YC Leader Supports High Investment Strategy in AI

CoinWorld reports:

Garry Tan, CEO of Y Combinator, believes that startups should not focus too early on reducing token costs when using AI agents. Instead, he advocates for founders to invest aggressively in computing power and tokens in the early stages, trading higher costs for stronger model capabilities and faster trial-and-error speeds.

Advocating for Maximum Capability

In the latest episode of the A16z podcast, Tan stated that founders need to allow themselves to consume a large amount of tokens. In his view, increasing the intensity of AI agent usage is akin to utilizing capabilities that will only become more widespread in the coming years.

His assessment is that while this approach may be expensive, it could be cost-effective for CEOs or founders. According to him, if the agents are "fully utilized," the annual cost could reach between $50,000 and $100,000.

The Focus is Not Just Spending Money

Tan is not merely encouraging higher spending. He emphasizes that the key is to solidify experimental results into repeatable processes. Once an agent successfully completes a task, the team should organize the process into standardized instructions for the system to execute automatically in the future.

He views these documentation files as a form of reusable digital labor. In other words, what startup teams gain from higher token costs in the early stages is not just a one-time output, but a workflow that can be continuously invoked later.

• The high-intensity usage cost mentioned by Tan: approximately $50,000 to $100,000 per year
• Target audience: CEOs, founders, and high-frequency trial-and-error teams
• Core method: high initial investment in experiments, then writing processes into reusable instructions

Diverging Views on Tokenmaxxing in Silicon Valley

Tan is one of the few tech industry figures publicly supporting "tokenmaxxing." Tokenmaxxing refers to the practice of actively allowing systems to consume more tokens when using AI agents, rather than prioritizing the reduction of usage and costs.

However, this idea is facing increasing scrutiny. Last week, Uber's technology head, Praveen Neppalli Naga, stated that the AI cost data observed by the company suggests that the so-called tokenmaxxing era may be coming to an end. The next phase of competition will not be about who spends more tokens, but who can use these resources more efficiently.

Scott Wu, CEO of AI programming company Cognition, also expressed similar views in a podcast this June. He believes that some companies have strayed from the focus when encouraging tokenmaxxing; measuring team performance should not be based on token consumption but rather on the final output.

Overall, this divergence reflects the industry's search for a balance between capability-first and efficiency-first approaches as AI applications enter a more refined stage. For startups, whether they are willing to bear higher costs upfront in exchange for faster product iterations and process solidification is becoming a new business choice.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.