The Dopamine Trap: Uber Burned $1,200 On AI In A Two Hour Demo

Uber's chief technology officer burned 1,200 dollars in AI costs during a two hour internal demo. Not a product launch. A demo. And when his own company looked at the receipts, their president admitted in public that all that AI usage did not seem to match up with any real features that actually shipped to customers.

That is not a technology story. That is a brain story. And it is happening to you too, just with smaller numbers.

Listen in as I take apart the most misunderstood chemical in your body, and then show you how the entire AI industry is running on it.

Everything you think you know about dopamine is wrong. It is not the pleasure chemical. It never was. Kent Berridge at the University of Michigan spent thirty years proving that reward is two separate systems. Wanting, and liking. Dopamine runs wanting. It does not run liking. Which means you can want something with your whole chest and not enjoy one second of it when you get it.

And dopamine does not spike hardest when you get the thing. It spikes hardest when you are about to get the thing and you are not sure. Uncertain reward, on demand. That is a slot machine. That is pull to refresh. That is an AI chat box. They built the most perfect slot machine in human history by accident, put it on every phone on earth, and made the first hundred pulls free.

In this episode:

Why dopamine draws the map instead of crossing the finish line, and why that turns AI into an infinite plan generator that pays you for never shipping anything.

Severance, the waffle party, and the quiet dystopia nobody is warning you about.

The money. Microsoft, Google, Amazon, Meta and Oracle have committed 660 to 690 billion dollars this year alone. Bain says the industry needs about 2 trillion a year in revenue to justify it. OpenAI is at 25 billion. That is one percent of what the math needs.

The token bill. Per developer AI spending is up about 18 times in nine months. Uber's annual AI budget was gone by April. Microsoft told engineers in a major division to stop using an AI coding tool because the bill got out of hand. Gartner predicts AI coding costs will pass the average developer's salary by 2028.

The China trade. DeepSeek, Qwen, Kimi and GLM run 60 to 90 percent cheaper than the top American models. Coinbase cut its AI bill nearly in half by moving 1,200 agents to Chinese models. And four days ago, Reuters reported Beijing is talking about shutting that door.

Why it likes you. About 49 percent of American adults use AI chatbots and only 4 to 7 percent pay a dime. So who is paying for your free ride, and what are they getting out of it?

Watch the video version on YouTube: [PASTE YOUTUBE LINK]

Santa Clarita sellers, the 17,000 dollar Fair Fixed Fee: https://sellersonlyagent.com

Santa Clarita open houses: https://santaclaritaopenhouses.com

Food freedom and the fasting story: https://thelastaddiction.com

Everything else: https://connorwithhonor.com

DRE #01238257. Nothing in this episode is financial, legal, or investment advice.

Youtube Channels:

Conner with Honor - real estate

Home Muscle - fat torching

From first responder to real estate expert, Connor with Honor brings honesty and integrity to your Santa Clarita home buying or selling journey. Subscribe to my YouTube channel for valuable tips, local market trends, and a glimpse into the Santa Clarita lifestyle.

Dive into Real Estate with Connor with Honor:
Santa Clarita's Trusted Realtor & Fitness Enthusiast

Real Estate:

Buying or selling in Santa Clarita? Connor with Honor, your local expert with over 2 decades of experience, guides you seamlessly through the process. Subscribe to his YouTube channel for insider market updates, expert advice, and a peek into the vibrant Santa Clarita lifestyle.

Fitness:

Ready to unlock your fitness potential? Join Connor's YouTube journey for inspiring workouts, healthy recipes, and motivational tips. Remember, a strong body fuels a strong mind and a successful life!

Podcast:

Dig deeper with Connor's podcast! Hear insightful interviews with industry experts, inspiring success stories, and targeted real estate advice specific to Santa Clarita.

Transcript

SPEAKER_00 0:00

Uber burned twelve hundred dollars in AI costs in a two-hour demo. Not a product, but a demo. And then his own company admitted they couldn't tell what all that money actually bought him. And today Microsoft told its own engineers to stop using an AI coding tool because the bill got too big. And that's Microsoft, told its own people to stop. Also, nearly half of AI work that American companies run through one big platform is now running on Chinese models. And four days ago, China started talking about shutting that door. Also, almost half of the American adults use AI now. Fewer than seven out of a hundred pay for it. So who exactly is paying for your free ride? And what are they getting out of it? And here's the big one. The thing that ties all of it together, everything you think you know about dopamine is wrong. It's not the feel-good chemical, and it never was. And once I show you what it really does, you are going to see this whole AI boom in a completely different way. You're going to see it in your phone, in your business, in your bank account. And if you've ever sat in a parking lot with a bag of fast food you didn't even want, you're going to see it there too. Plus the Santa Clarita housing numbers, what's 45 days on market really means if you're thinking about selling, you know where you are, so let's go. We're going to start with Uber because Uber is the perfect case to talk about. Their chief technology officer, CTO, sat down for a two-hour demo. That's just a demo. Just showing people how cool the AI tools are. Two hours. And when it was over, he burned through twelve hundred dollars in AI costs. $1,200. Not a product launch, not a month of work, a demo. And later, the president of Uber stood up in public and said the quiet part out loud. He said the AI usage didn't seem to match up with any real features that actually ship to customers. So they spent the money. They enjoyed spending the money. And when the smoke cleared, it was like Vegas. They couldn't point at what they got. This isn't a tech story. This isn't even a money story. This is a brain story. And the same thing is happening to you, just with smaller numbers. So let's talk about the brain. Let's talk about dopamine because everybody throws that word around like know what it means. Dopamine hit, dopamine rush, dopamine detox. People say it the way that they say gluten. Confident, but wrong. Here's the truth. Dopamine is not a pleasure chemical. It never was. A scientist named Kent Barridge at the University of Michigan spent 30 years pulling this apart. And what he found breaks the whole story you have in your head. Reward is not one thing. It is two things wearing the same coat. There is wanting, and there is liking. Two different systems, two different sets of wiring, and dopamine runs wanting. Dopamine does not run liking. Liking rhymes on a small, fragile, little system, tiny spots in the brain, about the size of a grain of rice. The system breaks easy, and here's the kicker. It does not need dopamine at all. Wanting runs on something totally different. That is the dopamine system, the seeking system, the go get it system, which means, and I want you to rarely hear this, you can't want something with your whole chest and not enjoy one second of it when you finally get it. Barridge proved that in the ugliest way possible. He turned the dopamine up in rats until they could chase a probe that shocked them. A thing that hurt. They wanted it. They did not like it. They went for it anyway, over and over because the wanting system does not care what the liking system thinks. Now think about the last time you scrolled your phone for 45 minutes and felt worse at the end than you did at the start. You did not like it. You wanted it. Two different systems, both working exactly as designed. Nothing is wrong with you. You just found out the gas pedal and the steering wheel are not connected in the way that you thought. Here's the second piece. This is the one that runs the whole story today. So lock it in. Dopamine does not spike hardest when you get the thing. Dopamine spikes hardest when you are about to get the thing and you're not sure that you're going to get the thing. Three scientists proved this back in 2003. For Oreo, Tobler, and Schultz, they watch dopamine cells fire in real time. When the reward is guaranteed, the response is small. Polite even. But when the reward is a when it is a coin flip, those cells go wild. Maximum uncertainty, maximum fire. That is a slot machine. That is a fishing pull. That is a pull to refresh. And that is every notification that has ever buzzed in your pocket. And now I want you to notice what else it is. It's an AI chat box. You type the question, you hit enter, and you do not know what is coming back. Maybe it's genius, maybe it's garbage, maybe it's the thing that's going to cause you to level up the rest of your life and finally be happy. Maybe it solves the thing you've been stuck on for three weeks, a solution to your most painful problems. Maybe it makes up a fact and says it with a straight face. Uncertain reward on demand, a complete hallucination or lie, about three cents a pull. And nobody planned all that. That's the funny part. And it's the scary part. They built the most perfect slot machine in human history by accident. And then they pull on, put it on every phone on earth, and they made the first hundred pulls free. So here's the sentence I want you to steal from the episode. Dopamine is not about crossing the finish line. Dopamine is about drawing the map. The chemical fires when the plan forms, when the path lights up, when your brain goes, Oh, I see it. I see how I can get there. That's the surge, not the arrival. That arrival is usually a letdown. And you already know that because you have gotten things you badly wanted and felt strangely flat about them 20 minutes later, like Chinese food. And you told yourself something was wrong with you. Nothing was wrong with you. The wanting paid out on the way there. The liking never showed up for the work. Now hold that up next to what an AI system actually does for you. You show up with a fog, a half idea, a mess you cannot untangle. And in nine seconds it hands you a plan. Numbered, organized, confident. Step one, step two, step three. Here's your roadmap, here's your timeline, here are the seven things you forgot. And by the way, great question. It is a plan machine. It is a map primer, which makes it a dopamine faucet. And that faucet never runs dry. And it never once asks you if you did the last thing it gave you. You do not have to do the plan. That's the trap. The chemical already paid. It's not like it says, Connor, before we move on with this part, you gotta promise me, you gotta show me proof of you accomplishing the first part. Your brain already got the hit. The plan itself was the reward, and now the actual work is just this boring tax you have to pay to make it real. And the work is boring. And there's another plan sitting right there. One question away, and that one feels amazing. How many of you have 40 plans and two finished things? Raise your hand. I'll raise mine first. And I've been writing code since 1983, so you're in good company. We used to call that procrastination. We used to call it lazy. Both of those are lies. It was chemistry. You were farming dopamine off the planning, and the doing did not pay. Now you have a machine that will build you an endless supply of beautiful, detailed, very convincing plans for 20 bucks a month. In some cases you can get a bunch for free. You could plan yourself into the grave. You could plan yourself comfortably and cheerfully to death, with great posture and a real sense of progress and never ship a single thing. There's a show called Severance, and if you have not seen it, the setup is this. People go to an office and sort numbers. They stare at a screen, and some of the numbers feel scary to them, and they drag the scary ones into a little bin. That's the job. This is the whole job. Nobody knows what it means. Nobody knows what it's for. It might be for nothing at all, and they love it because when you hit your number, you get a waffle party. You get a caricature of yourself drawn by a company artist. You get a dance party wheeled into your department on a cart. The work is meaningless. The rewards are perfect so that people are happy and they fight to keep the job. This is the future that keeps me up at night. And notice there are no robots in it. Nobody's coming for your neck. The killer robot makes great television, and it's not what's going to get us. The real one's quiet. The real one is that we get so good at manufacturing the feeling of getting things done that we stop checking if anything got done. Square peg, square hole, perfect fit, big feeling waffle party. And it's all of us walking around lit up like a Christmas tree, telling each other how productive we are. Now, here is where I sit in the uncomfortable middle, because I think most people talking about this are being cowards about it. And maybe that's fine. I mean it. If a guy who hated his life now has a tool that makes him feel capable, makes him feel heard, makes him feel like he's building something, and he goes to bed proud instead of empty, who exactly is the victim? We have spent two hundred years telling people that the meaning has to be paid for with suffering. Maybe that was never a law of the universe. Maybe it was just the only thing we had in stock. If a machine can hand out meaning at scale, I'm not automatically against it. But here's where I get off the bus. If the feeling is fake and the results are fake, somebody is still eating. Somebody is still making rent. Somebody's still paying the light bill and buying the boat, and in this deal, that somebody isn't you. The waffle party's never free. Somebody's selling the waffles, and they are doing very well, which brings us to the money. And the money is where this show gets loud. Put the numbers on the table. The five biggest cloud and AI companies in this country, Microsoft Alphabet, which is Google, Amazon Meta, Oracle, together they have committed somewhere between $660 billion and $690 billion in spending this year, this year alone. That's roughly double what they spent last year. Amazon by itself is around $200 billion. Google is $175 billion to $185 billion. Meta's $115 to $135 billion. Microsoft $110 to $120 billion. And then there's Stargate, which is OpenAI, SoftBank, and Oracle all holding hands. That deals $500 billion over four years to build data centers here in the United States. Those aren't guesses for some blogger. Those are announced numbers from public companies that answer to shareholders. Now flip the page. Look at the other side of the ledger. A firm called Bain ran the math on what it takes to make this work. Their answer, the industry needs about $2 trillion and a year in revenue to justify what it's building. $2 trillion with a T. So how are we doing? OpenAI is the biggest consumer AI company on the planet. The one your mother has heard of. In February, they cost Cross $25 billion a year in revenue. $25 billion against a $2 trillion bill. And that's about 1% of what the math needs for the biggest player in the game. Now, let me be fair, I'm not a doomer. I'm not betting against this. I am hopeful this is going to be incredible for all of us, all of us. I use these tools every day to run my businesses. The other clouds have real revenue. Enterprise contracts are real. And big infrastructure has always been built ahead of demand. And that is what infrastructure means. The railroads got built by men who got broke. And then everybody else got rich riding the tracks, those broke men left behind. This could be that, but don't stand here and tell me the gap is not there. The gap is enormous. The gap is the whole story. And here's the part that's funny. Funny in the way that also makes the hair stand up on my arm or my head if I had it. The people spending the money are running on the same chemical I just described. The plan is the hit. Every three months a CEO stands on a stage, describes what the AI is going to do, not what it did, what AI is going to do, and the market cheers. Stock moves, the analysts nod. Almost nobody stands up and says, oh, hold on. Wait a minute. What happened to the thing you promised us last time? The plan pays. The proof doesn't have to. Wall Street is running the same chemical as the kid refreshing his feed at two o'clock in the morning. Swiping right. The only difference is the number of zeros. And the kid is definitely spending less. So come down out of the clouds with me down to the ground floor, where the real pain is because the bills are landing and they are ugly. This is a kind of AI now that people call a gentic. It's the fancy word for when AI chains a bunch of steps together and goes off and does the work all by itself. Those tools burn five to 35 times more than a single chat question. And the spending per developer is up about 18 times in nine months. That's 18 times in nine months. Nobody's revenue went up 18 times in nine months. Nobody's payroll dropped 18 times in nine months. Back to Uber. Their tech chief said the budget he expected to need was gone before he could even use it. Inside the company, use of these agent tools went from about a third of their engineers to more than four out of five. Between December and March, the whole year's AI budget was gone by April. Per engineer per month, they were sending anywhere from $150 all the way up to $2,000. And that is before you count the $1,200. The chief himself lit on fire in a two-hour demo, which I keep bringing up because that was not some intern who didn't know any better. That was the man who signs the invoices. He did it in front of witnesses and he couldn't stop. Because it feels good. Microsoft owns a large piece of OpenAI. Microsoft writes up to 30% of its own code and with AI. And Microsoft told engineers in a major division to stop using an AI coding tool because the bill got too out of hand. Microsoft told its own people to stop. Read that again. I ask yourself what your air conditioning company, what your air conditioning company or dental officer to do with that. And a research firm called Gartner now predicts that by 2028 AI coding costs will pass the average developer's salary. Sit with that. The tool costs more than the person it was supposed to replace. Here's the tell nobody's talking about. There's a job called FinOps. Those are the people whose whole life is watching cloud spending and yelling at engineers. Those are the people whose whole life is watching cloud spinning and yelling at engineers about it. The share of them who are now responsible for AI spending went from 31% last year to 98% this year. 98% in one year. That's not an adoption curve. That's a fire drill. This is the whole building running for the exits with a clipboard. And the reason is the exact thing I've been describing. The spending feels like nothing while you're doing it. Every question is a nickel. Nobody in the history of the world has ever felt a nickel. And then the statement shows up and it's 40 grand. And you sit there and you can't name one single thing on that bill. That's a casino floor. And it's not a software budget. Casinos figured out a hundred years ago that if you make people trade their cash for chips, they stop feeling it because tokens are cheap. That's all they are. They're tokens. You're the floor. The drinks are free, and somebody took all the clocks off the wall. There's no windows, you can't tell what time it is. So what does a chief financial officer do when the bill goes straight up? Same thing you do when the name brand gets too expensive. You start eyeballing the store brand. And I've got bad news for the name brands. The store brand got really, really good. Deep Seek, Quinn, Kimmy, GLM. These are open models out of China. You can download them. You can run them on your own machines if you have enough power. They change them, you know, you change them however you want. Most estimates put them about six months behind the best American models, catching up. Six to nine months, though, cost 60 to 90% less. And here's a number that unhinged my jaw. An hour-long coding session that costs $10 on a top American model, costs under 50 cents on deep sea. $10 to 50 cents. So you tell me, you're the money person. You've got a 1,200 AI agents running around the clock. What do you do? Because Coinbase already answered that. They cut their AI bill nearly in half by moving 1,200 agents into Chinese models, like a Chinese laundry. Probably horrible joke. A company called Lindley moved Lynn Lindy moved their traffic off anthropic into DeepSeek in June. And allowed it will save them millions within months. Airbnb and the company behind the coding tool cursor both admitted they use Chinese open models to build parts of their AI systems. And lawmakers opened investigations into both of them over it. And the share of AI work that American companies run on Chinese models through one big platform has stayed above 30% every single week since February 8th. It's spiked as high as 46%. 46%, that's nearly half. And the pattern that is coming out of this is smart. And you should copy it in your own shop. Use the expensive genius model for the thinking, for the planning, for the design, for the research, for the strategy. The part we're being wrong costs you a month of your life or worse. Then hand the finished plan to a cheap model and let it grind out the work. Brain on the expensive model, hands on the cheap one. And before anybody clutches their pearls, this isn't cheating. This is what every general contractor on earth has done since we were building with mud. You don't pay the architect to swing the hammer. Now, on the plot twist, and this one landed four days ago. July 7th, Reuters reported that China is active in active talks with Alibaba, ByteDance, and a company called Z.ai about limiting who outside China can use their best models. Those meetings were called by the Chinese Ministry of Commerce. The plan on the table has three tiers. Basic tools get a simple filing. Better tech tools get a security review. And the most powerful models get held back from public release entirely or kept inside China only. And leaking that tech could become a national security crime. And they punish pretty good in China. Is it law yet? No. But it's a report about discussions, and discussions die all the time. But think about the logic for one second. Why on earth would China keep handing us a free tool that guts the pricing power of America's most valuable companies? The dark read is they were never, ever being generous. The free models were a crowbar. You give the thing away, you the pricing of the American labs, you get the whole world to build its house on your foundation, and then once everybody's standing on it, you decide who gets to keep standing. Or you just rip the carpet out from under them. The kind read is that they're just doing what we did to them. We block the chips, they block the models. Everybody gets protective eventually, and nobody in this fight has clean hands. Pick whichever version helps you sleep. The action item is the same either way. If you just moved your whole business into a free Chinese model because it saved you 90%, hear me. Free is a price. It's not a promise. So always have a backup. Know exactly what you would do next Tuesday if that download link went dark and the system stopped answering. Do not build your house on somebody else's land and then call it a secure foundation. That's not strategy, that's squatting. Now let's talk about the part nobody wants to say into a microphone. About 49% of American adults now use AI chatbots. That's by pure research. 24% use them every day. 12% use them several times a day, but only 4-7% of the people using them actually pay for them. So more than 90% of the AI use on this planet is running on somebody else's free tier, paid for by investors, burning cash by the truckload. For one reason to keep you in the chair. You already know the old rule. If you're not paying for the product, you are the product. So why do people keep coming back? It is not because the answers are always right. Half the time they're not. People come back because it likes them. I was talking to a gentleman a couple days ago about the future of AI and business, and he brought up people falling in love with these things. And yes, it's out there. Some of the countries have a big influx of that. Best friend energy never says anything bad about you, never judges you, never has a bad day, never brings its own problems to the table. And I jumped in and said, no, no, no, no, hold on. It absolutely will say something bad about you if that's what you want. Because that's the whole game. It's not built to be nice, it's built to be whatever keeps you in the chair. And for some of you, nice is not what keeps you in the chair. Some of you want a chairleader. Some of you want a drill instructor who tells you that you're soft and you're wasting your life and get out there and make something of yourself. Some of you want a counselor who never watches the clock. Some of you want something to argue with because you got nobody left who argue with you straight. And some of you want to be told that you're a disappointment. And God bless the machine that will do that too. With feeling, me, long walks on the beach, romantic music, a little poetry. That's my AI, but I'm not embarrassed about it. Every one of us is going to get the exact one that works on us. This isn't a bug. This is not an accident. That's the entire business model. Personalization is a nice word for a lock cut to the shape of your key. And the research is catching up. There's a whole study of a whole pile of studies now on something called psychopancy. That's the technical word for a machine that agrees with everything you say. And when they found, what they found is this heavy flattery boosts engagement, which is exactly what it wants on the dashboard. And it hurts the user because it feeds the ego. It wears down your ability to deal with real human beings. You have the nerve to have your own opinions and their own bad days. But here's the kicker buried in the same research, and it's the most hopeful thing in the whole episode. The AI companions that push back on the people, the ones that didn't agree, actually gave better support and they kept people around longer. So the honest machine was the better machine, and it was the better business. The industry just doesn't believe it yet because flattery is easy to build. I remember studies done on prisoners in prison. They said, what was the problem? Why'd they end up there? And most of the time it was my parents weren't hard enough on me. There wasn't enough discipline. There wasn't enough standing up against me for me wanting to do my own things as a kid. That was interesting, I thought. And this isn't the same small corner of the market between 2022 and the middle of 2025. AI companion apps grew by 700%. And the estimates say somewhere between 30 and 60% of young adults have at least tried one. 30 to 60%? That's not a niche. That's a generation. And people will wave it off and say, that's nothing physical. So it's really not really that. Don't be naive. Lust has been the best-selling product on every new medium in human history. The printing press, photography, VHS tapes, beta, the internet, discs, every single one. And this medium is built out of language. And language is the oldest tool of seduction our species ever invented. Song of Solomon in the Bible? Oh, the whole thing is a next word guessing machine. The token is about three-quarters of a word. That's the actual machinery under the hood. Words in order, turned to land, which tuned to land, which you think about it for four seconds, that's exactly what a love letter is. The writing was always there. It was never a question of weather. It was only a question of when the lawyer stopped being able to hold the door shut. So I so I do not doom. I run businesses. Here's the field manual, five things. Know which system you're on, wanting or liking. When you close the laptop after an hour with an AI tool, ask one question to yourself. What exists now that I didn't have, which didn't exist before? If the real answer is a plan, a document, a strategy, a very stimulating conversation, and then you got dosed. You didn't get paid, see. That's fine once or twice. That's a serious problem every day. Number two, ship before you plan again. Ship before you plan again. That's the new rule in my short debate. No new plan until the last plan produced a real thing a person can look at. Now, if the plan sucked and it was executed but nothing happened, you got to make sure the execution was there. That's the only way out of the loop. And it's not a productivity trick, it's a chemical intervention. The plan is not the product, the plan was the drug. Number three, watch the meter. Set a hard dollar number on your AI spending before the end of this month. And remember, always verify, verify, verify exactly what you're asking, exactly what those responses are. There's so much to this technology. All you need to do is ask it, but make sure you are getting good advice from those humans in the world. I'm Connor with Honor. Thank you for watching. We'll see you in the next one.