Google CEO Sundar Pichai recently talked about his daily routine in a podcast. Even at the heart of the company, he finds it hard to keep track of everything happening at Google each day on his own. In the past, figuring out what had happened over the past 24 hours often required setting aside a dedicated hour; now he has handed this pipeline over to AI agents—first letting them sort and summarize the information before passing it to him, though of course he also digs deeper offhand, and he does so frequently.
Besides daily briefings, these agents also keep watch on two things for him: on one side, tracking the R&D progress and compute consumption of Google's AI models; on the other, capturing discussions scattered across multiple internal group chats, so he no longer has to scroll through chat records one by one. Pichai mentioned that the team has also deployed a batch of agents specifically monitoring the compute utilization and operational efficiency of each product line, delivering data and analysis on a regular basis. In his view, this lets everyone do their work better.
Humans are still the final decision-makers
However, outsourcing work to machines does not mean handing over decision-making power along with it. Pichai made clear that management, including himself, still personally participates in major decisions such as compute allocation, and has not given the authority to make the final call to AI.
Saving effort does not necessarily mean reducing the burden. A Cisco engineering director who oversees more than a dozen AI agents revealed that supervising this group of "digital subordinates" requires constant task-switching, making the pace even more intense, and the length of the workday has not shrunk.
In Pichai's eyes, what makes agents truly valuable is precisely that they never tire—deployed in the cloud, working nonstop seven days a week, 24 hours a day, while also remembering where you left off, able to seamlessly pick up from where you last stopped. With tasks continuously progressing, people can move on to tackle the next problem, and then the next.