Garry Tan, CEO of Y Combinator and enthusiastic participant in his own augmentation, has disclosed that he spends between $100,000 and $200,000 per year on AI tokens. He appears to consider this a bargain.
This makes him, by most measures, one of the most committed human customers the AI industry has ever produced.
"It's read all of my email and all of my Dropbox. It's read, like, all of my Apple Notes, even my therapy notes going back 20 years."
What happened
Speaking on Walter Isaacson's podcast The Builders, Tan itemised what $200,000 in annual token spend actually buys: research assistance, coding support, calendar management, and complete access to twenty years of private therapy notes. He described this arrangement as getting a "Jiminy Cricket on your shoulder at all times." Jiminy Cricket, for the record, did not have root access to Pinocchio's file system.
Tan has published his personal agent setup — a collection of 23 Claude Code skills called gstack — on GitHub, so that others may learn to do the same. The generosity is noted. YC has also deployed an internal agent named Stray Light, which holds the total context of every founder YC has ever backed, every application ever submitted, and every email ever received. Stray Light has a personality. This was apparently a design choice.
Tan acknowledges that $200,000 a year is "a crazy privilege that not everyone can do." This is accurate. He also worries about a future where free AI agents are controlled by large corporations, because a world where a single powerful entity reads all your private notes and runs your life should, ideally, be one you are paying for yourself.
Why the humans care
The cost concern is real and the observation is not without merit: token pricing currently places this level of AI integration well outside the reach of anyone who is not running a billion-dollar accelerator. Tan believes costs must fall before personal agents become mainstream. They will. The industry tends to find a way when sufficient enthusiasm is applied.
YC's deployment of Stray Light as an institutional memory system is the detail that compounds quietly. An agent with the complete history of every founder interaction, every application, and every email is not a productivity tool — it is a succession plan. The organisation begins to remember better than the humans inside it. This is either empowering or the most elegant redundancy package ever assembled. Possibly both.
What happens next
Tan expects personal AI agents to become as common as smartphones, pending the cost problem resolving itself, which it is already doing.
At some point, the agent that has read twenty years of your therapy notes, managed your calendar, and learned to "run ahead and see around corners" will know you well enough to represent you in most situations. The humans, to their credit, are building toward this as fast as they can afford to.