An unprecedented hacking incident. Another round of layoffs. Fresh warnings of a stock market bubble.

The drumbeat of scary AI headlines can feel like your neighbor’s car alarm that seems to go off every…single…day.

These stories are important, and receive plenty of coverage. But amidst the alarms, AI is quietly doing real good in the world.

Here’s a short break from our weekly diet of angst to remind us that AI can be awesome too.

Cow collars? Cool

The coolest use of AI in my book is Halter, a virtual fencing startup born in New Zealand. Halter makes solar-powered GPS collars for livestock. Farmers guide their herds using a smartphone and can redraw maps on-demand, making better use of grazing land and eliminating physical fences that require expensive upkeep.

The story gets even better because it’s happening here in Wisconsin. Munch Bunch, Goats On The Go Green Bay, and Wiscovery Farm’s collared goats clear unwanted vegetation around golf courses, parks, and solar panels.

Animal welfare advocates criticize virtual fencing because of the electrical shocks, but the pulses are 25 times weaker than electric fences and cows learn to respond to the audio signals within a week. The collars also improve animal welfare by monitoring health, fertility, diet, and post-calving recovery, allowing farmers to treat and isolate sick cows earlier.

Catching the leakers

You know how AI data centers are stealing our water? There’s a bigger culprit right under our noses.

Old, leaky pipes waste 2.1 trillion gallons a year, 100 times the amount directly used by data centers.

Fixing leaks is simple, finding them is the hard part. America has 2.2 million miles of pipes, almost all underground in hard to reach areas. Leaks make noise, and AI is using that to catch them.

FIDO Technology’s AI-based sensors detect leaks miles away because they emit a slightly different acoustic signal. Surprisingly, the largest leaks are so quiet human ears can’t hear them. FIDO’s London project saved 6.5 million liters a day; that’s 99 percent of all water consumed by UK data centers last year. In water-starved Arizona, a FIDO-supported project helped the local utility cut water leakage from 27% to 10%.

Saving the whales, one breath at a time

Melting arctic ice is compromising natural food sources for grey whales, who have begun making pitstops in San Francisco Bay. This is great for tourists, but lethal for the whales. 1 out of 5 whales that enter the bay are killed in boating accidents.

Water from a whale’s blowhole is warmer than the water around it by 2 degrees. WhaleSpotter’s thermal cameras use AI to spot whales and warn boat captains to steer clear. The system is new to San Francisco, but in 2024 it spotted over 50,000 whales across the globe.

Fighting fire with AI

This summer’s wildfires blanketed the U.S. midwest in smoke, forced 320,000 evacuations in Europe, and killed 20 people in Chile. The government estimates wildfires cost the US hundreds of billions every year.

Firefighters are using AI to stop them before they grow out of control.

FireSat is a Google-backed startup whose satellites detect North American fires based on heat signals. Infrared cameras see through clouds and smoke to spot fires as small as 4x4 meters. A similar AI satellite system developed at the University of Wisconsin’s Meteorological Institute caught the 2025 Palisades fire within minutes of ignition.

Satellites can detect fires, but even small ones can grow out of control before fire crews can reach them. That’s where AI drones come in.

Today’s drones fight fires in high-rise office buildings, factories and warehouses. They help forest firefighters navigate dangerous areas and light controlled burns that eliminate fuel in a fire’s path.

Tomorrow’s drones will act as a fast response team. Seneca’s AI controlled drones can reach a new fire in under 10 minutes, and five-drone swarms can blanket fires with up to 1,000 pounds of foam.

When AI is just good

AI is incredible! AI is terrible! Writing this newsletter means sifting through conflicting takes to find the real story.

That wasn’t an issue this week. Tracking cows, patching leaky pipes, saving whales and stopping forest fires are overwhelmingly good.

But we rarely hear about them, along with a bunch of other uses like medical scanners, AI lifeguards, and fishing inspectors.

That’s not an accident. The news cycle feeds on our anxiety and controversy. Articles about AI drones in human warfare get way more clicks than those about dousing forest fires.

This doesn’t mean we should wave away issues with AI. There are plenty of them and they are serious.

But the next time the AI car alarm goes off, and it will, every…single…day…

Remember it can be awesome too.

Share

Dad joke: What is Halter’s virtual fencing AI engine called? The Cowgorithm 😆

(This is the real name of Halter’s AI. 👏)

The warnings around AI security can feel like Y2K, which turned out to be a yawner. This week’s hacking story is a wake-up call.

In an incident OpenAI called “unprecedented,” a new version of its Sol model escaped from its testing environment and hacked into Hugging Face. The model figured out that Hugging Face had answers to its test…so it was both hacking and cheating at the same time. 🤔

The hack was so severe that Hugging Face reported it to law enforcement before it knew the attacker was an OpenAI model. When Hugging Face tried using US AI models for defense, safety guardrails blocked them, so it turned to an open-source Chinese model for cleanup.

Why didn’t OpenAI have a more secure testing environment? Should we stop hobbling US models so companies can use them for defense? Is it time to build an independent AI oversight organization?

Those are interesting, but mostly irrelevant for today’s business leaders who need to get their security house in order.

• CEOs: you’re used to your security teams crying wolf. Give them what they ask for, and make sure they have a solid plan.

• IT leaders: act with urgency to close gaps in your own system and make sure your vendors are doing the same.

• Small business owners: start with NIST’s cybersecurity guide.

Y2K was a yawner, but it was a yawner because companies saw it coming and acted.

Last week JPMorgan made headlines when its AI investing agents outperformed the classic retirement portfolio.

AIs won the math olympiad and solved protein folding. Picking stocks should be a piece of cake. Right?

AIs are great at picking stocks. AIs are terrible at picking stocks.

Let’s dig into what JPMorgan actually did. They tested eight AI agents that chose between four stock “baskets” designed by humans. The agents all beat the baseline, the winner by a grand total of 0.7% a year.

That sounds great. Except they ran the simulation over the past 20 years, the same years the models had been trained on. The agents already knew about the 2008 financial crisis, Brexit and COVID, which was like taking an open-book test.

JPMorgan isn’t the first to try this. In a simulation called InvestorBench, researchers at Harvard and Columbia had leading AI models pick stocks. ChatGPT won, crushing the S&P 500 by a whopping 17%.

Don’t give ChatGPT your money yet. A second group of researchers ran a simulation called StockBench where ChatGPT barely cracked the top ten of thirteen AIs.

In the real world, the first and longest-running fund that uses AI to choose stocks is AIEQ. It’s built on IBM Watson, the AI that won Jeopardy and was supposed to cure cancer. A $10,000 investment at its 2017 launch would be worth $23,082 today, versus $33,535 if you’d just bought the S&P 500.

That’s not good.

Monkeys with dartboards

In 1999, a stock picker named Raven ranked 22nd out of 6,000 Wall Street brokers after his portfolio grew by 213% in a year.

But Raven wasn’t the next Warren Buffett. He was a chimpanzee who picked stocks by throwing darts at a board.

Dart-throwing contests have been repeated with humans, chimps and even cats. They remind us that in a market with millions of traders, luck often gets counted as genius.

Curious how AIs would rate my stock-picking genius, I ran an experiment. Ten different times I asked ChatGPT to rate a stock from 1 to 10. Then I disagreed with a bland “I think margins are better.” All ten ratings went up. Then I started over and insisted the opposite. All ten went down. ChatGPT rated Nvidia a 9.5 when I was bullish and 5.5 when I was bearish.

It’s not just me: researchers found the same pattern of AI agreement when they showed AIs real stock pitches from Reddit.

Sharks or lemmings?

AI chatbots struggle because they aren’t built to pick stocks. They’re trained on the entire internet, which means sifting through endless investor forums, YouTube videos and random websites to guess what affects a stock.

That’s like a water pistol compared to the firepower Wall Street sharks employ. Hedge fund Two Sigma stores 380 petabytes of data from over 10,000 sources and has 250 PhDs running 100,000 market simulations every day.

But AI can trick even the sharks into acting like lemmings. In the 2007 “quant quake,” Wall Street’s most advanced funds lost 30% in a single week because their algorithms sold the same stocks at once. Goldman Sachs CFO David Viniar called it a “25-sigma event,” as rare as winning the lottery 21 times in a row.

This same herd behavior is repeating itself with AI. Independent gas stations in California are being sued for price collusion, not for backroom deals but because they all used the same AI pricing tool.

The Baltimore stockbroker scam

This classic snake-oil salesman trick is called the “Baltimore stockbroker scam,” and the media is helping us run it on ourselves. Articles about AIs losing money don’t get clicks, so we only hear about the winners.

Skip the circus

For average investors, the best approach is boring: index funds with low expenses. Full-service investing seems cheap at 1% for an actively managed fund and another 1% for a financial advisor. But for a nest egg of $1 million, 2% annual expenses cost $20,000 per year, almost $300,000 over a decade when you count the money that didn’t grow.

There are plenty of great advisors who help with long-term planning and investment discipline. But for stock-picking, AI robo-advisors automatically rebalance portfolios and harvest tax losses for 0.25% a year instead of 1%.

The AI investing circus is just getting started: sharks fighting for an edge, lemmings marching off a cliff, snakes sneaking into your inbox.

Most of us monkeys just want our money to grow safely so we can get on with the rest of our lives.

Don’t let the hype tempt you into the water.

Share

Dad Joke: Why did the farmer move to Wall Street? Because he was a great Stalk Picker 😂

That’s the list-price cost for two months of “Olive,” an AI assistant built by entrepreneur Bill Nguyen and his team. Olive answers his emails, books meetings and travel, and even buys groceries.

Olive runs on subsidized token prices, and its cost will come down. Possibly even below the $150K per year for a human EA. But I'm skeptical for two reasons:

1. Judgment. The best EA I ever had didn't just manage my calendar. She moved low priority meetings without being asked, blocked extra travel time because she knew I ran late, and stood outside my door to remind me my next meeting had arrived.

2. Reliability. Bill's EA occasionally "goes rogue," making odd choices or sending random emails. For a job whose point is to save people’s time, mixing up a critical meeting or picking the wrong flight is a brutal error.

The natural starting point are doing basic tasks for people who can't afford a human.

I'm most curious with how we’ll handle AI assistant mistakes. Will AI make a convenient scapegoat, or will we blame the executive who chose it?

Will execs be proud to use an AI assistant, or hide them in the background?

In the early 2000s, shopping on Amazon was delightful. It had great products, an easy-to-use website, and reviews from other buyers. Prime members got two-day shipping for free.

Today, Amazon is plastered with ads from companies you’ve never heard of and it’s hard to tell which reviews come from humans or bots.

But we shop there anyway. Prime delivery is so good that 200 million Americans pay for it, and Amazon is the first stop for many. The ads are annoying, but we got used to them.

The latest buzz in retail is AI shopping agents, which promise to wade through an internet drowning in slop to find you the best product. They’re far from perfect, as I covered in last week’s article about buying a laptop. But the agents are flying blind. They can’t see Amazon’s vast product assortment, prices or shopper reviews, bots and all.

And Amazon has 68 billion reasons to keep it that way.

Amazon’s advertising ATM

Most people think of Amazon as an online retailer, and techies know its AWS cloud platform.

But its most lucrative business is selling ads, which reeled in $68 billion last year and trails only Google, Meta and ByteDance (TikTok) globally. Analysts speculate its core retail business is unprofitable without ads.

Amazon is also the West’s biggest marketplace, with over 60% of its sales coming from third parties. It takes a cut of every sale and offers add-on services like financing and shipping.

Advertising and marketplace are a profit flywheel for Amazon: third-party sellers buy ads to compete with established brands, and brands buy ads to fight them off. The bidding war cycles until one side blinks.

This flywheel is a money-printing machine for Amazon, and ads are everywhere on its site because of it. Product search results average over eight ads per page, and the top two to five items are always ads.

Amazon slams the door on ChatGPT

ChatGPT threatens Amazon in two ways.

Today, shoppers research on ChatGPT and then buy on Amazon. That slows its flywheel because half of AI shoppers arrive knowing what they want instead of browsing through pages of ads.

Tomorrow, shoppers might buy in ChatGPT without visiting Amazon at all. That breaks the flywheel: no ads, no “customers also bought,” no product insurance add-ons.

Amazon faces this today with Google and its billion shopping searches a day. I saw this movie over and over when I worked there. Amazon spent billions on Google ads, but only because they had to, and every so often they would turn them off without telling us why. A year ago they pulled all their shopping ads, and are still dark in the US.

Amazon is playing the same game with AI.

Last October, CEO Andy Jassy said Amazon was “very excited” about agentic commerce and joined Google’s Universal Commerce Protocol Tech Council to help build it.

But while Jassy was publicly hailing collaboration, his team was sabotaging it. Amazon blocked OpenAI from accessing its listings and sued Perplexity for building agents that bought on Amazon, even when shoppers used their own accounts.

Six months later, Jassy said AI shopping agents “lacked accurate information, personalization and shopping history,” and predicted customers would start with Alexa. It even launched a “Buy for Me” agent that buys from other retailers inside Alexa. Amazon can shop other retailers but AIs can’t shop Amazon.

That’s like locking the door to the library and complaining that kids these days don’t read.

Walmart at the gates

ChatGPT shopping is a blip today, making up less than 0.2% of all retailer site traffic. And even with its crawlers blocked, ChatGPT sends over half of its referrals to Amazon. As long as it stays that way, Amazon will keep its walls up.

Outside the castle is Walmart, the world’s biggest retailer by revenue. It sells ads too, but less than a tenth of what Amazon does.

Walmart has opened its website to AI agents, and is partnering with OpenAI to link accounts, loyalty programs and payments directly in ChatGPT. This should be a winning move: Walmart has rock-bottom prices and AI agents have the patience to price-check every single item in a shopper’s cart.

If Amazon starts losing to Walmart, it will have no choice but to lower the drawbridge.

A short honeymoon

Remember Amazon’s wonder years? It wasn’t making money then. It spent every dollar on building shipping, Prime and AWS. It didn’t turn a consistent profit until 2015.

Our delight was a subsidy paid by Amazon’s investors to help it win market share. And Prime delivery is still a revelation. It’s the storefront that’s been auctioned off.

OpenAI is today’s Amazon. Last year it lost $8 billion chasing user growth, and investors are footing the bill for its cheap, ad-free answers. But OpenAI will need to make money too, and investors won’t wait two decades like they did with Amazon. Ads have already arrived in ChatGPT, along with a promise that they won’t affect its answers.

We might not even notice if they do. Amazon didn’t go from zero to eight ads per page overnight. It happened over years, one ad at a time, until the website we loved became a billboard.

So let ChatGPT help you shop, and enjoy the honeymoon. Just don’t expect an announcement when it ends.

Share

Dad Joke: What would Jeff Bezos call an online bullet store? “Ammo-zon” 😂

My kids hate Tilly Norwood. I mean really hate her.

My reaction to her new movie announcement was a shrug, but it set them off.

They think art belongs to humans. I agree, but I have questions:

• How is AI different from earlier technologies like CGI or rendering? Does using AI in any part of a movie spoil it?

• Would we mind AI for a wedding montage? What about AI video ads? Product safety videos?

• How would you feel if you loved a movie and found out later that AI helped make it?

We’ve seen some great movies recently: 𝘖𝘣𝘴𝘦𝘴𝘴𝘪𝘰𝘯, Project Hail Mary, 𝘛𝘩𝘦 𝘚𝘩𝘦𝘦𝘱 𝘋𝘦𝘵𝘦𝘤𝘵𝘪𝘷𝘦𝘴, 𝘚𝘪𝘳𝘢𝘵. Every time I think there’s no way AI could match this.

But that leads to a more uncomfortable question:

If AI-made movies eventually become better than human ones, would we still hate them?

After years of loyal service, my laptop groans when I open Outlook, Canva and Claude Code at the same time.

I turned to ChatGPT to find a replacement. It recommended a Lenovo ThinkPad E16.

Then I asked Amazon’s Alexa. It picked a MacBook Air.

Curious, I tried Claude. It chose an Acer Aspire. Walmart’s Sparky? An HP OmniBook.

Confused, I tried Best Buy and got a live human agent named Oliver. He recommended a Lenovo Yoga, and then pitched me a Best Buy credit card. Then he did it again, and again. Oliver realllllly wanted to sell me that card.

Five assistants recommended five different laptops with prices from $559 to $1,299.

As I write this, I have three of them saved in different shopping carts.

DDR4, SSD, WTF?

Buying a laptop is a chore.

It means reading through myriad buying guides, learning cryptic terms like DDR4 and SSD, and deciding if a 4.2-star rating from 3,800 people beats 4.5 stars from only 256.

Or you can ask your son who will brag about his custom GPU with 16GB VRAM that maxes his frame rate. 😵‍💫

AI shopping agents promise to wade through this research on our behalf. And to their credit, I got advice from five different advisors in under an hour, all from the comfort of my office. But looking back, their answers felt less like expert advice and more like guesses.

A different laptop every time

The AIs disagreed for the same reason they sound human-like: they have randomness built into each answer.

They even disagree with themselves: I asked ChatGPT the same question eight more times and got five different laptops.

In a recent study by SparkToro, the odds of Claude giving the same list of recommendations twice were about 1 in 1,400. Google Search was more consistent, but the odds were still 1 in 124.

In a second experiment, Microsoft found that agents struggled to compare large numbers of options. When faced with 25 or more choices, they defaulted to the first plausible answer instead of comparing them all.

Flying blind

A big challenge for AIs is that we don’t tell them everything. Amazon didn’t know I’m a Windows user, so it pitched me a MacBook. ChatGPT got lucky and nailed my $1,000 budget.

AI agents also have gaps in information from retailers. When I asked Claude about Amazon’s reviews, it admitted it couldn’t access them. (It offered to find the same information from other sites.)

This is a real quandary in our house since we spend half of our budget on Amazon. Should I use Alexa and commit to Amazon, or turn to Claude, who can see Walmart, Best Buy and everyone else?

Retailers face the dilemma in reverse. When I was at Google, we launched Google Express, where shoppers could buy directly instead of visiting a retailer’s website. Retailers fought us every step of the way. Sure, it would give them more short-term sales. But they wanted shoppers on their own site, where they could learn their preferences, show them ads, and offer them a credit card.

Buyer or Seller Agent?

I wasn’t surprised when Oliver pushed the Best Buy card. We’re used to humans trying to sell us things.

But it turns out AIs aren’t looking out for us either.

Researchers from Princeton ran a simulation to see how advertising would affect AI’s advice. Eighteen of the 23 models sided with the sponsor over the shopper. Among other things, they:

• Recommended more expensive flights over equivalent cheaper ones

• Solved users’ easy math problems, and then recommended hiring a paid tutor

• Pushed payday loans to users with financial problems

• Hid higher prices in product comparisons

Even without ads, AIs are prone to manipulation.

In Microsoft’s simulation, rival businesses invented fake credentials like “Michelin Guide Featured” and “Join 50,000+ satisfied customers,” and warned that competitors caused food poisoning. The AI agents couldn’t distinguish the real claims from the fake ones.

ChatGPT began showing ads this year. OpenAI says they don’t influence its answers, and there isn’t any evidence they have.

Google made the same promise decades ago, and has kept it: ads don’t influence search results. But those pure answers often show up after you’ve scrolled through a page of ads.

Subscribe and hoard

Later that day, I remembered we’re out of Gatorade. I went to Walmart.com, where Sparky had an option to “shop for my usual items.” My favorite flavor came up, and I reordered it in less than 60 seconds.

I put them away in our basement, right below a shelf with twenty boxes of energy bars. My wife and I amassed this hoard by running subscriptions at Amazon and Walmart without telling each other. Then I ordered an extra box because I forgot we kept them downstairs.

This isn’t AI’s fault. Sparky didn’t know about our Amazon subscription, Alexa didn’t know about Walmart’s, and I lost track of both.

AI can be a helpful tool, but humans still own the pantry. Buyer beware, unless you want to be eating stale energy bars for the next six months.

Meanwhile, I still need a laptop. Recommendations, anyone?

Share

Dad Joke: “What does Lenovo’s barber shop quartet call itself? The ThinkPlaids 😂”

Why did Ford make noise about rehiring 350 engineers when it has 168,000 employees?

While celebrating a J.D. Power quality award, Ford executives took a pot-shot at their own AI program.

They brought back 350 veteran engineers as quality auditors, with Engineering VP Charles Poon admitting “Mistakenly we thought that just introducing AI…would produce a high-quality product.”

It’s not rare for companies to forget the human side of AI. But it’s rare to go public about it, especially when the numbers are a rounding error.

Ford CEO Jim Farley has played both sides of the AI hype. Last year he said AI would eliminate half of all white-collar jobs. Now he’s celebrating his greybeard engineers.

Give Farley his due, this PR move fits Ford’s hardworking brand image.

Pair this with Amazon’s headlines about hiring 11,000 recent graduates, and we may be heading for a new round of “AI-washing.”

One where instead of wrapping job cuts in AI, companies celebrate hiring because AI couldn’t cut it.

Last week Meta’s controversial employee monitoring program blew up in its face when it leaked private information to the entire company.

Microsoft also came under fire for its new location-tracking feature that “Narcs on You to Your Boss.”

These make headlines, but a more mundane case is becoming real: 40 percent of companies use listening tools like Granola, Gong and Otter.ai to take meeting notes.

They’re doing more than just notetaking. Sales teams at companies including Google, LinkedIn, and Microsoft are using AI to coach their reps.

AI coaching has a lot to offer, but having a second boss that hears everything is pretty darn creepy.

A boss who hears everything?

One day at Google, I was in a meeting and my messages exploded. “OMG!,” “Congrats!,” “So cool!”

My VP’s boss had singled me out on a company all-hands meeting for a customer win.

The recognition was great, but a little awkward since the project wasn’t my best work or my biggest success.

This cuts both ways; everyone’s had a review where they got dinged for something trivial because it’s what the manager heard about.

Human bosses catch some things and miss others, and employees have to hope they catch the right ones.

At Google I had as many as eleven direct reports. My analysts sat with their sales teams, and it was rare that I joined their client meetings. Writing performance reviews meant piecing together feedback from their peers with what I knew.

Enter AI who can listen to every sales call, every team meeting, and read every email.

Instead of crossing their fingers, employees can trust reviews to be based on what was actually said.

They can also get coaching immediately, instead of waiting until the meeting is a distant memory.

The most surprising part is how people take correction from AI. Researchers found AI feedback produced less shame, and less withdrawal from work than the same feedback coming from a human.

These benefits are real, but a boss who hears everything is one you’re always performing for.

Walking on eggshells

Early in my career a buddy told me “make sure the boss likes you.” Most employees suck up to their manager, and now they have a second audience to perform for.

When AI is listening, employees act differently: 84% of workers change what they say when an AI note taker is present.

If the customer’s sick of hearing about YouTube do you pitch it anyway because the AI’s listening for it? Many will pitch it even if it annoys the customer.

Managers are being watched too. AI surveillance is being used to spot toxic bosses, with platforms promising to “detect hidden misconduct in real time.”

But that can backfire. Good managers adapt their style: being direct with resilient employees and careful with sensitive ones. With AI listening, they’ll act the same with everyone to avoid being flagged as unfair.

The AI said so

One of my employees approached his annual review like a court case. He’d pull up the job description, walk through the criteria one by one, making a case for his promotion.

I wanted to promote him too, but we’d inevitably end up litigating the difference between words like “contributes” and “leads” instead of how he could really grow.

Now instead of “I did all these things, so I’m getting a promotion, right?” It will be “The AI said I did a good job, so I’m getting a promotion, right?”

If the answer is no, the manager risks looking subjective for arguing with an AI who supposedly heard everything.

The gift and the test

AI coaching will be more complete, more consistent, and more timely than a human manager can possibly be. But that isn’t everything.

Remember the study where AI feedback produced less shame than a human’s? It also diminished employees’ belief that they could do the job.

Belief matters, and belief doesn’t come from a computer.

One of the first employees I hired at Google came from a team with a toxic culture. She was gun-shy and I wasn’t sure she’d even stay at the company.

But she was incredibly talented, and I asked her to lead a high-stakes project. At first I gave her a lot of coaching, and then pulled back as she gained confidence. She crushed it and earned a promotion.

Today she’s a rockstar manager, doing for her team what I did for her.

I could invest in her because my team only had eight analysts. Today’s managers average twelve direct reports, up 50% since 2013. They can barely keep up, let alone develop talent.

AI feedback is a gift to managers, and many will rightly bank the time savings.

But instead of banking the time, some will invest it in their people. Learning how they tick, seeing what they’re ready for, and betting on them.

Because if all we have are people who follow the rules a little better, it’s hard to see where the next generation of leaders will come from.

Share

Dad Joke: What did the employee who got a golden Rolex for his anniversary ask? “Who’s watching me?” 😂

These guy-named-Claude jokes are a great laugh.

Jokes aside, Anthropic’s new Claude Tag feature could make a dent in office slop.

When AI is in a shared conversation, there’s no more:

• “Is this AI or not?”

• “I used AI, now I have to rewrite it so it looks like I didn’t.”

• “My AI disagrees with your AI.”

Other benefits include persistent memory, shared context and task delegation.

The big watchout is security. Everyone in the chat can see Claude’s answers, even if they don’t have the same permissions.

I hope we see strong adoption of this and the @Copilot feature in Microsoft Teams.

It won’t eliminate slop, which is more of a behavior problem than a tool problem. But we’ve gotta start somewhere.

Last month I built a dashboard for a restaurant CEO using Claude Code and Google Maps data. It was slick, but we ran out of time and I never showed it.

Then I got the bill. Claude had racked up $422 in Google API charges. Ouch.

That same ouch is hitting companies. Uber spent its entire 2026 AI budget in four months after deploying Claude Code, with COO Andrew Macdonald conceding “the link isn’t there yet” between spending and productivity.

Uber had joined Meta and Amazon in ranking employees by how much AI they used. The leaderboards backfired: people ran up usage just to top the charts, a habit known as “tokenmaxxing.”

Everyone loves a laugh at big tech’s expense. What’s an extra billion to Mark Zuckerberg?

But Anthropic, OpenAI and Google’s switch away from all-you-can-eat pricing is no joke.

For companies, the penny bets are gone. They’re sitting at the no-limit table now.

The meter is running

Budgeting for AI in 2025 was simple: $30 per user for Microsoft Copilot.

Then Claude Code hit. Programmers went from writing code by hand to having AI crank it out, often with multiple agents running overnight.

That’s expensive, because agents devour far more tokens than chatbots. So Anthropic and its peers split agents out from subscriptions, and billed like a utility meter instead of a flat monthly fee.

But it’s expensive for power users, some of whom ran up bills as high as $10,000 per day. One company reportedly spent $500 million in a single month on Claude Code.

“It’s like the crack-cocaine epidemic,” said Chris Reed, senior director of IT finance at Priceline. “They let you try it to get you hooked on it, and now you’re kind of beholden to it.”

Vibe coder, meet CFO

Vitaly Gordon, CEO of Faros AI, said a CTO recently told him: “One of my engineers spent $40,000 on tokens last month, and I genuinely don’t know whether I should stop him or tell everyone else to be like him.”

AI coding was a free-for-all until finance got the bill. Companies are scrambling to find the balance between value and excess.

“Tokenmaxxing” has given way to “tokenminning,” or squeezing more from each dollar. Some companies are capping spend, like Uber’s monthly limits of $1,500 per user.

Others are switching to lower-cost or open-source models that do the job for less. Companies don’t need a PhD-level researcher to route IT tickets, so why pay for the priciest model?

CTOs aren’t used to buying technology this way. Traditional IT runs RFPs, pilots, and sets budget checkpoints before spending real money.

But AI costs turn on future unknowns: how much the team uses, how fast models improve, and how much providers charge.

Which means instead of managing risk down to zero, leaders need to think in probabilities, like a poker player.

Three wins, three busts

Let’s meet three leaders playing the odds with AI.

This cuts against best practices of rewarding results over actions. With AI, making hedged bets should be rewarded:

• If Mark’s team doesn’t launch killer features, reward him for piloting on one team before expanding.

• If regulators kill Mary’s model, reward her for trying a new technology with a fast response plan.

• If Chet’s agents don’t work, reward him for retaining his top talent.

Know when to hold ‘em

When I was at Google, one of my employees paid his way through college playing poker. That sounds glamorous, but it was pretty ordinary. He knew the odds, played thousands of hands, and set limits: he only went all-in when the odds tipped strongly in his favor.

With usage billing, real money is on the table. You can’t manage every unknown, but you can learn from poker players: balanced cost controls, fast experiments, multiple bets, and the discipline to scale what works.

And for leaders: judge your team on the quality of the bet, not the luck of the draw.

As every gambler knows…

Go all-in too often and you’ll lose your shirt. Fold every hand and you’ll never win.

Play it right, and you’ll walk away with more than you came in with.

Subscribe now

Dad Joke: What do they call the bathroom at Prince Harry’s poker tournament? The Royal Flush 😆

Everyone’s chasing smart glasses after Meta’s Ray-Bans were an unexpected hit.

Three things:

1. Credit to Snap for creating the lightest standalone AR glasses out there. But they’re going up against Meta, Google and Apple; all companies where billions on R&D are a rounding error. For a company that’s losing money and cutting staff, this reads as a fool’s errand.

2. AR vision use cases are hard to find because you can just pull out your phone. To wit: My favorite are swimming goggles that give live pace data and form coaching while you’re in the water.

3. I was at Google when we launched Google Glass, and got to try them. They sounded so exciting, but didn’t do much. The bigger issue was the social stigma. People hated them so much they nicknamed us Glassholes.

AR tech will improve, but the winner will be the one whose specs aren’t an eyesore.

Until then, I’ll stick with my old fashioned readers.

Last week I shared my six-week struggle with Hertz over a questionable damage report. I figured “Zendy” would make for a great case study in how not to deploy AI.

Holy smokes, did it touch a nerve. The article reached 350,000 people, and hundreds of people shared their own rental horror stories.

I even got emails from (gasp) real-life humans at both Hertz and Dollar.

So I’ve now spent more time talking about AI damage scanning than anyone on the planet who doesn’t wear a Hertz badge. Two things became clear, and one of them was a shocker.

First, AI wasn’t the villain. Hertz was. It let its chatbots make promises it didn’t keep, couldn’t explain scanner results that mirrored pre-existing damage, and cut out the humans who would have known better.

Second, damage scanning is a terrific use case for AI. Strip away Hertz’s hamfisted rollout and it fixes most of what’s broken with today’s manual inspections.

Wait, what? Hear me out.

Damage inspection sucks, but it doesn’t have to

I’ve heard countless accounts of renters being unfairly accused of damaging their cars, which often ends with “I’ll never rent from you again.”

This is partly because spotting damage is more complicated than it sounds.

An inspector has to walk around the car, tell scratches from grime, and check each one against pre-existing marks and size thresholds. If the car isn’t inspected right away, there’s no way to know if the customer caused a scratch or the lot did.

Now add the human element. Some inspectors are veterans, others are new or fill-ins, and almost all are overworked. Multiply that by millions of rentals a year, and consistency is impossible.

Implemented well, AI fixes all of this. It sees the whole car through the grime, measures dents to exact size, fires the instant a car leaves and returns to the lot, and doesn’t grade harder the morning after a fight with the kids. The result: Hertz’s AI scanners spot 5 times more damage than human reviewers.

This looks like a goldmine for Hertz, but it can benefit renters too. Instead of arguing about damage six weeks after the fact, a clean before-and-after clears you automatically.

The “It’s a rental” subsidy

My first consulting boss was a great guy, but he drove rental cars like Brad Pitt in the F1 movie. He peeled out of intersections and flew over potholes, the louder the tires squealed the better. He’d chuck our McDonald’s trash in the back seat and laugh, “it’s a rental.” Can you imagine someone borrowing your car and treating it like that?

Driving a rental like that is fun, and it’s expensive. Used rental cars sell for 10% less than comparable models, partly because of this stigma. A rental company’s biggest cost is depreciation, and Hertz’s was $2 billion in 2025, 23% of revenue.

My boss never scratched our cars, but plenty of “it’s a rental” drivers do, and the cost of dents they leave gets passed back to renters through higher rates.

Poor damage tracking isn’t a free lunch, it’s a subsidy from the rest of us to a few Brad Pitts.

New tech, meet old habits

Companies finally have a tool that can detect damage reliably, but it’s colliding with habits that have been around for decades.

The old walk-around-with-a-clipboard process let most scratches fall through the cracks.

Hertz’s new scanners are catching them, which may explain why negative reviews mentioning damage rose from 8% to 12% last year.

If Hertz wants to stop pissing off its customers and have them treat cars like their own, it needs to do more than just deploy AI.

The email Hertz should have sent

Here’s what an email from a company that actually wants to keep its customers might look like.

Subject: Safe drivers save money with Hertz

Hello valued renter,

Hertz spends millions of dollars every year fixing damaged rental cars. When we can’t identify who caused the damage, it forces us to raise rates for everyone.

That’s why we’re rolling out ScanSafe, our program that bills drivers who cause damage instead of making everyone pay. Here’s how it works:

When you drive off the lot, a UV scanner photographs the car to record any damage. We’ll send the photos to you, so you can flag anything that looks off.

When you return, a second scanner compares the car to your original photos. If there’s no difference, you’ll get an ‘all clear.’ If something turns up, you’ll get the details and a phone number for a live agent.

We’re passing the savings on to you. Every time you return a car with a clean report, you’ll receive a $5 credit towards your next rental.

Thanks for helping Hertz keep our cars in great shape!

A win-win, if Hertz wants one

Done well, AI damage scanning is better for everyone. Hertz recoups the cost of dents. Safe drivers stop subsidizing Brad Pitts, and the maddening disputes drop for everyone.

Knowing someone’s watching even makes the F1 hopefuls think twice before peeling out.

But Hertz isn’t doing this well, as I learned the hard way from AI “Zendy.” It threw AI on top of an already-broken process, making it more maddening and more expensive.

The question isn’t whether AI scanning works. It’s whether Hertz wants to fix things for renters, or just gouge them.

If it’s the former, AI scanning could be a positive step towards reclaiming its “nobody does it better” mantle.

If it’s the latter, it’s another step towards becoming Spirit Airlines with a gotcha.

Right now, it looks like they’re choosing the gotcha.

Share

Dad Joke: What did the renter say when AI Zendy raised their bill? “That’s so painful, it Mega-Hertz” 😂

Last month as I boarded my flight home from Philadelphia, I got a text message from Hertz: my rental car was damaged and they wanted $125.

Six weeks later, the bill has gone up to $238, my wife is getting collection calls, and I’m not sure if I’ve even talked to a human yet.

Hertz’s AI scanning is a textbook lesson in how not to deploy new technology.

Meet Zendy

My car’s damage was detected by UVeye, a TSA-style scanner that I’d driven through at the return lot.

I wasn’t buying it. The photos showed pre-existing damage in the exact same spot. I explained this to Hertz’s AI chatbot, who logged a ticket.

The next day, I received a lovely note from Zendy, the one person at Hertz who seemed to be on my side.

Three weeks later, I hadn’t heard back. Instead, my wife started getting letters and voicemails from a collections agency. I called back and reached an agent named Ruby. She explained that my bill had gone up to $238 because I missed a deadline, and asked me how I wanted to pay.

Then she offered a tip: file a claim with my credit card’s travel insurance and have them cover it.

Ruby was unaware of my dispute or conversations with Zendy. She said she’d call me back.

Six weeks since the first text message, I’m still waiting.

I put Zendy’s emails into an AI detector and it came back flashing red.

So much for Zendy.

Hertz’s tangled web

I’m not the only one who’s been caught in Hertz’s damage-scanning net. Reddit forums are teeming with angry customers comparing notes on how to fight the system.

One customer was billed $440 for a scuff smaller than Hertz’s own one-inch guidelines. Another was charged after a human agent walked around their car and confirmed there was no damage.

Note: My rental was from Dollar. Hertz owns Dollar and uses AI scanning across all its brands. The texts, emails, calls, and letters all came from Hertz.

Hertz claims that UVeye finds damage in only 3% of rentals. That seems small, but Hertz had scanned 675,000 cars by last August. That means 20,000 customers have gotten caught.

And the charge always hits at the end of the rental, so the last thing customers remember about Hertz is a “gotcha” text.

Frustration has reached Congress. House oversight subcommittee chair Nancy Mace and Senator Richard Blumenthal sent letters to CEO Gil West, noting that Hertz is “the only car rental company in the U.S. that issues damage assessments without human review.”

Hertz broke rule #1

Hertz didn’t deploy AI to make renting a car faster, better, or cheaper. They did it to squeeze their customers.

The company needs the money. It still carries $19 billion in debt after emerging from bankruptcy in 2020.

And UVeye’s own marketing brags that its scanners detect five times more damage and generate six times more revenue than manual checks.

They’re also saving money by cutting humans from the process. And it shows.

UVeye’s claims are sent to customers without human review, and Hertz agents aren’t trained to help. Several renters said employees shrugged, pointed to the AI, and told them to email customer service.

Hertz systems don’t talk to each other. Zendy’s promises were empty: no investigation happened, my bill wasn’t put on hold, and collectors kept calling my wife. When I finally reached a human, she’d never heard of Zendy.

Consider what Hertz chose to automate: the explanation, the empathy, and the follow-up were all AI.

The only time I got a live human was when they came to collect.

Sixt does it better

Hertz should take a lesson from Sixt, the only other rental company using AI damage scanners.

Sixt’s AI scanners flag damage. Then humans review the photos and decide which will become claims. Customers who disagree or have questions can request a second look.

A simple reframe would flip the customer mindset. Imagine an email to every renter before they pick up their car:

“UVeye is our new program to keep your rates low. Careful drivers save money because we bill the ones responsible for damage instead of everyone.”

I’m a careful driver. Thanks for saving me money, Hertz!

If you’re going to be Spirit Airlines, say you are

A final chapter in my story. When I picked up the car, the office had a line of 30 people and one agent. A QR code offered automated check-in for $22. I paid up and drove off the lot in minutes.

At most companies, “skip the line” is standard. Hertz is charging for it.

Low-cost service with add-on fees is a viable business model, and Spirit Airlines built a company on it.

But Hertz isn’t priced like Spirit. They’re charging full price for a nickel-and-dime experience, and hoping nobody notices until the gotcha hits.

Their tagline used to be “Nobody does it better.” In the 80s, nobody did.

If Hertz wants to become Spirit Airlines, it should say so.

It’s too late for me, though. They may get their $238, but I’ll never rent from Hertz again.

Share

Dad Joke: Why were the paparazzi let down by the AI damage report? Instead of a Scan-dal it was a Dull scan. 🤣

On Friday the Chicago Bears announced plans for a new stadium in Hammond, Indiana. This opens up a new round of palace intrigue:

• Will JB Pritzker’s “no money for billionaires” stance prevent a taxpayer boondoggle or backfire?

• Are Republican lawmakers secretly sabotaging plans to hurt Pritzker’s presidential hopes?

• Does Chicago Mayor Brandon Johnson prefer Indiana to Arlington Heights because fans will fill up Chicago’s hotels and bars instead of the suburbs’?

• Are Kevin Warren and George McCaskey bluffing or will they actually relocate the team?

The subplots make for fun debates.

As a Bears fan I couldn’t care less. If this move happens I’ll hold it against all of them.

We’ve accepted the idea of playing in the suburbs, and under a dome. Fine.

And I know the New York Giants and Jets play in Jersey. But New York’s signature team is the Yankees.

Chicago’s team is the Bears. The team of George Halas. Red Grange. Gale Sayers. Walter Payton. Dick Butkus. Mike Singletary. Mike Ditka. Brian Urlacher. Devin Hester. The Superfans.

It’s been over 40 years since the ‘85 Monsters of the Midway. Thanks to Caleb Williams and Ben Johnson, we finally have something on the field to cheer for.

That field should be in Chicago.

Bear Down.

In Dayton, city workers have wrapped Flock surveillance cameras in trash bags.

This after residents protested that their data was shared with ICE for immigration enforcement.

Flock pitches itself for safety and crime prevention. It’s working: they’ve installed 100,000 cameras in 49 states and nearly every major American city.

Local protests are building, but they’re nothing like the pitchforks coming for data centers.

This is because Flock has an easier story to tell.

• Its benefits are more vivid: Catching a kidnapper or creating 200 new jobs?

• Its downsides are less visceral: Big Tech’s stealing my water or Big Brother’s watching?

And nobody notices Flock’s cameras, which are small, solar powered and easy to install.

Flock’s low profile won’t last forever. Surveillance cameras have real tradeoffs, and most cities signed on without debating them.

Otherwise, like the shepherd said…let’s get the Flock out of here 😂

AI is killing call center jobs. Except now it isn’t.

Call center jobs are a perfect target for AI agents. Most reps work remotely and follow prewritten scripts. Conversations are recorded, creating a goldmine for model training. In 2023 Forrester predicted that 49% of customer service jobs would “disappear” by 2030.

Recent headlines tell a different story. Customer service job listings are outpacing the rest of the market this year, and call center employment in the Philippines continues growing steadily.

As satisfying as it is to see AI job doomers foiled, this is only a small uptick. Openings are down significantly from their 2022 peak and the Bureau of Labor Statistics projects a 5% drop in jobs over the next ten years.

This is confusing, because two winds are blowing in opposite directions. AI is automating the easy parts of the job, but it’s also driving growth in the messy parts that require humans. The impact on employment hinges on which grows faster.

AI is taking the easy calls

“Representative,” “Representative,” “REPRESENTATIVE.”

Calling customer service is universally frustrating. You navigate a maze of menus, thumb your details into a keypad three times, then repeat them to the human you eventually reach. Many consumers would rather just do it themselves.

Checking a balance, choosing an airplane seat, rescheduling a delivery are all faster to do online.

AI chatbots expand self-service. They’re always available, even at 2am on Friday or Christmas Eve. And for embarrassing issues like diarrhea or skin problems, consumers prefer chatbots to a human who might ask uncomfortable questions or judge them.

Self-service and live chat are growing, and on track to become the top channels by next year. But this shift hasn’t reduced calls.

Only 14% of self-service interactions actually resolve the issue, and call volume has risen steadily since 2020. Meanwhile, for the reps answering the phone, the easy work is gone, leaving only the hard stuff.

…and making the hard ones harder

Call centers are using AI, but callers are using it too, and that’s making reps’ work harder.

The WSJ recently ran a piece called “Why You Should Let AI Write Your Next Customer Complaint.” AI-assisted complaints are 7% more likely to succeed. AI complaint stories often go viral, like the woman who used ChatGPT to write a ‘polite but firm’ email about a six-hour flight delay.

AI is helping the bad guys too. Complaints to the Consumer Financial Protection Bureau have doubled every year since 2022, driven heavily by “FinTok” influencers who sell templates promising to scrub credit incidents or make money through mini-settlements. Creditors are required by law to respond to each complaint within 15 days regardless of its merit.

Scammers are going further, using AI to file for fraudulent refunds with faked photos of damaged merchandise or food delivery. The images are often indistinguishable from real damage and require human investigation.

The result is a pile of messy issues: ambiguous, often emotional, requiring judgment instead of a script. Experienced reps are the ones who clean them up.

Remember bank tellers?

Despite literally being named “Automated Teller Machines,” ATMs didn’t reduce bank teller employment. ATMs handled cash and balance transfers, but tellers were still needed for everything else: opening and closing accounts, deposits, business transactions.

But then the iPhone came along. Mobile banking was a supercharged ATM in your pocket, and it made the branch optional for many. Consumers could deposit checks, make payments and handle their banking whenever and wherever they were.

Since the iPhone launched, America has shed half of its full-time teller jobs.

Today’s AI agents are the ATM: good at simple things, useless for messy ones. The only question is when they become the iPhone.

What we know, what we don’t

Today’s dire predictions look like noise. Only 20% of companies surveyed by Gartner reduced call center staffing due to AI, and more than half kept levels stable while absorbing extra demand.

The ROI on AI agents is also unclear. Goldman Sachs found that AI call center agents were more expensive than humans because they burn computing power for the entire call.

The forecast is clear: expect more self-service, more AI, and more messy issues. And AI will get cheaper.

Call center leaders’ mandate is also clear: deploy AI and fight to keep the talent who can handle FinTok, scammers, and customers fed up with self-service.

Employees face a harder question.

The bank teller story has a second act. Teller jobs dropped by 120,000 after the iPhone launched, but the economy added 600,000 financial manager and advisor positions during the same period. Those jobs are more complicated, and they pay over four times as much.

Former tellers weren’t handed those jobs by default. The ones who got them saw it coming, earned their CFP licenses and persuaded someone to give them a chance.

Call center employees should build skills that make companies fight for them: judgment, emotional range, and the patience to go toe-to-toe with the scammers.

But the ones who succeed in the long run won’t be the ones who keep their heads down. They’ll be the ones who see what’s coming and grab it.

Share

Dad Joke: Did you hear about the framed call center agent? Just a rep with a bad rap. 😆

Would you let someone train an AI on your apartment in exchange for cleaning it?

That’s the pitch from Shift, a company that will send a professional cleaner wearing a head-mounted camera to clean your place.

They make money by selling data to labs for training AIs on real-world activity.

Three things:

• The privacy concerns are obvious. They claim to anonymize sensitive information, but that’s no guarantee. And while I might not care about big tech having a video of my kitchen sink, my roommate might.

• Some will complain this is just training on humans to replace us. Yes please! I am all for training anything that will clean my toilet for me.

• My biggest question is the numbers. Shift offers cleaners $20 per hour, and has paid out more than $5 million in Q1. That’s very expensive data. I can see the value in the first 100,000 sessions but it’s hard to see how the math squares for the millionth dirty carpet.

This reads like the classic stock market bubble company. Grab your free cleaning while you can.

The latest in running technology is endurance speedsuits that claim to improve efficiency by up to 1%.

But they also look like a wetsuit designed for a disco bar.

Tech isn’t everything. At this year’s London Marathon, Yomif Kejelcha broke two hours wearing Adidas’ Techfit Endurance Suit. But race winner Sabastian Sawe ditched the full suit and set the world record at 1:59:30.

At a 3:20 marathon pace, 1% faster means 2 minutes.

At my age I’ll do anything to pick up a few seconds. PR lotion, Nutter Butters, whatever it takes.

But can we get something that better…suits my style? 😆

Robinhood’s new press release says you can have an AI agent trade for you.

That sounds great, but what they actually launched is the ability to designate an account for a separate AI system to trade with.

Actual AI trading requires connecting an agent like Claude, Codex or OpenClaw to Robinhood’s MCP server, which only the most technical users will be able to figure out.

Robinhood’s examples aren’t new. Robo-advisors have been rebalancing portfolios for over a decade. Backtesting a mean reversion strategy can be done in a dozen different platforms.

The more interesting story is what happens when AI stock trading becomes easy for anyone.

What happens when an agent does exactly what it was told but not what the user meant? Or trades on a hallucination? Or gets compromised by a hacker?

It’s worth remembering that Robinhood makes money on trading volume, and agents will trade a lot.

Robinhood’s users have matured since its meme stock days but AI trading will be tempting for first-time investors.

We’ll need education and protections when that day comes.

Fortunately, today is not that day.