The CEO of the company building the agents is using agents to run the company — and still spending hours checking their work.
The Summary
- Sundar Pichai uses AI agents to track Google's model development, compute allocation, and internal discussions — tasks that previously took him an hour daily just to stay current
- The agents summarize updates and monitor resource usage, but humans still make the final calls on consequential decisions
- Managing these agents is becoming its own skillset, creating new overhead rather than pure efficiency gains
The Signal
Google's CEO is running the same experiment as everyone else: can AI agents actually make you more effective, or do they just shift where you spend your time. Pichai told Sources podcast host Alex Heath that his agents handle three specific tasks: synthesizing model development updates, tracking compute resource allocation across product areas, and following internal chat threads. The work still happens. Someone just has to teach the bots what matters.
The interesting part is what Pichai admits he still does himself. He goes deeper "often." He personally makes decisions about compute allocation. The agents compress information, but they don't replace judgment. This tracks with what we're seeing across companies deploying agents: the technology handles volume, humans handle nuance.
"Managing agents is becoming a skill in its own right, giving some workers a new set of headaches."
The tell is in what Business Insider reported separately: managing agents is making everyone a boss without the pay bump. You're not doing less work. You're doing different work. Instead of reading 50 Slack threads, you're:
- Defining what the agent should track
- Checking whether it's tracking the right things
- Verifying outputs before making decisions
- Preventing duplicate work when multiple agents overlap
- Keeping agents from wandering outside their scope
That's management. Pichai has agents monitoring Google's compute utilization and efficiency by product area. Those agents need telemetry access, boundary definitions, and someone who understands what "normal" looks like so they can flag what's abnormal. The CEO might spend less time reading. He's spending more time being a product manager for his personal agent stack.
This is the pattern emerging across Web4 infrastructure. Agents don't eliminate work. They change its shape. The companies winning this transition are the ones building tools that make agent management less like herding cats. The workers adapting fastest are treating agents like junior employees: clear instructions, regular check-ins, trust but verify.
The Implication
If the CEO of Google is still personally reviewing agent outputs and making the final calls, your middle manager definitely is. The pitch that agents will "free you up for strategic work" is half true. They will free you from some tasks. Then you'll spend that time managing the agents. The skill to develop now is knowing what to delegate and what to own. Pichai delegates information synthesis. He owns resource allocation decisions. Draw that line clearly in your own work, or you'll end up managing bots full-time while the actual decisions pile up.