Why The OpenClaw Founder Spent $1.3M on AI Tokens

When a buyer asks AI for a solution like yours, does your business come up? Most companies have no idea. And by the time they find out, they've already lost a deal to someone who did. HubSpot AEO helps you show up in those moments with the right answers buyers are looking for.

Before the first click, before the first form fill. That's the moment HubSpot AEO is built for. Check out HubSpot.com, the agentic customer platform for growing businesses. Open Claw's founder, Peter Steinberger, he spent $1.3 million on tokens in the span of 30 days.

So he's like, people are freaking out over my token spends, but here's what nobody sees. So people are like, they're freaking out about it, the $1.3 million. It's like, how do you justify it? So he's actually justifying it here.

And I think it's helpful for people to understand what the perceived ROI might be. So here's what he says. He says, here's what nobody sees. Part of what excites me so much about working at Open Claw is that I'm trying to answer the question.

How would we build software in the future if tokens don't matter? We constantly run 100 codecs in the cloud, reviewing every PR, every issue. If a fix on main lands, Claw Sweeper will eventually find that six-month-old issue and close it with an exact reference. So he's just saying all these things.

Okay, security issues, right? De-duplicating things, complex theft, ephemeral machines, log into Telegram. So basically, he's trying to, I guess he's just saying that he's improving this quickly at scale. And he's saying that all this automation allows us to run this project extremely lean, which is interesting because there's a lot of people that actually use OpenClaw.

But here's the thing. We'll just leave it at that for a moment because I want to say that Hermes is starting to overtake OpenClaw. But Hermes has a much, I would say they have a more significant team, but OpenClaw has been able to get by so far until now where he's been saying that, hey, they're actually building a team around this. So it's interesting to see this because it's like, oh, this is open source.

How is this so good? It's only one person doing it. But the reality now is like what's covering for it is actually that token spend. That token spend is actually what you would have spent on maybe payroll in the past.

Yeah. What did they spend the $1.3 million on? All this stuff over here, all this automation to help with just maintaining. But what was the end result?

What did he get at the end of the 30 days? I think his point is saying that the $1.3 million that I'm spending a month on this keeps OpenClaw stable with a very lean team. Got it. Yeah.

So it's more so maintenance. Yeah. Yes, so maintenance security issues and a lot of tests that they need to run. That's all he's saying over here. Scans for comments for spam and blocking issues, things like that.

So instead of paying 1.3 to people, it's tokens. The tokens may or may not be cheaper than humans. So 12 million in payroll and how many people are using Open Claw right now. I'm actually curious to see, so let me see.

Yeah, I have no idea how many people are using it. Although I do think Open Claw's growth has slowed down because of Claw. Hermes is kicking its butt now.

Yeah, so I know you can say Claude too, and I think a lot of them are building these solutions, but Hermes is the one that's really eating their lunch. It's taken up like a third of the GitHub stars now. Okay, here, I have some data for you. Just in most of the corporate world, though, I don't hear Hermes too much too often.

Yeah, you can say Claude. Yeah, I hear people, everyone using Claude. I actually think Claude Cowork, I'm sorry, at least of right now, as of this recording, Claude Cowork is a piece of crap. And they've had like these managed agents and things like that.

So I don't know, like, and Claude's obviously going to claim that a lot of people are using it, but if you use it once, you're considered a user, right? So like, what do you say about that? So monthly use visitors for OpenClaw is 38 million.

Running instances, it says 500,000 active users, 3.2 million. So take that for what it is. And then 350,000 GitHub stars and 48,000 forks. And then 1,000 active developers for contributors.

Yeah. So, and I'm actually curious. So Hermes number of users and GitHub. Okay, so Hermes is 158,000 GitHub stars, 25,000 forks, hundreds of thousands globally, contributors hundreds, but they do have an active team as well.

I don't know number of users here. But anyway, all that to say is that is how someone justifies $1.3 million in token spend in 30 days. Yeah. It's crazy.

It's a big amount. Yeah. You look like you're highlighted on one. Do you want me to pick the next one? Here, I'll pick the next one. Go for it. Pick the next one.

So, you know, a lot of people are talking about this. They built this AI tool that replaces CMOs, right? And I want to talk about this, right? So someone here, so Guri Singh,

Your competitors are about to hire an AICMO and you don't even know it exists yet. Higgs field just launched supercomputer and most people think it's a video tool. It's not. It's an AICMO you hire once and it runs your marketing while you sleep.

One prompt, it picks the workflow, routes the thinking to GPT 5.5 Pro and you give it your TikTok, your meta ads, your X, it studies your niche, ships new content, reads analytics. Every run sharpens the next one. You stop opening 14 tabs, you stop hiring four freelancers, you stop guessing what hook to try next. Okay.

So here's the thing about these types of posts. These posts really annoy me because the thing is this, who's the one that's going to come up with the strategies? Who's the one that has a taste ultimately on this stuff? And by the way, this supercomputer from Hicksfield, it eats up tokens like crazy and you can only use it like a couple of times.

Like if you're on a free trial, you can probably use it like once and that's it, right? But it's interesting to me, Neil, because you have the Grand Theft Auto CEO. He said something on an interview. He said what he's learned from seeing so many games being created over the years,

You could have thousands and thousands of games created every year. You might have one to five hits, and that's it. So you know what he said, Neil? He said that in terms of AI replacing humans, it cannot replace the taste piece because you can make as many apps as you want right now, as many games as you want.

It can't replace that. Same thing goes for CMO. A really good CMO with taste and that experience, that taste comes from good experience, no CMO that's good is going to be replaced for a long time. No, the thing that a CMO has, it's not just taste, it's experience.

Which comes from taste. If you've been doing something for a long time, that experience is really hard to replace. And the easiest thing to ask yourself is, how often have you used LLM and found it to be off? Every time I ask this question, almost everyone raises their hand.

You asked that last week. Yeah, I asked it last week and almost everyone raises their hand. And the people who don't raise their hand is kind of scary because it's off. There's instances of when if you use it where it is off.

And if you don't realize that it's off, you're doing things a really wrong way, especially if you're using it for work. So you should always double check. It doesn't mean that it's off majority of the time. It's just off a percentage of the time, whatever that percentage may be.

Some of those mistakes could be costly and can cost an organization millions and millions of dollars, sometimes even more than that. This is why you have experienced talent. Quick break for SaaS founders listening. You know how content marketing works.

You know how AI SEO works. And you've been seeing competitors ranking for your keywords all day long. but you don't have the time to write and hiring a content team feels premature or just out of the budget. That's exactly why we built ClickFlow.

It helps create AI SEO content that actually sounds like your brand. We're talking thought leadership, we're talking product content, comparison pages without having to add more headcount. It also helps you do things such as building FAQs, internal linking, and telling you which old posts are hurting your domain. One user reported saving 90 hours per month, and that was a part-time employee.

You can get a 14-day free trial at quickflow.com. Scale your content without scaling your team. Back to the show. So speaking of experience and talent, Neil, do you know what the Airbnb CEO said, Brian Chesky?

No. So he said, I don't think people managers will have any value in the future. So he said the people most screwed by AI will be the people managers. So, you know, my thinking on this, Neil, is and I'm curious to get your thoughts, too.

When we think about hiring at my company now, you can't just be a manager. You have to be an individual contributor. So, you know, you talked about the concept of player coaches. I think everybody has to be a doer in a company now.

If you're a peer people manager and we both have very people intensive businesses, I think you're in trouble. And so, like, what are your thoughts on this right now? And we can kind of go back and forth on it.

Sorry, someone ended up texting me. I thought it was an emergency. Neil's not paying attention. It's probably related to a deal. No, no, it's not even that.

You know how iPhone summarizes your texts? I don't have notifications on. Try texting me right now. No, I don't have notifications on either. Yeah, then how are you seeing the texts?

I have the text app actually open on my phone. Why are you opening it in the middle of a podcast? It's because I have to move it in the house. So it's like a lot of stuff I want.

So someone's texting me and they're just like. You might as well share a story because. Sure. So someone was texting me about something silly. Technically, first they called me and I didn't.

I ignore the call. So my calls don't go on silent. I can see when a call comes up. And then they started texting me and I can see the text. And Apple is summarizing how it's important.

And, you know, I should call back sooner than later to discuss a bathtub. But I'm like, this is not important. And I just texted back. I'm like, the designer is on site.

Just have them look. I'm like, if there's anything else that's wrong, then text me. But other than that, I'm like, I don't know why you're bugging me. Yeah, I see.

I see. I see what this is. This, to me, is AI failure because it's telling me it's important when it really isn't important. Oh, I know what you mean.

Yeah. It'll say like... Sometimes it'll say like, oh, this is a big... This is emergency right now. And check it. And when you look at it, it's like it wasn't an emergency.

Correct. It's just... You know, Apple summarizing. I'm not saying it's bad overall. I do like the summaries, but sometimes it's just off. Another example of how AI can be off.

Dude, speaking of we're just talking about compute capacity. So OpenAI just announced this. They're releasing OpenAI Guaranteed Capacity, a new offering that enables customers to guarantee long term access to OpenAI Compute. Dude, if I can buy at a fat discount right now, I would consider it.

Are they selling it at a fat discount? Yeah. So I don't know about that. But now it's called guaranteed capacity helps customers plan ahead for critical workloads in a compute constrained world.

Look at how they're marketing it. So you have to talk to sales about this. It says plan for capacity, but scale with capacity you can be certain about.

Plan with predictable compute. It doesn't say anything about discounts. I'll tell you that. Yeah. Yeah, no, people will buy in advance if you can get discounts.

But let's come back to the statement here. So Brian Chesky, Airbnb CEO, says, I don't think that people managers will have any value in the future. And I think what he really means by that is that pure people managers are in trouble. But if you're a people manager and you know how to individually contribute, you'll be fine.

Yeah, it goes back to being a player coach. I do believe there's going to be people managers in the future. I've always believed that managers, at least in my company, our managers are all player coaches. I don't know one manager that's not a player coach in which they'll get their hands dirty and help manage.

The reason I don't believe managers go to zero, let's use marketing for example. If you've been in the industry for 20 years, so like we have an ad agency similar to Eric, ours is called NP Digital. And we have managers who have been in the ad agency space literally more than 20, 30 years. And then we have new people as well who have been doing it for two, three years.

What's valuable about offices, and I know I'm switching, but it'll make sense in a bit. What's valuable about offices is when people come in person, the younger people get to learn from the people with more experience and it helps them improve their skills faster, at least from what we've seen. This is also why people managers are important is because it's

If they're a player coach and they're good at what they do, they have a lot of experience. Even if it's virtual, it's not in person. The younger people may not be learning as quickly, but the younger people will gain a lot of knowledge by just being around seasoned managers who are willing to share and help them out. You know, at least with my observation, Neil, with the younger people right now that just pick it up, it's very easy to train kids on AI when they're, and I call them kids because they're in college or some of them coming out of high school, when it's an empty slate and they're opening to just listen to everything.

Now, that being said, you still need to mix it with experience. So I think we're going to see an influx of more hiring from college grads and even high school grads.

And that's going to be a good thing because people are like, oh my God, all these kids are not going to have jobs. I think a lot of them are going to have jobs. Yes. I think a lot of them will.

And we will end the episode on that. Goodbye.