Garry Tan Spends $200K a Year on AI Tokens - What Small Business ROI Actually Looks Like

To decide whether automating a process makes financial sense, compare the current monthly labor cost of the task against the expected token spend plus build cost of an agent that can carry 60-80% of it. Start with the single workflow that burns the most repetitive hours on stable inputs - invoice processing, lead triage, document review - not the one that looks shiny in a demo.

On Walter Isaacson's podcast The Builders, Y Combinator CEO Garry Tan told TechCrunch he spends $100,000 to $200,000 a year on AI agent tokens. His agents read his email, Dropbox and 20 years of Apple Notes, and YC runs an internal agent called Stray Light with context on every founder they have ever backed. Tan called the spend a crazy privilege and said personal agents will not go mainstream until costs drop.

What this means for a small business

Tan's figure is a ceiling for a venture CEO, not a budget template. For a 5-20 person company, a well-scoped agent on one workflow usually sits at $50-500 per month in tokens - the math works when it replaces 20-80 hours of human work per month. The test is boring: measure how many hours per week the task eats, multiply by loaded cost, subtract build and running costs. If the first-year ratio is not at least 3x, pick a different process. Document-heavy work clears this bar quickly - a tender analysis pipeline I built runs on cents per document and saves days of manual review per bid.

Why it matters: Token cost is the easy part of the equation - the discipline is picking one process with measurable hours and stable inputs, instead of trying to put an agent on everything the way Tan can afford to.