Subscribe
Share
Share
Embed
A daily briefing on the AI systems, products, companies, and policy shifts that are just becoming possible.
Want a podcast for your own topics? Join early access: https://www.barelypossible.to/waitlist/?source_path=public_feed&feed_source=rss
Okay kiddos, it's your boy Tony DeLuca, and welcome back to Barely Possible. We've got a lighter menu today, but don't let that fool you, because there's a couple of items on here that'll tell you more about where the money's really going in this AI moment than any of the big splashy headlines from the last two weeks. So grab your coffee, sit down, and let's have at it.
Here's what we're chewing on. We've got a president's memecoin that vaporized three-point-eight billion dollars out of nearly a million wallets. We've got Amazon quietly turning out the lights on Mechanical Turk, which, if you know your AI history, is a much bigger deal than it sounds. We've got almost ninety new unicorns minted in half a year. We've got a little company called Bending Spoons that most of you have never heard of, but that owns AOL and Vimeo and just went public. We've got Mistral shipping a math-proving model. And then, because I like to keep you honest, a little science palate cleanser about how the Earth got cooked in its first half a billion years. Let's start where the pain is.
Let me tell you about the TRUMP memecoin, because there's a fresh analysis out that puts a real number on the damage. This is Anthony Ha's reporting, and the number is this: nearly one million people have lost a combined three-point-eight billion dollars buying President Donald Trump's TRUMP memecoin, while Trump himself made six hundred thirty-six million dollars. Read that back slowly. A million people out three-point-eight billion. One man up six hundred thirty-six million. That is not an accident of the market. That is the shape of the thing working exactly as designed.
Now, I want to be careful and fair here about how these things actually work, because the mechanics matter. A memecoin like this launches, a big chunk of the supply is held or controlled by the people close to the launch, and the money that flows in from the crowd is what creates the price. When the crowd buys in at the top, they're literally the exit liquidity for whoever got in first and got in cheap. The insiders collect the fees, they collect the early position, and when the enthusiasm fades, the little guy is left holding a token worth a fraction of what he paid. This is the oldest pattern in the world dressed up in new clothes. It's a raffle where the house also holds most of the tickets.
And look, I'm a Bronx guy. I've watched people lose money on the ponies, on scratch-offs, on cousin Vinny's can't-miss investment. There's a difference, though, between a bad bet you walk into with your eyes open and a bad bet where the guy running the table is the sitting President of the United States, and the branding is his name and his face. That's what makes this one land different. It's not just financial, it's the sheer conflict-of-interest of it. Somebody in the highest office in the country ran a token, cleared six hundred thirty-six million, and a million regular people ate the loss.
Why does this matter to you, the builder, the founder? Because crypto and web3 keep trying to rehabilitate their reputation, keep trying to say, no, no, the memecoin degeneracy is a sideshow, the real work is stablecoins and settlement rails and tokenized assets. And there's truth in that. But every time a number like three-point-eight billion in losses hits the wire, it makes the whole category harder to build in. It makes regulators itchier, it makes serious institutional money more cautious, it makes your compliance conversations longer. If you're building anything legitimate in this space, the memecoin casino is not your friend. It's the loud drunk at the end of the bar making the whole place look bad to the health inspector. Keep that in mind when you're pitching, because your investors are reading these same headlines.
Now let's shift from money that vanished to something that's disappearing more quietly. Amazon is shutting down Mechanical Turk. Or, more precisely, according to Anthony Ha's reporting, Amazon will stop accepting new customers for Mechanical Turk, and these may be the last days of the platform. And I want to sit with this one for a minute, because if you don't know the history, you'll miss why it's poetic.
Mechanical Turk launched back in 2005. The name itself is a reference to an eighteenth-century chess-playing machine that was supposedly automated but actually had a real human hidden inside making the moves. Amazon named their platform that on purpose, with a wink, because that's exactly what it was. It was artificial artificial intelligence. You'd post a task that computers couldn't do well, labeling images, transcribing audio, checking whether a photo contained a stop sign, moderating content, and a distributed crowd of real human workers all over the world would do those tasks for pennies apiece. It was the plumbing underneath a huge amount of what we called machine learning for the better part of two decades.
Here's the thing that should make you stop and think. Every big image recognition dataset, every content moderation system, a huge amount of the labeled training data that made the last generation of AI possible, a lot of that got built on the backs of Mechanical Turk workers. Human beings, doing the tedious labeling work, so that models could learn. The machine had a person inside it the whole time. And now, the models those workers helped train have gotten good enough that Amazon doesn't see a future in the platform anymore. The tool that AI was built on top of is being retired because AI got good enough to replace the tool.
That is a genuinely full-circle moment, and I don't want to be glib about it, because there's a labor story underneath the poetry. There were real people who made real, if meager, money on Turk. Researchers who ran studies on it. Small operations who used it to get data labeled cheaply. This wasn't a big glamorous product, so it's going out with a whisper, not a press conference. But for anyone building in AI, Mechanical Turk closing its doors to new customers is a marker. It's the sunsetting of the crowd-labor era of machine learning, the era where the intelligence was rented human attention, one HIT at a time.
And here's the practical read for founders. If your product still leans on a human-in-the-loop crowdsourcing layer for labeling, moderation, or data cleanup, you should be asking hard questions about your dependencies and your alternatives right now. The vendors in that space are consolidating and, in Amazon's case, exiting. The economics that made crowd labor cheap are getting undercut by models that can do first-pass labeling themselves. That doesn't mean you fire all the humans, quality still needs review, but the shape of the workflow is changing under you. Don't get caught flat-footed when the platform you depend on quietly stops taking new customers.
That thread, machines replacing the human scaffolding underneath them, connects to something the AI Daily Brief was chewing on this week, and I want to touch it briefly because it's the productive version of the same idea. Cloud Code's creator Boris Cherny put out a framework describing five archetypes he sees on his team: the prototyper, the builder, the sweeper, the grower, and the maintainer. The prototyper churns out ideas, most of which don't ship. The builder turns a prototype into production. The sweeper simplifies and optimizes and unships. The grower iterates toward product-market fit. The maintainer keeps a mature system reliable at scale. And Cherny's point, which I think is the useful one, is that these aren't tied to job titles anymore. A designer might be a prototyper, an engineer might be a maintainer, and the mix your team needs depends on where your product is in its life.
I'll give you my plain read on it. The reason this matters, and the reason it connects to the Turk story, is that when the tedious middle of a job gets absorbed by agents, what's left is the disposition, the judgment, the taste. You stop being defined by the function you were hired into and start being defined by which of these modes you're actually good at. Nathaniel Whittemore added a few externally-facing roles to Cherny's list, an editor who decides which prototypes deserve to live, a scout who brings the outside signal in, a risk steward who keeps the whole fast-moving machine from flying off the rails. I don't need you to memorize the taxonomy. The one thing worth taking away: as making gets cheap, the scarce skill isn't producing the thing, it's deciding what's worth producing and catching the risk before it bites you. That's a hiring lens and a self-assessment lens both. That's all I'll say on it, because I'm not here to recap another podcast segment by segment.
Now let's talk about money that's showing up rather than disappearing. TechCrunch has a rundown from Dominic-Madori Davis, and the headline is that almost ninety new unicorns have been minted so far this year, with AI igniting what they're calling an investor frenzy, more startups hitting billion-dollar valuations every single month.
Now, I want you to hold two things in your head at once here, because I'm the skeptical guy and I'm not gonna let a big shiny number go by without kicking the tires. Ninety new unicorns in roughly half a year is a torrid pace. On the one hand, that reflects something real. There is a genuine wave of AI companies solving genuine problems, and there's a mountain of capital chasing them. On the other hand, and you knew there was an other hand coming, a unicorn is just a valuation. It's a number a group of investors agreed to on paper in a private round. It is not revenue, it is not profit, it is not durability. A billion-dollar paper valuation and a functioning business are two very different animals, and in a frenzy, the gap between them gets wide.
We've been circling the AI-hype theme on this show for a few days now, the Jersey Mike's IPO name-dropping AI, the memory-chip shortage, all of it. So I don't want to just wave my hands and say hype, hype, hype. Let me give you the specific, useful angle instead. When you mint ninety unicorns in six months, you are pulling forward a lot of valuation on the promise of future performance. Some of these will earn it. Many won't. And the ones that don't are the ones that raised at a billion, then can't grow into that number, and then have to do a down round or get quietly acqui-hired for parts. So if you're a founder raising right now, understand the environment you're raising in. Capital is loose, valuations are frothy, and that's great for the term sheet and dangerous for the next round. Raising at a nosebleed valuation feels like winning until you have to defend it eighteen months later against real revenue. Take the money, but be honest with yourself about the multiple you just signed up to justify.
And here's the thing that ties the unicorn frenzy to the memecoin story, and I promise this connection is real and not forced. Both are stories about valuation running ahead of substance. In the memecoin, the substance was basically zero and the crowd found out fast and brutally. In the unicorn wave, the substance is real but the valuations may still be ahead of it, and the reckoning, if it comes, comes slower and quieter, in down rounds and shutdowns eighteen months out instead of a token chart falling off a cliff in an afternoon. Same disease, different metabolism. As a builder, your job is to be on the substance side of that gap, not the froth side.
Speaking of substance without the froth, let me tell you about a company that most of you have genuinely never heard of, and that's the whole point. Anna Heim at TechCrunch has a piece walking through Bending Spoons, which she describes as the little-known AOL and Vimeo owner that's now public. And this is one of my favorite kinds of stories, because it's the quiet operator hiding in plain sight.
Bending Spoons remains largely unknown, even though its portfolio of products has served more than a billion people. Let that sink in. A billion people. A company you couldn't pick out of a lineup. This is an Italian company that built its reputation on mobile apps and then pivoted into a very specific and, frankly, very clever strategy: buying up established, sometimes tired, sometimes neglected tech brands and running them for cash. They picked up Evernote, they picked up the meditation app, they picked up Vimeo, and yes, they picked up AOL, the ghost of the nineteen-nineties internet that somehow still has a real user base and real revenue.
Why does a company like this matter to you? Because it's the anti-unicorn. While ninety startups are getting minted on the promise of the future, Bending Spoons built a real business on the deeply unglamorous work of buying the past and squeezing value out of it. There's an art to it. You take a brand people already know and trust, you cut the bloat, you tighten the monetization, and you run it like an operator instead of a dreamer. It's not sexy. Nobody's writing breathless threads about the AOL turnaround. But they just went public on the back of it, serving a billion people across the portfolio.
And there's a lesson buried in there for the current moment. Everybody wants to build the shiny new AI thing. Fewer people want to do the boring, disciplined work of running products that already have distribution and users and revenue. But distribution is the hardest thing to build and the easiest thing to buy if you're patient and you've got the operational chops. Bending Spoons figured out that owning the boring, working thing can be a better business than chasing the exciting, unproven thing. In a frenzy, that's contrarian. And going public in this environment, off a portfolio of legacy brands rather than an AI moonshot, is a quiet little statement about what a durable business actually looks like. File it away.
Now let's shift from the operators to the researchers, because Mistral put out a recent report a few days back that's worth a quick mention, especially for the builders who care about where AI reasoning is actually getting reliable. It's called Leanstral 1.5, and the tagline is Proof Abundance for All.
Here's the short version, and I'm going to keep it short because I don't do kernel-level deep dives on this show. Leanstral is a model aimed at formal mathematical proof, specifically working with Lean, which is a proof assistant, a piece of software mathematicians and computer scientists use to write proofs that a machine can check for correctness, step by step, with zero hand-waving. The dream here is that instead of an AI confidently telling you something that sounds right and might be garbage, you get an AI that produces a proof the machine can verify is actually, provably correct.
And that, to me, is the interesting angle for anyone building serious systems. The whole knock on large language models has been the hallucination problem, the confident nonsense. Formal verification is one of the few genuine answers to that, because in a formal proof system, you can't bluff. The checker either accepts your proof or it doesn't. So a model that gets good at generating formally verifiable proofs is a model whose output you can actually trust in the domains where it works. That's a different flavor of AI than the chatbot that sounds smart. It's AI you can audit.
Now, the name Proof Abundance for All is doing some marketing work, and I'm skeptical of the word abundance the way I'm skeptical of every word ending in -ify. But the direction is real and it matters. If formal proof generation gets cheap and reliable, the near-term payoff isn't your average app. It's in verified software, in security-critical code, in the parts of the world where being probably right isn't good enough and you need to be provably right. If you're building anything where correctness is life-or-death or money-or-jail, keep an eye on this line of work. That's the whole segment on it, we've got the link in the show notes for the ones who want to go deep.
And honestly, that ties back to a thing I keep coming back to on this show. The most valuable AI isn't always the flashiest. Bending Spoons found value in boring old brands. Leanstral finds value in the unglamorous, rigorous work of proof. The froth is in the memecoins and the ninety-unicorns-in-six-months headline. The substance is in the operators and the verifiers. As a builder, you want to know which side of that line you're standing on.
Alright, let me clean up a couple of loose items on the menu before I let you go.
There was a TechCrunch item about Uber's European expansion hitting a speed bump, and I want to be straight with you about it, because context matters. This is Anthony Ha again, but the underlying story goes back to February, when Uber announced plans to launch in seven new European markets in 2026. The fresh wrinkle is that five of those seven launches are now reportedly on hold. So this isn't brand-new news out of nowhere, it's a follow-up on a plan from earlier this year that's stalling. And the plain read is: even a giant like Uber runs into the wall of European regulation, local labor rules, and market-by-market friction. Europe is not a single market you can steamroll, it's twenty-seven kitchens with twenty-seven sets of rules. If you're a founder dreaming about European expansion because the TAM slide looks juicy, remember that Uber, with all its money and lawyers, just quietly shelved most of a seven-market rollout. Expansion into Europe is death by a thousand regulatory paper cuts. Plan accordingly.
And I'll give the desk-gadget listicle a pass, that one's a shopping guide, not a story, and I respect your time too much to read you a list of stuff to buy.
Now, before I sign off, let me give you the palate cleanser I promised, because I think you've earned a little wonder after all this talk of money vanishing and platforms dying. Ars Technica, Jacek Krywko, has a piece about Earth's earliest history, and the headline is The missing 500 million: Cosmic bombardment melted Earth's first crust.
Here's the gist. There's a period in Earth's history called the Hadean, named after Hades, because it was hell. The very first crust of our planet, the first several hundred million years. And for a long time the assumption was that the intense heat of that era came mostly from the interior, from the planet's own molten guts. But this new work suggests a big chunk of that heat came from the outside, from a relentless cosmic bombardment. Impact after impact after impact, so many and so large that they literally melted and re-melted Earth's first crust. The heat of the Hadean, in other words, may have come from getting hit as much as from within.
And I love this because it's a reminder that the world we walk around on, the solid ground we take for granted, was forged in a period of absolute violence from the sky. The crust under your feet is the scar tissue of a half a billion years of getting pummeled. There's no builder lesson here, no product angle, I'm not going to insult you by inventing one. Sometimes it's just good to remember the planet itself is a survivor of a rougher startup phase than any of us will ever go through. Half a billion years of getting cooked and it turned into all this. Puts a bad Series A down round in perspective, doesn't it.
So let me tie it all together. Today was a story about substance versus froth. The TRUMP memecoin, three-point-eight billion in losses against one man's six hundred thirty-six million, that's froth at its most predatory. The ninety unicorns, valuation running ahead of proof. And on the other side, Bending Spoons quietly running a billion-user portfolio of boring brands, Leanstral chasing provably correct output, and Mechanical Turk closing the book on the human-labeled era that all of modern AI was quietly built on top of. The machine finally learned to do what the person inside it used to do. Where you want to be standing, as a founder, is on the substance side of every one of those lines.
That's the menu, kiddos. If today taught us anything, it's that the loudest number in the room, whether it's three-point-eight billion vaporized or ninety unicorns minted, usually tells you less than the quiet operator nobody's writing about. Watch the quiet ones. This has been Barely Possible. I'm Tony DeLuca, thanks for spending a little time with me, and I'll catch you on the next one. Be good to each other out there.