How I Cut My $7,500 Claude Cost to Almost $0

Here's how I decreased basically my $7,500 a month AI token spend to zero, or it's trending towards zero. So you can see this over here. So Neil, look at this. So you can see this month to date so far, $6,700, right? You can see the graphs over here. It's like going up to like $400 a day or whatever. And I'm not saying I'm spending the most because some people hit me up on Instagram. I'm spending $30 grand a month. Congratulations. So anyway, my point is you can see the graph starting to go down.

Okay. So what's the red versus the white?

That's just the different models. So red would be Sonnet, and then you have Opus over here as well, and then you have Opus like 4.7 and all that. So my point is, Neil, I got it down somehow, something broke over here, like yesterday it went up. So I'm gonna have to look into that. But all I did, Neil, was I changed my model hierarchy. And so what I mean by that is, I have— I paid $200 a month for Claude Max, okay? And then I said, what are you using first versus second versus third? And then OpenAI OAuth is number 2. ChatGPT 5.5 is amazing. And then the last one is a fallback to the API. And I realized what happened yesterday. What happened with my OpenCLaw is it kept screwing up. It kept calling the old API because it didn't realize that it had the CLI. So something's broken there. And so that was fixed as of this morning, actually. So that was fixed. So you do need to continue to monitor your costs, but I got this as low as I got this to $20. So it was trending towards zero and I'm going to remove all the cron jobs. And so now it's a good deal.

That's a great deal. Here's something that I saw interesting the other day, because I was talking to Heath and we were talking about Mac Studios because he has quite a few of them.

Yeah. He has like 15, right?

I don't know.

He has a lot. So you know what happened? Remember Thomas from SaneBox? So they interviewed him for his podcast and he's like, yeah, Heath has 15 Mac Minis. How many do you have? I was like, I have one.

So He's been buying like Mac Studios as well, not just Minis. And I didn't know this, he was breaking it down to me. He's just like, dude, he's like, yeah, it costs a lot for memory. Like you can't even get any of these computers anytime soon. I'm like, what are you talking about? So I go to the Apple website.

Oh, nothing's available.

Nothing's available. And the memory storage is way lower than what you actually need for the model.

I was trying to buy the 512 and then 2 weeks later it was only down to 256. Now I can only get the 96. Is the 96 even available? For the memory? Uh, 96, right?

Uh, you can get 96 or 256, but your delivery time is nowhere near soon and they don't offer, uh, 512 anymore.

Yeah. So I use, um, so my plan is to wait for the new ones. They're going to sell out quickly, but I want to buy like 2 of the new ones, but I bought 2 of the DGX Sparks. That's why, because the Mac Minis are out. And so Neil, to your point, the, the Quinn model, I think Alibaba, their open source model, almost equivalent to Opus 4.6. Opus 4.7 is out right now, but it's really good. It's really solid. And so what we talked about earlier in this podcast, my Camry fleet has to run on these DGX Sparks, and then we're also going to have open source as well. So I might just eliminate the API from, at least from my personal, but I think for my company, you're still going to have to keep paying for API because you want them to have the best models.

Yep.

Yeah. Just like you want your kids to have the best food, the best schooling, and all these things.

Yes, that is true. It all ends up adding up to being a lot of money. But yeah, dude, these computers, like I was going on the checkout, Like these things are expensive. I didn't realize how expensive these can be.

It's like 16 grand for a fully loaded one.

Uh-huh. I was like, this is like a car almost.

Yeah, it's, but it's worth it. It's just a productive car, way more productive than a car.

Yes.

Yeah. Okay. So while, while Neil's looking for something right now, I'm going to go for the next one.

Okay.

So Salesforce just admitted it, switching from per seat to usage pricing made customer spend go up 83%, Neil, 83%. So earlier we just talked about how Salesforce is going all in on APIs, CLIs, all these things, right? So Jason Lemkin, the SaaStr or the EchoSign guy, okay, he said they reduced their seats from 10+ to 2 human seats and 1 API seat. And yet we now pay $22,000 a year, 83% up from $12,000. Why? Our 20+ AI agents use Salesforce 100 times more than than we did a year ago. So they now have an AI VP of marketing, AI VP customer success, and run 4 third-party AI GTM agents, including Agentforce.

Is their pricing model still by seat, but it's for AI? Is that what they switched it to?

API, the API cost.

Oh, the API usage is going up instead of the seats. So they're spending less on the seats, but they're spending way more on the API.

Yeah, which you can try to optimize it as much as you can. It's still going to keep going up. It's just about dampening the cost.

So he's spending more money in total.

Yeah, you're spending more money. So 83% up, like 12 grand to 22 grand, right? And so, I think we're going to continue to see this. Now, here's the flip side of that, Neil. Notion. Do you use Notion? Do you pay for Notion?

We probably pay in our company, but I don't use it.

I don't use it either, right? So here's the thing. Notion, they're all over AI, right? They're like, okay, you know, it's— they've been accelerating their AI adoption, all this stuff. They just realized they hadn't used Notion in months, even with the strong AI product enhancements, because their agents don't need Notion, right? They do need Salesforce. What do the agents need? They need .md files, that's what they need. You're okay just using Obsidian for that.

Wait, is this Notion realizing or is this Jason Lemkin realizing?

Jason Lemkin realizing he doesn't need Notion.

I thought you were like, Notion realized they don't need Notion.

Yeah, they're going to shut their company down.

I'm like, wow, I don't know who runs marketing there, but I would fire that person.

Yeah, so I think that's really interesting. Per seat versus usage-based, I think we're going to continue to see this change because even the API for SEO stuff, the way we're planning to do it now is, okay, freemium, you get some access and then you're going to pay on usage. I'm sure you're going to— you're thinking about that for your SaaS products too. Yes.

And then a lot of companies, if you look at FINN, it is for ticket resolved, which, you know, people just want— I can go both ways. I know a lot of people want to pay for outcomes.

It's hard to measure too.

It's hard to measure. I personally, as a business owner, prefer monthly costs that are fixed.

Yeah, because it's more predictable.

Very much more predictable.

Yep.

Now if we use it a lot more, eh, all right, I get you, you may have charging, just like with emails. Like you know how your emails is fixed, but if they, if you send out X amount more emails, you have X more contacts, you go into a higher band.

Yep.

The reason I personally like that, and you nailed it, is I like predictability in our cash flow, in our accounting, our business, what we're going to generate each quarter. Like I don't like this ambiguity of, oh, you just pay for whatever you use. Well, if you just get spammed, And a lot of, you know, Fin fixes a lot of support tickets that were just spam or bullcrap. Like what, you just ended up spending thousands of extra dollars for something that wasn't needed?

It's tough. I think it's a tough job for CFOs in the next few years because they're going to, because your enterprise company, before you're trying to protect as much, but now it's like you have to pay on usage for the best Claude tokens, the best OpenAI tokens. I do think, You know, the agents that we're selling right now for the revenue agent side, we're basically going to charge like a base fee. There's a management fee, but also there's token overages too. And we're probably just going to mark up whatever it is that they're spending because we don't want to hold them back. If you're getting this intelligence that could do more for you, you probably want to pay more for it. And so unless you don't want to grow as fast, so like what are you supposed to do? Yeah. Yeah. But speaking of Fin, by the way, they launched a sales agent. Did you see that?

I did not see that.

They launched a sales agent. They're betaing it right now, but I think they're trying to charge people based on like a per lead basis. So we'll see how that goes.

And it just goes out there and tries to get you leads?

I think it's more of an inbound. So it comes to your website and then they take on those leads.

So they take on leads that I've already taken on myself and then they'll do the sales for you?

Yeah.

And then they charge you a percentage?

I think they'll charge you just more than $1. I think they'll make, maybe they'll say for like, I don't know, $500, $100, $50 or something like that.

So, but they'll help you close the deal?

Yeah. Yeah, and maybe this thing will just help you with self-serve. So I don't know all the details.

Yeah, I don't see it working for enterprise, but for self-serve, it could potentially help.

Quick break. If you're doing AI SEO or content marketing in general, you know what the problem is. The problem is you're spending hours writing yourself, or you're paying agencies lots and lots of money, and you're getting AI slop that sounds just like everyone else. That's why we built ClickFlow. ClickFlow creates production-grade content that actually sounds Like you. Plus, it helps you with other areas such as building internal links, creating FAQs, and more. One customer even reported saving 90+ hours a month just by using ClickFlow. You can get a 14-day free trial just by going to clickflow.com. No more AI slop, just production-grade content that drives traffic and conversions. Back to the show. Yep. All right, so let's talk about this one over here. So Neil's tired already.

So wait, before we get into the next one, you want to know why we're tired? Why? Dude, my son comes, he says 3 o'clock in the morning. I usually wake up like at 4 something, 5 at the latest, but he's like, check in on me. So my kids have this thing where like they want you to check in on them. So I'll put them to bed. And I usually do the nighttime routine. Then I surprise check in on them within like 5 minutes of putting them to bed so they know that I'm watching. We have cameras in their room, not in a creepy way, but like the nanny cams to make sure they're okay.

Yeah.

You know, they don't just try to roam out the front door. It notifies us because you don't want your kids just trying to like leave the house or do something that they're not supposed to. Or if they cry, it notifies us, uh, like something could be wrong. And my son, like at 3 AM, he's like, check up on me. So I go to his room like, hey, everything okay? And he's just like, I had a nightmare. I was so scared. And he's like shaking. So then he's like, will you lay with me, Daddy? Please. I'm so scared. So I just lay in his bed.

Is he in his own room?

He's in his own room, yeah. So then I'm laying with him and I'm just like trying to fall asleep and he's like grabbing me tight and shaking.

Oh, so you don't have sleep.

So I didn't have much sleep. And then he's like, are you gonna leave in the morning to go to the gym? And I'm like, yeah, but I'm like, you'll be okay. You'll be sleeping by then. He's like, but then I won't know if you're leaving and then the monster can come in. Like, there's no monster. He's like, can we go lay in, in, uh, In your room instead. So I'm like, all right. So we go lay in my room because he's like, at least mommy will be there in the morning while you go to the gym. And then he's just excited to sleep in our bed because he doesn't do it. So he can't fall asleep. And he's like trying to— he's waking up my wife, then she's not going to get sleep. I'm like, let's go back into your room. So then I didn't get any sleep after pretty much 3:00 AM.

Oh, you should have done this. Okay. Here's how you solve it. Creatine, 15 to 20 grams of creatine, you're locked back in.

Me?

Yeah. When you, when you don't get good sleep, you take 15, 20 grams, it locks you back in. It, it's, it's your, it's like as if you had good sleep.

But does it help you get good sleep when someone's pulling on your arm and not sleeping themselves?

You feel like poo poo right now, right? If you have 15, 20 grams of creatine, you'll feel good.

Oh, you're saying right now?

Yeah, right now.

Not when he's waking me up like at 3 AM.

No, no, you just deal with that.

Yeah.

Yeah. Yeah.

I'm like, I don't know if that'll solve. No. But yeah. So if I put 15 to 20 grams, you're saying I can take a quick nap?

You don't even need a nap. You'll bring you, like, I've had nights where I had bad sleep. There's nights where I didn't sleep at all and like, I, I, Zero sleep. And I took the 15, 20 grams. I was good. I made it through the day.

Oh, cuz it's like caffeine, right?

It's not caffeine. No, it's just, it, it, I forgot what it is exactly, but what I've, so what people have been talking about recently with creatine is not just good for building your muscles, it's good for cognitive benefits. But because your body eats the 5 grams so quickly, um, in other areas, if you want the brain benefits, 15 to 20 grams, that's what you do. And it'll, it'll cancel out the bad sleep.

So I've been using it for brain benefits for a long time. I take it daily. I take 10 grams. I don't do 15 to 20, but I always make sure I do 10. Excuse me, I do 15, 20.

Yeah.

I didn't know it was 15 to 20.

Look at, look at, uh, Dr. Rhonda Patrick. She talks about it all the time. There's, there's enough studies on this already.

Yeah. And is it 15 all at once or you space it out?

I just do it all at once. That's why I, that's why I have to pee so much when I come here.

Oh, 'cause, oh, 'cause you drink a lot of water 'cause doing—

Well, no, no, I take a lot of supplements in the morning and that 15 grams, 15, 20 grams plus maybe I, I take all the other supplements. Maybe I'm taking like 50 to 60 grams of stuff. So it's a lot.

On a side note, Eric and I are not giving you creative or medical or health advice.

This is, we're giving, we're just talking to each other as friends.

Okay.

That's all. None of the disclaimers Next one. All right. So 70% of brands don't show up when an AI agent goes shopping. Yours probably doesn't either. Okay, so if we look at this over here, so I think this is a game changer, Neil. Google's design.md file is the 50x brand scaling system. So not enough people are talking about this right now. So Google just came out with— they released this. They're trying to have a new standard, like just as there's an agents.md file that agents will use, there's like these standard.md files.

Just like they used— for everyone listening, everyone remember their old AMP framework?

Yeah. This is a new framework that they're trying to standardize for everyone to use. That's why they open sourced this format. But the example that I just gave a little earlier where I showed you my landing page, right? That was used using— here, I'm going to pull this back up again. So this is using design.md. And all I did, Neil, was I took their design.md. I went to GitHub, okay? I took their repo for this and I said, hey, I want you to make a design.md for ClickFlow, for Single Grain, for Single Brain. and for Carrot, and it has a design.md for each one. And this does look like the ClickFlow website. And so my, my, my whole thing is this, you know, if you and I as marketers, we want to spin up ads, we want to spin up carousels, we want to spin up landing pages, whatever it is that we want to do now, this makes it even faster, which is why I— and by the way, when you have a design.md, it also helps agents discover it as well, right? And so I think this is very big for marketers, and I don't think many people are talking about it. So I think most brands are going to be behind. Because they aren't AI-readable yet. I think this part of it will help change that.

So for anyone who's listening that's an agency, this could actually be very useful to you, and here's why. Not just what Eric's talking about, but there was an agency recently who had around 40% of their revenue from AI consulting and integrating all these new types of protocols. They just sold for a 30+ times multiple of EBITDA.

30+ times?

I think it was 32 or 33 or something, 30+ times.

Wow.

Fully 40% of their revenue was coming from this kind of stuff. And there was some structure on the deal, but a large chunk, if not—

How much structure?

Well, majority of the money was upfront.

Okay, that's great structure.

This is what I was at least told. I would assume around 70% of it upfront.

Yeah, that's great. That's a great structure, guys.

Even if 60% was upfront, it's such a good deal.

You know, it's funny. So we're close to hiring this person right now. And then he's like, yeah, the current company that they're at, the private equity firm doesn't believe in AI, right? So he's the one that's trying to push AI fluency. And yet at the same time, You have these multiples that are floating around. So look, I'll just put it to you this way. The classic playbook now is you basically take your ad variants, all the landing pages, you feed it into this design.md, you have an agent take care of it. It can make— you go from like 1 asset to 500 assets. The key thing, Neil, with this is whenever you're making this stuff, like these slides, you want consistency. And there hasn't been the ability to have good consistency, at least for me, until now. Right.

So, but dude, this is something that's interesting for you going back to the agency part because your agency, I don't mean this in a bad way. I mean it in a positive way because you have less headcount and because you have a smaller revenue total than larger agencies, it's easier for you to get your percentage up to like 40, 50% to some of this kind of stuff.

40, 50x.

No, 40, 50% being AI related.

Oh yeah, yeah, yeah, yeah, yeah.

And then, right, because percentage-wise, it should be easier for you. And if your contracts state that and you're actually doing that work, you can go to market. You know, I'm not going to give Eric's numbers. I don't know them, but let's just use simple numbers. If, and I'm just going to use clean, easy numbers. If he's doing $10 million in revenue, $2 million in profit, and I'm not saying that's the numbers, I'm just using round numbers. You know, 30x multiple on $2 million is $60 million. And on the $60 million, even if they give you $40 million upfront or $30 million upfront, that's such amazing—

This podcast is going to make everyone into an AI agency.

Yes.

That's what you're doing right now, Neil. But no, Neil's not wrong. And here's the thing, I actually think there's an advantage to being really like for startups right now, for example, or even smaller companies because you can make the changes more swiftly. Now, that being said, there's trade-offs to it, right? But I will just say that I have said internally in my company, I was like, what a good time to be alive right now, right? What a good time to be exactly where we are right now because we can do what we want. And like, I've also said this too, in terms of AI fluency at my company, I push it so hard. There's AI fluency standup every day. I'm in that standup every day, right?

Yeah.

And I'm just like, guys, like, if you become better, great. If you quit, great, right? Like, I will take both outcomes. And so I'm like, The train has left, guys. It's left many, many, many, many times. And the analogy I also use, Neil, is like, if you and I were riding horses, do you know how long it would take for us to get to New York from LA?

New York to LA on horses?

Horseback. You and I riding a horse together.

2 months, 3 months?

Exactly 3 months. Okay. 90 days. Okay. Now, if you and I got in a truck, a pickup truck, how long does it take for us to go from LA to New York?

Train LA to New York, 1 week. A truck, you're saying? Like nonstop?

Yeah. You might have to stop for pee and food and all that.

3 days, 4 days?

Exactly, yes. 4 days. Okay. So, 22.5 times faster. So, the analogy I give people now, because a lot of people talk about how they work in like 2025 clock speed, right? I say, look, man, you might've been in a horse last year and you might have really strong strategic thinking and all that, but you're still on a horse right now. Now, if you get in a pickup truck, not only can you drive way faster, your capacity is a lot more. You can carry a much bigger load now, right? And so that's my analogy for everyone. I'm like, right guys, let's all come along, right? Let's all come along. And I want everyone to get better. And then if you quit, also good.

Yeah.

So guys, that is it for today. Please let us know what you think about this new format, new-ish format, and we'll see you tomorrow.