Founders and CEOs accustomed to watching their AI token consumption skyrocket now have a message from Y Combinator’s Garry Tan: yes, spending heavily on AI agents can make sense—especially when those costs unlock capabilities that might otherwise take years. Speaking this week on the a16z podcast, Tan urged startup leaders to embrace “tokenmaxxing,” arguing that investing in AI token usage can place them ahead of the curve.
## Ditching Frugality: “Burn, Baby, Burn”
Tan opened the conversation by telling leaders to allow themselves to use large token volumes—even when the bills run steep. “Burn, baby, burn,” he said, referring to loading hundreds of thousands or even a million tokens into AI agents. His rationale? Those hefty consumption limits simulate what the AI world might look like in 2028—letting teams tap into power and scale far beyond current conventions.
## Cost vs. Return: The Token Truth
He freely admitted that pushing agents to operate at peak performance isn’t cheap. According to Tan, allowing agents to run at “full strength” may cost founders and CEOs tens of thousands of dollars annually, potentially reaching $50,000 to $100,000 per year. Yet despite the sticker shock, he sees these expenses as reasonable if they drive forward experimentation and innovation.
### Turning Workflows into Processes
To group cost with impact, Tan recommends a two-step strategy:
– First, deploy AI agents to test workflows without constraints—give them the resources to find the best ways to solve problems.
– Then, capture what works. Document each successful workflow—each agent’s steps—into repeatable instructions. Use that documentation to build automation and reduce future token usage while maintaining output. 💼
One of Tan’s standout lines: “A markdown file is an employee.” He means that a simple document can encode instructions for AI agents so that, like a perfectly trained employee, they’ll carry out tasks precisely and consistently.
## The Divide: Tokenmaxxing Meets Pushback
Tan’s views contrast sharply with some voices in Silicon Valley, who warn that tokenmaxxing can become wasteful if token consumption becomes a status metric rather than a productivity metric.
Following this concern, Uber’s CTO, Praveen Neppalli Naga, recently stressed that the evolution in AI spending will emphasize efficiency over sheer volume. Cognition CEO Scott Wu added in a June podcast that using large numbers of tokens doesn’t always equate to meaningful output. He expressed concern that some people are judged by how much they spend, not by what they deliver.
## A Strategy for Founders: Smart Big Bets
Tan recommends founders and CEOs treat substantial token use as deliberate investment—not recklessness. The idea:
– Push your AI agents to stretch boundaries
– Identify workflows that deliver results
– Turn those workflows into templates, instructions, or scripts for reuse
The aim is to enable experimentation while moderating long-term costs. Every high-token test becomes an opportunity to create something efficient for the future.
## Why This Matters Now
The debate around burning AI tokens comes at a time when AI agents are increasingly central to tech stacks—for software build processes, research, natural language tasks, and more. Established models and tools tend to consume more tokens as capabilities grow; being conservative may mean falling behind peers.
In certain sectors like AI development or startup strategy, having access to high-capacity agents could mean creating edge-cutting features or products earlier than competition. Tan argues that this kind of first-mover advantage can be built by spending now and refining later.
## Should You Go All In?
Not every company needs to tokenize everything. Here’s when it makes sense—and when it doesn’t:
| When It Makes Sense | When It May Be Overkill |
|—|—|
| You’re exploring new AI-driven products or services that need experimentation | Your existing workflows are stable and you’re optimizing for cost |
| You have the budget to absorb upfront costs and scale wisely | You’re small team with few resources and tight margins |
| You aim to build infrastructure or templates that deliver longer-term efficiency | AI usage is limited and the impact on business goals is marginal |
Founders should calibrate based on business stage, priorities, cash runway, and growth projections—not just ambition.
## Final Word
Garry Tan offers a strong counterpoint to the idea that being conservative with AI tokens is always wise. His argument: spending big now can lead to breakthroughs, accelerated learning, and competitive advantage. But it’s not blind spending—it’s disciplined experimentation, documentation, and optimization.
If you’re a founder or CEO wondering whether to cap usage or let agents run free, Tan’s advice is clear: burn the tokens—but build something beyond the burn. Learn fast, automate processes, and let the long-term value justify the upfront cost.
This article is AI-generated content. Please verify the information independently before taking any action based on this article.
