Garry Tan: Investing Heavily in AI to Get Ahead

Garry Tan Encourages Massive Investment in AI Tokens

Garry Tan, an influential figure at Y Combinator, has recently advocated for massive investment in AI tokens. According to him, founders who adopt this approach can access AI capabilities now that will only be available to the majority in the future. Tan referred to this strategy as tokenmaxxing, despite criticism from some segments of Silicon Valley who deem it costly and unnecessary.

In a recent episode of the A16z podcast, Garry Tan delivered a bold message to founders concerned about the high costs associated with intensive use of AI tokens. He encouraged them not to hesitate to spend massively, even if it involves significant expenses. "Burn, baby, burn," he stated, emphasizing the importance of maximizing the use of AI agents.

A Futuristic Vision of AI

Tan explained that intensive use of tokens allows founders to live as if they were already in 2028. He illustrated his point by suggesting loading a million tokens or more, which he believes grants access to a level of AI capability that most will not reach for several years. However, he acknowledged that this advantage comes at a considerable cost, estimated between $50,000 and $100,000 per year to fully utilize these agents.

Despite the high cost, Tan believes this approach is wise for CEOs and founders, as it allows them to leverage AI in an innovative and forward-thinking manner. He emphasized that this costly experimentation can be profitable in the long run.

Monetizing AI Investment

Tan proposed a method to monetize these AI expenses. Once an AI agent successfully accomplishes a task, he recommends making that process reusable. This can be done by creating a markdown file containing instructions that the AI can automatically follow to reproduce the task indefinitely. "A markdown file is an employee," he asserted, highlighting that these instructions enable the AI to perform perfect work every time.

The Debate Around Tokenmaxxing

Garry Tan stands out as one of the few tech leaders advocating for tokenmaxxing. This practice, which involves allowing AI agents to consume a large number of tokens, is criticized by other experts in Silicon Valley who consider it unnecessary.

Recently, Praveen Neppalli Naga, a technical lead at Uber, expressed reservations about this approach. He noted that the rideshare industry is beginning to see signs indicating the end of the tokenmaxxing era. According to him, the next phase will not be defined by the amount of tokens spent, but by the efficiency of their use.

In a June podcast, Scott Wu, CEO of Cognition, also criticized tokenmaxxing. He pointed out that companies have sometimes lost sight of the goal by focusing on the quantity of tokens used rather than the results achieved. "Let's try to rank people by the amount of results they actually produce," he suggested, questioning the current approach of the industry.