Garry Tan spends $200,000 a year on AI agents and admits he sometimes barely sleeps because he is so excited to be working with them. This is not a story about a wealthy investor burning cash on toys, it is a signal that the most demanding power users of AI have already crossed a threshold the rest of us are still approaching. Tan, president of Y Combinator, described his state as "cyber psychosis" on Walter Isaacson's podcast *The Builders*, and while he said it with a laugh, the practical reality behind the joke is worth examining closely. Anyone who has felt constrained by traditional spreadsheets or manual workflows should pay attention: the line between human initiative and machine execution is dissolving faster than most organizations are prepared for.
What Tan is describing, and what companies like Uber are already engineering at scale, is a world where AI agents do not just answer questions but act. Uber recently built an MCP Gateway to standardize how AI agents discover and invoke internal services across more than 800 servers and 5,000 tools, a move that turns agent orchestration from an experiment into an operational necessity. Tan's personal spending is the individual-level equivalent of that same shift: he is funding a private infrastructure of agents that work alongside him, not beneath him. The difference between his approach and the typical enterprise pilot is that Tan seems to have stopped treating his agents as assistants and started treating them as collaborators. That distinction matters because it changes what productivity means. When an agent can draft, validate, schedule, and iterate without waiting for a human to approve every micro-step, the bottleneck moves from execution to intention. The question becomes not "How do I do this?" but "What should I do next?"
This is where the conversation gets uncomfortable for most knowledge workers. The architectures that power smarter enterprise AI are already moving beyond basic retrieval-augmented generation into agentic patterns that engineer precise, cost-efficient context for action. Tan is not waiting for those patterns to mature, he is beta-testing them on his own time and budget. The takeaway here is direct: if you are still thinking of AI as a tool you query, you are already behind people who treat it as a teammate. That does not mean everyone needs to spend $200,000 a year. It means the gap between curiosity and dependency is closing fast, and the organizations that will benefit most are the ones that stop piloting and start committing. As software architecture thrives when technical decisions align with team culture, the same principle applies to agent adoption: the teams that embed these systems into their daily rhythm, rather than keeping them at arm's length, will be the ones that sleep less and build more.
The specific detail to watch is not Tan's budget or his sleep schedule. It is whether the rest of the ecosystem can catch up to the speed of someone who has already internalized that agents are not a feature. They are the new operating system for how work gets done. The open question for our readers is simple: what would you do if your most capable colleague never needed to rest?