equipment maker Cisco rolled out a personal AI agent for all of its 90,000 global employees this week.

The AI agent, called MyAgent, is personalized to each employee, Cisco Executive Vice President of Operations Thimaya Subaiya recently told me in an exclusive interview. 

Cisco’s AI agent program is one of the first examples of a widespread agent rollout inside a large company, and one where the autonomous bots are set to accomplish a variety of tasks on behalf of their humans.

AI agents have become more powerful and capable of performing different tasks, from writing code to answering customer support inquiries and analyzing financial data. Agent adoption also has been growing among businesses—helping drive the usage of tokens and compute that pushed Nvidia to yet another blowout quarter yesterday.

But there are still plenty of agent rollout challenges tech leaders are grappling with, including “AI agent sprawl,” cybersecurity risks and token costs.

At Cisco, the decision to give every employee an AI agent was driven by the need to consolidate the AI tools it used inside the company. The company then expanded its agent rollout to encourage employees to use AI to get everyday tasks done.

“I truly believe that you cannot replace humans entirely with AI,” Subaiya told me. “What you can do is very successfully augment people with AI.”

One way employees have used their AI agents is to manage their inbox: summarizing emails, filtering out spam messages and automatically drafting responses. Other examples include summarizing and analyzing market news and providing predictions on stock-market movement.

Each Cisco agent pulls relevant context about an employee in real time through secure enterprise connectors based on existing user permissions. The agent is accessed through a desktop application or a mobile app, and employees can chat with it through the Cisco-owned Webex videoconferencing software.

Cisco’s agents are designed so they can call upon enterprise systems and over 800 subagents on the back end. Those subagents perform the tasks employees prompt them for, such as tracking variances in a sales forecast and automatically sending an anomaly notification.

The agents are built only to interact with the humans they support. In other words, agents don’t interact with other agents or other employees—they only take requests from their assigned humans, and any external action requires explicit human signoff, Subaiya said.

Controlling the data that agents are able to access is also a key component of their design. The company built a “policy server” that prevents agents from doing things like deleting data sets, and stops Cisco data from being used to train third-party models, Subaiya said.

To manage the higher costs of using always-on AI agents for every employee, Cisco is using tools including a token-monitoring feature from Splunk, the cybersecurity and analytics company it bought in 2024.

The company also has built an “intelligent router”—which functions like a model harness—to automatically direct tasks to the most capable and cost-efficient AI model. On top of that, Cisco operates its own data center, where it has graphics processing units, or GPUs, that run open-weight models.

Roughly 50% to 60% of Cisco’s AI requests are routed through open-weight models, and 20% to 30% rely on software automation. The remainder is “a very small percentage” that actually goes out to a foundation model, according to Subaiya.

How is your company using AI agents? Has your agent adoption meaningfully increased over time? Send your feedback to me at belle.lin@wsj.com (if you’re reading this in your inbox, you can just hit reply).

From Steven Rosenbush, chief of the enterprise technology bureau at the WSJ Leadership Institute:

I reported yesterday that AI research lab Deep Cogito had raised $43 million to fund the development of specialized AI models that companies can own, helping them control proprietary data and optimize performance while keeping costs in check.

The Series A round was led by TQ Ventures, with participation from other investment firms Benchmark, Nexus Venture Partners, Atreides Management and South Park Commons, plus cybersecurity company Zscaler. This brings Deep Cogito’s total funding to more than $56 million.

Deep Cogito can play an important role in building an open-weight ecosystem in the U.S., according to Schuster Tanger, co-founding partner at TQ Ventures.

“We think of them becoming one of the defining companies building open-weight models and specialized enterprise intelligence. We think there’s an aperture for a highly proficient American solution,” Tanger told me.

DIY AI. Deep Cogito was founded in 2024 by two former Google employees who worked on its AI search products, Drishan Arora, who is chief executive, and Dhruv Malrana, chief product officer. Their research is focused on the post-training phase, after a model has been broadly trained on enormous amounts of more general information.

Arora believes that companies will become increasingly open to the idea of building their own models, as long as they have external help. Currently, only a limited number of companies have the necessary in-house skills and resources.

“The idea that you are going to use open models…for your own proprietary intelligence, and optimize it for your particular content has resonated with a lot of folks. I believe there’s an inflection point here. This is going to happen more and more,” Arora told me. 

You can read my full column here. I would love to know if your company is open to the idea of building specialized open-weight models that it owns and controls.

Nvidia reported record sales of $96.2 billion for the quarter ended in July, beating analyst expectations of $92.3 billion. Chief Financial Officer Colette Kress said she expected 70% revenue growth in the company’s 2028 fiscal year, which, as The Wall Street Journal notes, beat the 45% growth that analysts polled by FactSet had predicted. As the Journal writes:

“Paired with the chip giant’s quarter of blowout earnings that also came in well ahead of estimates, the forecast helped calm rising worries throughout the market over dizzying spending on AI infrastructure and the degree to which Nvidia has extended itself to help customers keep building data centers.”

Notably, the company said that its growth could be even higher. Kress “said customer demand was set to double but revenue would be limited by supply constraints,” the FT noted in its coverage.

Demand for AI is exploding, as our report on Cisco in today’s newsletter shows. Nvidia and other companies simply can’t deliver everything that customers want because they are limited by capacity. It’s no wonder that so much capital keeps pouring into the development of AI at every level, from infrastructure companies to model developers.

• Google is moving a team focused on the risks and societal impact of AI out of Google DeepMind, the research lab developing its frontier AI models, The Wall Street Journal reports. The change in reporting structure—which follows a reorganization that saw DeepMind co-founder and Chief Executive Demis Hassabis step aside into a new chairman role—has prompted some members of the team to raise concerns that its work will be negatively affected.

• Salesforce posted higher second-quarter profit and revenue on rising demand for its AI and data offerings, and expanded its partnership with Anthropic. The partnership, dubbed Claudeforce, combines its broad platform of business tools with the AI company’s Claude system, WSJ reports.