Token maxing — loading 800K-1M tokens into agent requests at full compute — lets founders operate at 2028 capability levels today, and the expensive cost is justified for CEO-level work.
Garry describes using Open Claw and Hermes Agent at full strength with massive context windows, calling it 'living in 2028' — expensive but transformative for founders and CEOs. ✦ AI generated
Garry Tan · a16z Podcast · 2026-08-12 · original ↗
starts at this moment · 22:22
“Talk about how YC is organizing around loops and how you're advising your companies to.”
if you really want to token max, you actually have to use something like Hermes Agent or Open Claw. Um and then you have to like tune it all the way up. Like you're just like, give me like let me load a million tokens or 800,000 tokens in to any given request and like have that be in your like soul.md. Yeah. But when you do that, like I think that you basically get to live in 2028. Like, you know, it costs, I don't know, 50 or 100,000 dollars a year to like use the agents at full strength, like full 150 IQ on every request. But you'll feel it more or less immediately. It just costs like a crazy amount. But for a CEO or for a founder, it actually makes a lot of sense to do that and you have to give yourself permission to token max in that way.
verbatim transcript · starts at 22:22
22:22any given task. >> Yes. >> And so, um, what's interesting is like I think there are a couple things at work right now, like, um, the frontier model labs that are, you you know, letting you use ChatGPT or Claw, I think that they're still pretty constrained on and pretty constrained on, um, amount of compute they want to give you. And so, if you really want to token max, you
22:46actually have to use something like Hermes Agent or Open Claw. >> Mhm. >> Um, and then you have to like tune it all the way up. Like you're just like, give me like let me load a million tokens or 800,000 tokens in >> Yeah. >> to any given request and like have that be in your like soul.md. >> Yeah. But when you do that, like I think
23:04that you basically get to live in 2028. Like, you know, it costs, I don't know, 50 or 100,000 dollars a year to like use the agents at full strength, like full 150 IQ on every request. But you'll feel it more or less immediately. It just costs like a crazy amount. But for a CEO or for a founder, it actually makes a lot of sense to do that and you
23:31have to give yourself permission to token max in that way. Um, but you know, and then what you get is like I get you get to live in 2028 today. >> Amazing. >> Um, but I think like going back to what founders can do, um, if you live that way, not only do you want to do that task, you want to like skillify it. Like you you do some like
23:53feat of strength and then you turn it into a markdown file plus code plus tests that can be reused and like put into a cron job. >> Yes. >> And so, uh, I think we're just seeing like across the board, um, a markdown file is an employee. >> Yeah. >> And it's an employee that, uh, will do the job perfectly every single time and it will do it like as many times as you
- ·Load 800K–1M tokens per agent request at full compute
- ·Requires Hermes Agent or Open Claw tuned to maximum
- ·Costs $50K–$100K per year to operate at full strength
- ·Grants '150 IQ on every request' — immediate capability boost
- ·Expensive but transformative for founders
- ·Effect is felt immediately after enabling token maxing
- ·Garry Tan: 'You have to give yourself permission'
- ·Annual cost justified by capability unlock for decision-makers