YC chief Garry Tan's $200,000 agent bill shows the real cost of AI
Y Combinator's Garry Tan says his AI agents cost up to $200,000 a year in tokens. Here is what that price tag signals for Indian founders paying in dollars.
Key points
- Y Combinator chief executive Garry Tan said on Walter Isaacson's podcast The Builders that he runs AI agents for work and personal life.
- Tan put his token spending at about $100,000 to $200,000 a year, and said agents are too expensive right now to go mainstream.
- For Indian startups billed in dollars, the cost gap strengthens the case for subsidised compute and cheaper local models.
Photo: Dariusz Sankowski / Wikimedia Commons, CC0 1.0 (cropped). Representational image.
The News
Garry Tan, chief executive of the startup accelerator Y Combinator, says he spends "about maybe $100[,000] to $200,000 a year on tokens" running his AI agents. He made the admission on an episode of Walter Isaacson's podcast The Builders, as reported by TechCrunch on 2 November 2026.
Tan called the bill "a crazy privilege that not everyone can do." He uses agents for podcast research, interview preparation, coding and a long list of work projects. He also said his agent has read all of his email, his Dropbox, his Apple Notes and 20 years of his therapy notes.
Y Combinator itself now runs an internal agent called Stray Light, which Tan said is rebuilding how the accelerator operates. It holds a "company brain" with context on every founder YC has backed, everyone who has emailed YC and every application. Tan has also published gstack, a set of 23 Claude Code skills, on GitHub.
Why It Matters
The figure is a rare public price tag on what heavy agent use costs today. Tan himself concedes the point: personal agents will only go mainstream if prices fall, because "It's too expensive right now." That puts cost, not capability, at the centre of the agent debate.
History suggests the price will move. When GPT-4 launched in March 2023, OpenAI charged $30 per million input tokens; GPT-4o arrived in May 2024 at $5, a drop of more than 80 per cent in 14 months. If that curve holds, a six-figure habit for one power user could become a monthly subscription line for a small team within a few years.
Tan's other argument is about control. He criticised free agents run by Big Tech, naming Meta, which he said "controls the algorithm", and argued for personal agents the user programs and owns. For founders, that frames a clear product gap between consumer assistants and user-controlled tools.
Indian Angle
For Indian founders and developers, token spend is billed in dollars, so every rupee move against the dollar changes the bill. A budget that a Silicon Valley executive calls a privilege would swallow the entire annual tech spend of many seed-stage Indian startups. That cost gap is one reason the government's IndiaAI Mission, approved in March 2024 with an outlay of Rs 10,372 crore, offers subsidised GPU compute to local builders, and why home-grown model makers such as Sarvam AI are pitching cheaper, India-hosted alternatives.
The privacy dimension is just as relevant here. Tan's agent reads email, files and therapy notes. Indian companies building similar agents, or deploying them for staff, must handle that kind of personal data under the Digital Personal Data Protection Act, 2023, which requires clear consent and purpose limits. An internal "company brain" holding data on every applicant, as YC describes, would need careful design to fit that framework.
There is also a direct link to the YC pipeline. Indian founders are a regular presence in YC batches, and Stray Light means their applications and emails now feed an AI system that shapes how the accelerator works.
FAQ
How much does Garry Tan spend on AI tokens?
Tan said he spends about $100,000 to $200,000 a year on tokens for his AI agents. He described this as a privilege few people can afford, and said costs need to fall before personal agents become mainstream.
What is Stray Light?
Stray Light is Y Combinator's internal AI agent. According to Tan, it has a personality and a "company brain" with context on every founder YC has funded, everyone who has emailed YC and all applications, and it is changing how YC operates.
What does this mean for Indian startups?
Agent costs are priced in dollars, so heavy use is far costlier relative to Indian startup budgets. Founders may lean on subsidised compute under the IndiaAI Mission or cheaper local models, and must design agents that handle personal data in line with the DPDP Act, 2023.
Where can I hear the full interview?
Tan's comments come from an episode of Walter Isaacson's podcast The Builders. TechCrunch's report, linked below, summarises the key remarks, including his views on Big Tech control of agents.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.