Token Spend is OUT OF CONTROL! Tricks to save your company!

[00:00:00]So, picture this. Uber set an entire AI coding budget for all of 2026 and burned through it in 4 months. Their own CTO torched $1,200 in tokens during a 2-hour internal demo. And the CEO called the moment that we heard the real number head exploding. Now, here's the part that should stop you cold. There are companies right now paying more AI for more for AI tokens than they are paying the people that they laid off to make room for AI. That bill that everyone's pretending wasn't coming, well, it's here. And the one end analyst found that in the rush to max out token spend, teams are wasting over 44% on just on bug fixes. So, today I'm going to break down some ways that your token spend has gone out of control and the tricks to claw it back before finance comes knocking on your door. How long do you think your company's technology budget will last when developers start racking up $15,000 monthly bills for a single coding agent? The tech industry is spending a staggering [music] $1.4 trillion on AI infrastructure, but only generated $613 billion in actual returns. Companies are spending more money on cloud token costs than they were on the employees they recently let go. The free ride of subsidized AI is over and the massive bill is finally coming due for enterprise teams everywhere. I'm going to help you out and let you guys know some great solutions for this. Let's dive into this topic today.

[00:01:23]Welcome to Startup Pack. I'm Spencer and here at Startup Pack we love to build custom software solutions for companies.

[00:01:27]With a decade of executive leadership as a fractional CTO and 25 years in software development, I help transform tech teams and products, including building out custom AI solutions. So, let's talk about the number nobody wants to look at. For 2 years the AI labs sold everybody on intelligence is too cheap to meter.

[00:01:43]Now, Sam Altman said it, people believed it and 2026 the meter showed up and it reads like an air conditioning bill in Phoenix in July and trust me, I know about that. I live in Phoenix.

[00:01:53]The subsidies are drying up, the caps are coming due, and the whole lot of teams are about to find out exactly how much their free coding agent actually costs. Now, the massive token reckoning is here and it's catching management teams completely off guard. The days of trading cloud agents like an infinite unmuted resource are crashing down fast.

[00:02:10]Before we get into the details here, please make sure you leave me a comment down below. It's absolutely my favorite thing. Reading your comments is genuinely one of my favorite. Now, one of the other things too, we're going to having a free giveaway of one of these inference boxes use it running an open mono agent.ai. So, stay till the end cuz I'm going to give you the details on that.

[00:02:28]Now, let's dive into some of these articles that I've got here because these are some good ones that I want to dig through with you guys today.

[00:02:35]All right. Of course, my buddy here is Simon Willison. Uber caps usage of AI tools like Claude code to manage costs.

[00:02:42]Now, we talked about this here, but he's saying they're limiting $1,500 monthly per tool strike as a rational policy.

[00:02:49]That's $1,500 monthly per employee per software development. Think about that for a minute. That's an insane amount.

[00:02:57]>> [cough and clears throat] >> Two years ago, if I had told you that we'd be paying $1,500 a month for any tool per month, that's incredible. But, that's what Uber's capping this spend at. That's just absolutely insane to me.

[00:03:09]All right. The token reckoning is here.

[00:03:10]It's not what you think. Token waste is a service of product of product productive use, not waste for the sake of waste could be the biggest problem in AI right now.

[00:03:19]Basically, he's saying here that because we if we go and cap everything here, exactly a year later at cost we're looking at less like a cheap intelligence and more like an AC bill in Phoenix. That's actually where I got that from. But, it's actually pretty funny cuz that's actually what it is.

[00:03:31]Now, breaking. Anthropic gets sued for fraud. A customer filed a federal class action lawsuit Monday. And if you don't think there's going to be a lot of people jumping on this one, claiming he had 15% of weekly allowance in just one 5-hour sprint. The actual usage provided by the max 5x and max 20 plan is far below the advertised amount.

[00:03:48]They're calling it a marketing fraud.

[00:03:50]So, Dario, buckle up cuz I actually think this class action lawsuit could get some legs.

[00:03:56]So, this guy says, "Yeah, selling $14,000 for $200 a month, a masterful gambit, Mr. Altman. Very good stuff."

[00:04:02]Cuz recently we purchased one of Anthropic and Open AI subscription plans and randomly ran long horizon coding tasks using until we exhausted the weekly limit.

[00:04:11]It's why I believe that a $200 per month plan maxes out at $2,000 worth, but they actually found that they could get a $200 a month to go to $14,000 worth of token spend. If you don't think Anthropic and Open AI are about to pull rug pull this, no way they're going to run these economics. Uh and so, look at this. A $20 per month Claude code can run almost $400 per month of token cost, $200 per month 8,000. On Open AI, $200 per month 14,000. You think that's going to survive the CFO uh going into IPO?

[00:04:42]Dream on, little broomstick cowboy.

[00:04:45]Anthropic is extremely expensive, and I think many people are urgently looking for alternatives. This is Microsoft's head of AI.

[00:04:51]I've got an alternative for you. Uh we've got one here.

[00:04:55]It's called openmonoagent.ai. We've built this ourselves. It's fantastic.

[00:04:59]It's growing very quickly. Uh you don't believe me, we're already up to uh you know, 106 1.6 thousand stars on on GitHub. Growing very quickly. AI should not have a meter. Should be unlimited tokens forever. Here's our democratized thesis. AI should be a shouldn't be a subscription you rent, it should be infrastructure you own sitting on your desk serving your code answering only to you.

[00:05:21]Real deal. We're actually rolling out VS Code plugins this week as well as web search. Shout out to the guys for building that awesome stuff. Team's been working hard.

[00:05:31]All right.

[00:05:32]The era of subsidized AI may soon be coming to an end. I think you can just get rid of the may and says is going to an end. As SpaceX, Open AI, and Anthropic look to IPO, the focus will inevitably shift from growth at all costs to show me the EPS and profitability.

[00:05:45]For most of 2024 and 2025, AI agents are expensive was the thesis. Here is the leaked memo and enthusiast forum sorry, there was an antidote. A leaked memo here and enthusiast post there, the numbers were not concrete enough to publish, but now we're seeing numbers like this.

[00:06:02]And this tells you everything you need to know.

[00:06:05]So, the numbers aren't going to be like there for very long, right? Microsoft canceled internal cloud code license this week after token-based billing made the cost untenable even for a company with effectively infinite cloud resources. Uber CTO sent an internal memo warning the company burned through its entire 26 AI budget in just 4 months.

[00:06:25]It appears that OpenAI has moved all Codex users to token-based billing using a credit system aligned with API pricing. Some companies had received a 2-month-long introductory period are now receiving a limited amount of pooled credits per user.

[00:06:37]So, they go through and look at the Codex rate. So, even Codex is starting to cap this saying, "Hey, if you want to see this, this is how much credits you get."

[00:06:45]That's not going to last very long.

[00:06:47]Okay.

[00:06:49]Um So, I love that everybody's like, "Dang, you make investments in a business for some sort of return and you make money, too." So, ultimately the question is whether the industry's massive AI spending will deliver meaningful payoff.

[00:07:01]At some point, you have to make money.

[00:07:02]You make investments in a business so that you can generate returns and make money. Surprise!

[00:07:08]So, Business Insider says, "In a lot of ways, companies are losing more money today implementing this technology than they were 2 years ago." said Goldman Sachs head of low global equity serve research.

[00:07:20]Good stuff here. Okay. All right, I like this one cuz this was a chuckle.

[00:07:24]$15,000 out of your checking account. Is it what I think it is? Somebody thinks they're getting a ring. Nope, they're paying their Claude Code bill.

[00:07:32]All right. First, it was MIT and McKinsey. Now, Bain finds that returns to corporate AI investments are disappointing.

[00:07:39]AI investments are circular bet as ROI disappoints, Bain survey shows.

[00:07:43]We can click on this, but we're going to get blocked and you know, that's the way it goes. So, hearing that T-Mobile attempted to limit its Claude enterprise users as to as little as $30 a month from a cap of $2,000 a month in tokens, but faced immediate outcry. They're now saying $2,000 is a temporary number and that tiered spending caps will be introduced by the end of the week.

[00:08:02]Again, I can't even fathom like a $2,000 bill per employee.

[00:08:07]Uh somebody just is really anyways. The industry reportedly spent 1.4 trillion and still isn't profitable. A website called is AI profitable is tracking the economics of AI doom boom, sorry. I think I said doom. That was a little for for reading slip. To determine if it's prof- profitable and answer is surprise, no.

[00:08:25]As of May 2026, AI industry has spent roughly 1.4 trillion on model development, data centers, chips, networking and other AI infrastructure over the same period. It's generated about 630 13 billion dollars in revenue.

[00:08:36]Biggest losses belong to the leading companies. Boom boom boom boom all companies we recognize, right?

[00:08:41]According to the dashboard, Nvidia has generated an estimated 478 billion in AI revenue against 225 billion in AI related spending for roughly a profit of 215. So, the person who's made money, it's There you go. It's uh Nvidia, right?

[00:08:55]They're the only ones who've made money in this.

[00:08:57]The figures are compiled from public filings, earning reports, analyst analysis estimates, etc. The estimates also exclude indirect benefits from AI such as improved search, advan- advertising and software sales.

[00:09:09]So, it's kind of tricky because somebody like Google is still making money hand-over-fist on their advertising department, but they're actually one of the few who could possibly say that that's actually helping.

[00:09:18]Now, that's about 18,000 euro 90 million dollars a year. Okay, Uber sets a $1,500 a month cap on some AI coding tools for staff.

[00:09:27]Uber officially reeling in the Claude budget after blowing their AI budget earlier this year. So, again, it's about 90 million dollars spent. Man, I hope Uber can generate 90 million dollars of revenue more on that. I'd be pretty mad if I was the CFO.

[00:09:40]OpenAI's Sam Altman says the AI spending issue has kind of become a meme, adding that token costs went went from an issue that never came up to all of a sudden a huge issue. Surprise.

[00:09:51]In Ed here says, "Yeah, almost like every AI company obfuscated and subsidized the cost for years and suddenly decided the enterprises they need uh that they needed to pay the token rates."

[00:10:02]So, Sam Altman warns OpenAI and Anthropic are experiencing severe pullbacks on AI spending as companies put significant restraints on spending to restrict costs.

[00:10:10]The company warns investors, "It's the first time it's happened in AI and something we never expected." This part is just what gets me. Something that we never expected. The build-out costs are are aren't sustainable to allow profitability to hyperscalers or end users.

[00:10:25]Uh this is like and that's where I guess it's just crazy to me. And why we reason why we built openmono.ai is because we want people to be able to build and understand that within we're we're within 3 to 5% of the large frontier models totally for free.

[00:10:39]You don't need a crazy hardware on this.

[00:10:41]All you have to do is go here. You can check out this is the hardware you need.

[00:10:45]Not a lot of tough hardware, right? You can even use one of these little nook boxes, which like I mentioned earlier, we are doing a free giveaway on. So, you definitely want to make sure you take advantage of this. Come here to openmonoagent.ai, click on the free giveaway, enter to make sure you went enter to win for this free inference box that we're giving.

[00:11:04]Cuz see here at Starter Pack, we really believe in building custom software solutions. And we love building these out for our clients and building infrastructure that they own from top to bottom.

[00:11:13]So, one of the big reasons people stay glued to expensive cloud agents is the agent can pull in some fresh information. So, we just integrated web search into openmonoagent. And this is great because now your local agent can reach current information out on the web without you shipping your whole workflow off to someone else's server.

[00:11:30]So, you can get the capability you actually wanted without the per token capability tax that you didn't want.

[00:11:36]It's the same practical philosophy that I bring to every client. Real utility, not a feature you brag about and never use. Local doesn't have to mean limited, and we just proved it with Open Mono Agent.

[00:11:46]The whole token reckoning happened because pricing was opaque and could change on a whim. Open Mono Agent is open source, and it's hosted right on your own server. So, you can read exactly what it does, you can fork it, you can run it without trusting a single marketing claim.

[00:12:01]There's no intro period quietly ending and no surprise migration to credit-based billing 6 months later.

[00:12:06]The apprenticeships and boot like the we onboard people and teach even our juniors how to use this tool. It's very simple to get set up.

[00:12:16]So, I keep saying that skill beats hype, and the best way to build skill is to actually get your hands on the tool and start using it. That's why we're running the free giveaway because we with the link that we that I showed you before, because we want you to be able to get your hands on this. But, you can still do it. You can download and install this This box you see right behind here, it's a $350 box I got off of Facebook Marketplace. I put a $1,000 GPU into it.

[00:12:38]Boom, I'm running my own AI server that can comfortably run a massive amount of AI workload, all for free, just for the cost of electricity.

[00:12:47]Now, I'm curious to hear what you guys think. Do you agree? Do you disagree?

[00:12:50]Love to have a great conversation. Make sure you like and subscribe. Check out startup hack.com if we can help you out in building custom software solutions, and check out openmonoagent.ai and click on the free giveaway because we'd love for you to be able to enter in the free giveaway, and we will catch you guys to