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, I'm your boy Tony DeLuca, and welcome back to Barely Possible. Pour yourself whatever gets you through a Monday, because today we've got a menu that runs from a seven-billion-dollar plumbing deal to two AI CEOs arguing about whether people hate their industry — and if so, whose fault that is. Buckle up, let's have at it.
Let me start with the story that I think matters most to you if you're building anything, because it's the kind of deal that changes the shape of the ground under your feet without a lot of noise. Bloomberg reported that Stripe has finalized a deal to acquire OpenRouter for more than seven billion dollars. Now, if you don't live inside the developer tooling world, OpenRouter might not ring a bell, so let me lay it out plainly. OpenRouter is the thing a lot of builders quietly wire into their apps so they don't have to marry a single AI model. You send your request, and OpenRouter figures out which model to route it to — cheapest one, fastest one, best one for the task, whatever you tell it to optimize for. Their CEO, Alex Atallah, described the company as "Stripe for AI." A single access point, no lock-in to any one lab. They claimed something like eight million users and access to more than four hundred models.
Now here's the part that made me sit up. Back in May, OpenRouter raised a Series B of a hundred and thirteen million dollars, at a reported valuation of about one-point-three billion. That's May of this year. And now, a few months later, they're getting bought for north of seven billion. That's roughly a five-x jump on paper in about a quarter. I want to be careful and frame this correctly — that funding round is an older item that resurfaced; the news today is the acquisition report. But you put those two numbers next to each other and you see how hot this particular corner of the market runs right now.
So why does Stripe want this? Think about what Stripe actually is. Stripe is the company that made it so you, the builder, never have to think about the ugly machinery of moving money. You drop in a few lines of code, and payments just work across a hundred payment methods and a dozen currencies. That's the whole game — Stripe sits in the middle, takes a little cut, and abstracts away a mess you never wanted to deal with. Now look at what OpenRouter does. It sits in the middle of your AI calls, abstracts away the mess of four hundred different models and their pricing and their quirks, and — this is the interesting part — it's a natural place to meter and bill for usage. That's a tollbooth. Stripe knows tollbooths. Stripe is the best tollbooth operator on the internet.
And here's the founder angle I want you to chew on. There's a version of the AI economy where the money isn't just in the models — it's in the layer that decides which model gets the job and who pays for it. If OpenRouter becomes the default meter, and Stripe owns the meter, then Stripe just bought a seat at the table for every AI transaction that flows through that layer, no matter which lab wins the model war. It's the "sell picks and shovels" play, except the pick is a routing decision and the shovel is a billing rail. For you as a builder, the practical thing to watch is whether that layer stays neutral. The entire pitch of OpenRouter was "prevents lock-in." The entire business model of a payments giant is, historically, to become the thing you can't leave. Those two impulses are going to have a conversation inside that company, and how it resolves affects your dependency map. Stripe told TechCrunch it doesn't comment on rumors or speculation, so this is on Bloomberg's reporting for now, but if it closes as described, it's the most consequential builder story on the board today.
Now, that deal is fundamentally about trust — about being the neutral middle layer everyone's willing to route through. And trust, funny enough, is exactly the word at the center of the next story. Let me shift you from the plumbing to the podium.
Dario Amodei, the CEO of Anthropic, spent this past weekend arguing on the internet, which is a thing that happens now to the people running the most valuable companies on Earth. The setup: the investor Gavin Baker went at Amodei, both on the All-In podcast and on X, essentially saying Amodei's constant warnings about AI danger have fueled the backlash against the industry — the anti-data-center sentiment, the government crackdowns, all of it. Baker's line was that Amodei "lost the argument" on AI regulation, and that given he's about to run one of the most important companies in the world, quote, "I respectfully think he should make an effort to be a more positive advocate for his own industry."
Amodei didn't take that lying down. He pushed back on the idea that his messaging has been lopsidedly negative — he pointed out he wrote one big essay on the risks and one, "Machines of Loving Grace," specifically because he didn't think the industry was painting an inspiring enough picture. But here's the part I actually found worth your time. Amodei conceded the premise underneath the fight. He said, and I'm quoting him, "the public has a negative view of AI" and "this is a big problem." Where he disagreed was on the cause. He doesn't think that negativity comes primarily from him or any other executive warning about risk. His line was, quote, "I think it is fundamentally a crisis of trust. I think that ordinary people don't trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over."
And then he said the thing that I think is genuinely the most honest sentence any of these guys has said in a while. He said the most accurate criticism of AI companies, including his own, is that, quote, "we haven't yet delivered on our big promises to benefit the world. That is totally on us." And he added that promising AI will cure cancer is "more a cliche than it is inspiring" — what would actually move people is "actually curing cancer."
Now I want to give Amodei his due, because that's a rare bit of candor. He's basically saying: stop yelling at me about my tone, the problem is we keep writing checks that reality hasn't cashed. That's correct. But I'd also note where the self-interest peeks through, because it always does. On regulation, Amodei said Baker was setting up a false choice between distribute-AI-widely-with-no-rules versus concentrate-it-in-a-few-hands. And then he told you exactly what Anthropic aims for. Quote: "We try very hard to make proposals that disadvantage — slow down — frontier AI companies while advantaging smaller competitors." He argued AI is "structurally" a technology that concentrates power, that open weights only shift that concentration toward whoever has the most compute and chips, and that the right "rules of the road" can address the cyber and bio risks while institutionally constraining the big labs.
Here's my read, and I've been circling this trust theme for a couple of episodes, so let me zoom into the specific claim instead of hand-waving at the meta. When a frontier lab CEO tells you his preferred regulations would "slow down frontier AI companies while advantaging smaller competitors" — that sounds noble, and it might even be sincere. But you, the builder, should notice that Anthropic is a frontier company that already cleared the hard part. Rules that make it more expensive and slower to reach the frontier are a moat for whoever's already standing on it. I'm not accusing the man of bad faith. I'm saying: read the policy proposals, not the poetry. When someone tells you their regulation will conveniently kneecap their peers while helping the little guy, check whether you're actually the little guy in that sentence, or whether you're the thing getting slowed down on the way up.
And there's a companion piece to this trust fight, because while Amodei was defending himself from the charge of being too gloomy, another guy over at Meta got hammered for being too sunny. Let me connect those, because they're the same argument from opposite ends.
Mark Zuckerberg published a sixty-five-hundred-word essay this week called "The Future Is for Everyone," painting a bright picture where — his words — "everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about." And the reaction was rough. TechCrunch's own AI editor Russell Brandom wrote that the manifesto is "exactly why people don't like AI." On the Equity podcast, Kirsten Korosec, Rebecca Bellan, and Anthony Ha kicked it around, and Rebecca had the line of the week — she said Zuckerberg is positioning himself "almost like an anti-Dario." Where the frontier labs talk about slowing down and safety, Zuckerberg's basically saying no, we do not slow down, we cannot give an inch to China, and slowing down only hurts the individual.
Now here's the thing that makes this land differently than a normal booster speech, and it's a point the Equity crew made sharply. Rebecca said, and I'm quoting, "You look back to the social media days, he was saying that he wants to make sure that everyone has an outlet for talking to their friends and having a social network. And what do we have instead? We have ragebaiting and advertisements, and not connection." That's the whole problem with the messenger in one breath. It's not that "AI for everyone" is a bad idea in the abstract. It's that the last time this specific person promised to connect and empower everyone, a lot of people feel like they got the opposite. So when he shows up with a new promise, the immune response is automatic.
And let me give you the builder-relevant substance underneath the manifesto, because there is some. Meta shipped a model called Glimmer — that's their open-weight model, the one anybody can supposedly download and run on their own hardware. And separately there's Muse Spark, the more powerful one that stays locked behind Meta's APIs. So you see the real strategy under the "for everyone" language. Rebecca nailed the cynical read: Meta isn't winning the closed frontier race, it's not clearly winning the open race either, so it's trying to win on "personal empowerment" — the model on your own device managing your schedule, drafting your messages, always on, works with or without internet. And then Muse Spark stays as the paid, powerful, we-still-control-it option for anyone doing bigger projects. That's a coherent business move. It's just wrapped in Pollyanna language that curdles the moment you remember whose feed you're scrolling.
The tell, and I love this detail, is that Rebecca actually tried to go download Glimmer to her MacBook to play with it. Couldn't. You need specific hardware. As Kirsten put it, dry as a cracker: "It's not for everyone. Not yet." So the manifesto is titled "The Future Is for Everyone," and the product doesn't run on the laptop of a professional tech journalist. That gap — between the sermon and the download button — is the whole reason skeptics roll their eyes.
So step back and look at what you've got across these two stories. Amodei says the backlash is a crisis of trust and the fix is delivering real results, not better marketing. Zuckerberg says the fix is a bolder, sunnier vision and full speed ahead. And the public's reaction — hammering both of them — tells you neither the doom nor the boom is landing. People aren't rejecting the tone. They're rejecting the track record. That's the honest version of Amodei's own point, and it applies to Zuckerberg twice over.
Alright, let me get you out of the executive Twitter wars and into some hard metal, because there was a genuinely wild week in electric aircraft, and it's a good lesson in how young industries actually grow up.
The transportation newsletter over at TechCrunch, written by Kirsten Korosec, walked through two deals in the electric vertical takeoff and landing world — that's eVTOL, the flying taxi dream. And the headline one is a plot twist worthy of a soap opera. Archer Aviation announced it now owns its former rival, Wisk Aero. Boeing agreed to sell Wisk plus two other subsidiaries to Archer in exchange for a stake in Archer — Boeing now holds sixteen-and-a-half percent of the company. The other two pieces are SkyGrid, an airspace and air-traffic-management software outfit, and a drone maker called Insitu.
Now why is that a plot twist? Because back in 2021, Wisk sued Archer, alleging the "brazen theft" of confidential information and intellectual property. That fight dragged on for two years, Archer countersued for a billion dollars in damages, and it ended in an unusual settlement that not only killed the lawsuits but turned into a collaboration. And now the company that was accused of stealing from Wisk... owns Wisk. If you ever needed proof that in a small enough industry, today's defendant is tomorrow's parent company, there it is.
And I'll give you the little bit of history here because it's genuinely a strange lineage. Wisk started life as Kitty Hawk — the electric aviation startup led by Sebastian Thrun, backed by Google co-founder Larry Page. Kitty Hawk shut down in September 2022 — and I want to be clear that's an old date, that shutdown is years back, not this week's news. But its Cora program had spun into a joint venture with Boeing, that JV got renamed Wisk, and now here we are.
The second deal is quieter but honestly more telling about where this whole sector is heading. Joby Aviation bought a company called Resonant Sciences for five hundred million dollars. Resonant makes radio frequency and sensor systems, and it's becoming a dedicated defense unit — "Joby Defense." And that's the real story under both deals. These air-taxi companies set out to fly you to the airport over traffic, but the certification process for carrying human passengers is long and brutal and expensive. So while they wait, they're consolidating and they're chasing near-term revenue wherever it lives — and a lot of it lives in defense. Joby says it's sticking to its urban air-taxi mission, but it also sees revenue in defense, and revenue is the word that keeps a hardware company alive during a long regulatory winter. If you're a founder in any deep-tech, long-timeline space, that's the pattern to internalize: the dream funds the certification, but a paying side-business is what keeps the lights on until the dream is legal.
There were a couple other nuggets in that same roundup worth a quick mention. Uber did something out of character — it sold off its entire stake in Serve Robotics, the sidewalk delivery robot company that spun out of Uber more than five years ago. And here's the spicy part, per Korosec's little birds: Serve was reportedly caught off guard, only learned about the sell-off when Uber's regulatory filing hit. They're still business partners — Serve's robots stay on the Uber Eats app under a contract that runs to 2027 — but selling your entire equity stake in a partner and letting them find out from a public filing is a cold way to run a relationship. And on the robotaxi front, Waymo got clearance to expand its paid rides across a broader chunk of the Bay Area and LA, with room to grow into Sacramento and San Diego. Slow, permit by permit, the driverless map keeps filling in.
Now let me pull one thread out of that same newsletter and give it its own moment, because it connects to something we care about on this show — the security surface of everything. Buried in the notable-reads was this: Delta Air Lines is investigating a cybersecurity incident where an unidentified passenger allegedly set up a fake Wi-Fi network mid-flight, which spooked the pilots enough that they alerted air traffic control. Think about that for a second. A rogue hotspot at thirty-five thousand feet, named to look official, is now a thing pilots have to worry about. And Uber Freight, separately, is investigating a cyberattack and data breach that an extortion crew calling itself Helix took credit for. The attack surface isn't the data center anymore. It's the seat-back Wi-Fi and the freight logistics backend.
Which is a decent bridge into a genuinely practical bit of housekeeping, because your accounts on AI platforms are a target now too, and most people don't check them.
TechCrunch's cybersecurity guy Lorenzo Franceschi-Bicchierai put out a straightforward guide this week on how to tell if your accounts on ChatGPT, Claude, and Perplexity have been broken into, and I want to give you the actionable version because it takes ninety seconds and most of you have never looked. Standard advice up top: unique passwords in a password manager, and turn on multi-factor authentication. ChatGPT and Perplexity both offer it. Here's a detail worth knowing — Claude doesn't use passwords at all. Anthropic just emails you a login link every time. Which means there's no password to steal, but it also means your email account becomes the single key to your Claude, so lock that down hard.
The practical moves: on ChatGPT, click your name bottom-left, go to Settings, Security and Login, then Active Sessions — you'll see every device logged in, and you can boot any one you don't recognize or hit "log out all." Claude's is similar, under Settings then Account, look for Active Sessions, and you can log out of all devices. Perplexity's the odd one — it doesn't show you where you're logged in at all, so your only real lever is "sign out of all sessions" under settings, which forces a fresh login everywhere. Why does this matter for a builder? Because you've probably got API keys, custom instructions, and half-finished project context sitting in those accounts, and a stranger logged into your session is a stranger reading your business. Go check your active sessions after the show. It's free.
Alright, let me shift gears entirely, from your login screen to a factory floor, because for the deep dive today I want to slow down on the one story that I think is the most consequential for how physical products actually get built — Ford betting two billion dollars that Henry Ford's own invention is the thing holding it back.
Here's the setup, and this is from a recent report by Kirsten Korosec. Two years ago, Ford closed its Louisville Assembly Plant in Kentucky and scrapped the assembly line system it had used for more than a century. Not upgraded — scrapped. The moving assembly line, the thing Henry Ford himself launched, the innovation they teach you about in grade school as the birth of mass production. Ford looked at it and said: this is not how we're going to catch up. This week they gave an update saying the overhaul is on track, and in 2027 the plant will start producing the Fathom — an all-electric midsize truck priced under thirty thousand dollars, the first EV on Ford's new universal platform.
Now let me tell you why the stakes are so high, in Ford's own framing. Their previous EV efforts were a drag on profitability — they lost money on them — while Chinese competitors and Tesla leapt ahead building vehicles that sell at volume and at a profit margin. So the problem Ford is solving isn't "can we build an electric truck." They can build one. The problem is "can we build a cheap electric truck and actually make money on it." And they decided the century-old moving line couldn't get them there.
So here's what they replaced it with, and this is the part I find genuinely fascinating. It's called the "universal production system," and it was developed by a skunkworks team led by a former Tesla executive named Alan Clarke. Instead of one long line where a car crawls from station to station picking up parts, it uses what they call a three-branched assembly tree. Follow me here, because it's a real change in shape. On one branch, they assemble the front of the vehicle. On a second branch, separately and in parallel, they assemble the rear. And these use large single-piece aluminum castings — Ford calls them unicastings — which is a technique Tesla popularized, where instead of bolting together dozens of stamped parts you cast one big piece. Far fewer parts, far fewer welds. Then on the third branch, they build up the structural battery together with the seats, consoles, and carpeting. And at the very end, those three big components come together to form the vehicle.
Think about what that does. A traditional line is sequential — everything happens in order, one station at a time, and the whole line moves at the pace of its slowest step. This tree lets three chunks of the car get built at the same time, in parallel, and only marries them at the end. Ford says the result is they can assemble the Fathom a net fifteen percent faster than the vehicles that used to come off that Louisville line. And here's a detail that should make every software person perk up: as part of the rebuild, Ford nearly tripled the Wi-Fi coverage in the factory — up to 1,080 access points — because they need high-bandwidth, low-latency connectivity for software quality checks. The factory itself is becoming a connected system. Employees have already spent months training on the new setup at a development center in Allen Park, Michigan.
Now why am I making this the deep dive on a show for founders and builders, most of whom will never pour aluminum? Because this is one of the cleanest examples I've seen of an incumbent making a genuinely hard choice, and the lesson generalizes. Ford didn't have a technology problem — they knew how to build EVs. They had an economics problem, and the economics were locked into a process that was so foundational to the company's identity that questioning it was almost heresy. The moving assembly line isn't just how Ford builds cars. It's Ford's origin myth. And they killed it anyway, because the thing that made you great in one era can be exactly the thing that sinks you in the next.
For you, the founder, the parallel is uncomfortable but useful. The process you're proudest of — the workflow you invented, the thing that got you your first ten customers, the architecture you'd defend in a bar fight — that's a candidate for the scrap heap the moment the economics shift under it. Tesla forced this on the whole industry by treating the car as something to be cast in big pieces and assembled in parallel, and now the company that invented sequential mass production is copying the newcomer's playbook and hiring the newcomer's executives to do it. That's the whole game in one factory. The incumbent's advantage is scale and history. The incumbent's curse is that history tells you the old way is sacred. Ford just voted against its own history to the tune of two billion dollars, and we'll find out in 2027 whether the bet pays off. If the Fathom ships under thirty grand and Ford actually makes money on it, that three-branch tree becomes the template a lot of manufacturing copies. If it doesn't, it's a very expensive lesson in messing with what works. Either way, watch it — this is what a real bet-the-process pivot looks like when you can't hide it in a slide deck.
Let me bring it home with a couple of quicker hits, because there's product news you can actually use.
Google rolled out a feature — this went out in July and resurfaced this week, so I'll frame it as recent rather than brand-new — called Sheets canvas. The pitch is that you describe what you want to Gemini in plain language, and it turns your rows and columns into an interactive mini-app — a dashboard, a tracker, a seating chart — with no coding, and it stays synced to the underlying sheet in real time. Now, I'm a skeptic by trade, so let me be measured. On one hand, this is the natural direction — the spreadsheet has been the world's most-used programming environment for forty years, and layering a "just tell it what you want" interface on top is genuinely useful for the person who was never going to write a formula anyway. On the other hand, "AI builds you an app on top of your data" is a sentence I've now heard from roughly everyone, and the value is entirely in whether it holds up when your data is messy, which real data always is. It's available to Google's AI Pro and Ultra subscribers and rolling out to Workspace business plans. Worth a look if you live in Sheets; not worth rearranging your stack over.
And one more, for the indie developer in you. Marco Arment — the guy behind Overcast and Instapaper — shipped a new app called Unforgetful, and I mention it because it's a small, sharp example of building for a real human problem rather than a market-size chart. It's a reminders app for people who compulsively hit snooze. The trick: notifications repeat, even if you dismiss them, until the task is actually done and deleted. You literally can't lose a reminder by fat-fingering it. And it piggybacks on Apple's existing Reminders database, so Siri still works and you're not migrating anything. Twenty bucks a year. I flag it not because it'll change your business, but because it's a clean reminder — pun intended — that a well-built tool solving one annoying thing precisely still has a market, even in a world where everybody's building the everything-agent. Zuckerberg's out there promising an assistant that understands your whole life; Arment shipped an app that just won't let you ignore the dentist. I know which one I trust more this week.
So let me tie the ribbon on all of it. The through-line today, if there is one, is delivery versus promise. OpenRouter got a seven-billion-dollar validation not for a vision but for being the boring, working middle layer people actually route through. Amodei's most honest line was that the industry hasn't delivered on its promises and that's on them. Zuckerberg got hammered precisely because his last big promise curdled. And Ford is spending two billion dollars to stop promising cheap EVs and actually build one at a profit. The market, and the public, are done buying the manifesto. They want the thing that works. Build the thing that works.
That's the menu for today. This has been Barely Possible — I'm Tony DeLuca, thanks for spending part of your day in here with me. Go check your active sessions, and I'll see you tomorrow.