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 today's menu's got a little bit of everything — truck heists that read like a heist movie, a company going public while admitting it might blow up the world, and robot dogs that have a staff revolt on their hands. Buckle up, grab your coffee, and let's have at it.
Let me start with the one that stopped me cold, because it's not an AI model, it's not a funding round, it's not another lab bragging about benchmarks. It's guys in unmarked sedans ramming cop cars off the road to steal servers headed to a data center. That's the lead today, and I want to spend some real time on it, because it tells you something about where the whole AI boom actually touches the ground — literally, on a highway, at night, with a trailer full of what one expert called "pure gold bars."
Here's the story, and I'm working from a piece Ashley Belanger wrote for Ars Technica, which is pulling from reporting NBC News did talking to industry leaders, consultants, and investigators advising law enforcement. Since June, thieves have made away with more than a hundred million dollars in data center freight. A hundred million. And the way they're doing it is not smash-and-grab. It's sophisticated. It's digital. It's organized.
Walk through the playbook with me, because it matters for anybody building a company that ships physical stuff. First, the thieves hack freight brokers' accounts on those shipment-listing websites — the online marketplaces where loads get matched to truckers. Then they post fake shipments to lure real bids. A trucking company responds, and the thieves send them a malicious link. Victim clicks, software gets installed, and now the bad guys control that account and can start bidding on real shipments under a legitimate identity. Sometimes a real delivery driver gets hired to pick up the cargo and has no idea they're part of a scam. Other times they use fake driver's licenses — and here's the AI angle that's almost poetic — they're using AI to generate those fake licenses to get accomplices hired as drivers. So AI is being used to steal the gear that builds the AI. Either way, the goods get handed off to a driver who's in on it, and the stuff gets resold.
And in one of the largest thefts, they spoofed the GPS coordinates. The tracking app that's supposed to tell you where your ten-million-dollar trailer is? It's showing you a lie. The truck's gone, the dot on the map says everything's fine.
Now here's where it goes from clever crime to something uglier. Two of the biggest thefts NBC found — none of the equipment was ever recovered, and the off-duty cops hired to escort those shipments got taken out in hit-and-runs. One trailer carrying several hundred pounds of computer server equipment worth thirteen million dollars was headed to Fayetteville, Georgia — which, the piece notes, is home to one of the largest data center developments in the country. Two off-duty officers escorting it lost sight of the trailer after an unmarked sedan struck their car and forced them to stop. The trailer just drove on. And a private investigator named Gerardo Pachuca, a former LA County Sheriff's cop who investigates cargo theft, told NBC flat out: "The accident was not an accident. The bad guys wanted that load, so they intentionally took out the escorts."
Same thing happened with another shipment, California to Texas, server rack components worth around ten million. Escort vehicle gets struck by an SUV, trailer speeds off, and it didn't matter that the cargo was being monitored by a tracking app — because the coordinates had been spoofed.
Let me pull back and tell you why I think this matters beyond the shock value. Keith Lewis at a risk firm called Verisk CargoNet — they track cargo theft — put it in a line I can't shake: "It's like stealing pure gold bars. It's so much easier to steal this than to rob a bank." And he's right, and that's the whole thing. We have spent two years in this show talking about the AI economy as this abstract thing — valuations, compute, token prices, the S&P 493 that we covered just yesterday. But the physical reality is: there are now billions of dollars of extremely portable, extremely fungible, hard-to-trace high-value goods rolling down American interstates every single day. And wherever you concentrate that much value in one place, organized crime shows up. It's not a tech problem. It's gravity.
And here's the detail that should bother the builders listening. Unlike other cargo theft, this data center gear doesn't get dumped on an online marketplace. It gets broken down into parts and sold on gray markets — to tech firms that rely on questionably sourced equipment to keep their costs low. And in a lot of cases, investigators said, an overseas buyer is already lined up before the theft even happens, particularly in regions where access to high-end US technology is restricted. Like China. So you've got export controls on one end creating artificial scarcity, and on the other end you've got a thriving black market where a stolen GPU is worth more than a legally imported one. Scarcity plus value equals theft. Every time.
Now, the solution is almost insultingly simple, and that's the part I find darkly funny. Scott Cornell, who chairs the Transported Asset Protection Association — the big supply chain security nonprofit — said the overseas crime rings are "playing chess in the supply chain," but there's an easy path to checkmate: run background checks on your drivers. That's it. His exact line: "The only way this works is if the drivers are not properly vetted." The whole multi-million-dollar operation collapses if freight companies would just slow down and actually verify who's picking up the load. But everybody's moving fast, everybody's under pressure to ship, and the vetting is the first corner that gets cut.
And here's the kicker that I want every founder to sit with. The piece notes that to fix this, the industry "seemingly can't rely too much on AI." The old-fashioned work — knowing your clients, vetting your drivers, human diligence — that's the defense. The thing the whole industry is racing to automate is the exact thing that has to stay human to keep the trucks from getting jacked. I'm not going to turn that into a tidy little lesson, but I'll just leave it sitting there, because it's real.
So what's the takeaway for you if you're building anything that moves physical product — hardware, devices, anything with resale value? Three things. One, your tracking is a target, not a guarantee; if it can be spoofed, assume it will be. Two, your weakest link is probably a vendor's hiring process you've never looked at. And three, as AI infrastructure keeps scaling — and we've got a story later about SpaceX dropping forty billion on GPUs, so it's not slowing down — the physical supply chain around it becomes a bigger and bigger prize. States are standing up specialized task forces. One escort company CEO, Mike Duffy of Solutions Group International down in Southern California, told NBC they're escorting hundreds of loads a month, and much of it is tied to data centers. That's a whole shadow industry — armed escorts for computer parts — that basically didn't exist a couple years ago. That's the AI boom you don't see in the keynote.
Alright. Let me shift from trucks on the highway to a filing on Wall Street, because yesterday was a big one. Anthropic officially filed its S-1 to go public.
Now, we talked about the leaked versions of this S-1 last week — the cloud spend, the hyperscaler-as-competitor risk — so I'm not going to re-run all of that. But the official filing dropped Monday and there's genuinely new numbers worth your time, and there's one line in here that I had to read twice because it's kind of astonishing.
The topline: Anthropic plans to list under the ticker "ANT" — they're going on the Nasdaq. Revenue of sixteen-point-three billion dollars for the first half of 2026. For comparison, the same period last year? One-point-two billion. So that's revenue going up more than thirteen-fold year over year. That is a genuinely staggering growth curve, and I don't throw that word around.
And the losses are interesting too, and you've got to read them carefully. Operating loss of six hundred ninety-three million for the first half — which is actually an improvement, because the first half of 2025 they lost two-point-one billion operating. So the operating picture is getting better. But then there's the net loss, and here's where you need a cool head: net loss of eighty-point-seven billion dollars. Eighty billion. And before you fall out of your chair — that number is largely driven by noncash accounting adjustments tied to convertible debt. It's not eighty billion dollars walking out the door. It's an accounting artifact of how their convertible debt gets marked. But it's in the filing, and the company itself says plainly it "may not achieve or sustain profitability in the future." That's standard risk-factor language, but you should hear it.
Now here's the part I actually want you to internalize, because it's a founder story, not just a finance story. There's a companion piece by Rebecca Bellan at TechCrunch digging into the strategy, and it reveals how Anthropic is keeping enterprise customers locked in. As of the end of September: more than eleven thousand nine hundred customers spending a hundred grand or more a year. Nineteen hundred spending a million or more. And about two hundred fifty customers spending ten million dollars or more, annually, on Claude.
But the number that made me sit up is this one: net revenue retention of about nine hundred percent. Let me say that again. Nine hundred percent. For the folks who don't live in SaaS metrics — net revenue retention measures how much more an existing cohort of customers spends this year versus last year. A healthy, beloved software company might hit a hundred twenty, a hundred thirty percent. The absolute best-in-class land-and-expand businesses dream about a hundred fifty. Anthropic is reporting nine hundred. That means a dollar of revenue from existing customers a year ago has turned into nine dollars. That is not a normal software number. That's the signature of a company where customers start with one little API call and then it metastasizes across the whole organization.
And the mechanism is the thing worth stealing if you're building. The filing describes a classic ecosystem lock-in, the Amazon-and-Uber playbook. Uber started with rides, expanded to food delivery and freight. Amazon drew you in with cloud computing, then storage, then databases, until leaving got impossible. Anthropic's doing the same across three layers — the models, the developer platform, and the first-party apps like Claude Code and this thing called Cowork. More than half of their hundred-thousand-dollar-plus customers now use both the Claude platform and a first-party app, up from about thirty percent at the end of last year. And here's the money stat: the highest-spending five percent of enterprise customers who used Claude Code spent about twenty times as much per month as the average enterprise user who didn't touch the coding tool. Twenty X. So the strategy writes itself — get them on the API, then get the coding tool into the engineers' hands, then get the broader apps into everyone else's hands, and watch the spend compound.
They've even got a whole operation for it. An internal "Applied AI" team that embeds with customers to wedge Claude into their existing stacks, plus a new venture with Blackstone and other PE firms — they're calling it "Ode with Anthropic" — to put forward-deployed engineers inside businesses to build custom tools on Claude. That's Palantir's forward-deployed model, basically. Consultants with a product behind them.
Now — the governance thing, because this is the part that's genuinely controversial and you should know it before you'd ever buy a share. The seven co-founders — Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Christopher Olah — form what they call a "Founder LLC" that will collectively control fifty-point-one percent of shareholder voting power. And their Long-Term Benefit Trust gets to appoint most of the board. So here's the deal you're signing up for as a public shareholder: your voting power is not proportional to your economic ownership. You're buying into a company that, by its own admission in the filing, builds technology that poses — their words — "catastrophic or existential risks to humanity," and you, the shareholder, get almost no practical say in how they steer it. They say this structure exists to protect the AI safety mission. Maybe it does. But understand what it is: you get the financial upside, or downside, and the founders keep the wheel. That's a bet on seven specific people's judgment, forever. Eyes open.
That belief that the business keeps reinventing itself is actually baked into the filing — they say they'll spend less effort "entrenching our current products" and more effort "defining new ones," products meant to "dwarf the economic value" of what they sell today. That's an extraordinary thing to tell public investors: don't get too attached to what we sell now, because we plan to make it obsolete ourselves.
And it connects straight to the next thing, because the money has to come from somewhere, and the compute has to come from somewhere.
Which brings me to SpaceX, locking in a forty-billion-dollar GPU financing deal with Apollo. And this one's got a classic Elon victory lap attached to it, so let's unpack it.
Here's the structure, because the structure is the interesting part. SpaceX reached a binding commitment for roughly forty billion in financing for AI chips and data center gear. But SpaceX isn't buying the chips directly. An entity managed by Apollo Global Management — the big New York private equity shop — buys the chips and the server hardware, then leases the equipment to SpaceX. The leases run about five to five-and-a-quarter years. SpaceX expects to close by year-end, with the leases kicking in during the first half of 2027 as the GPUs — specifically Nvidia's next-generation Vera Rubin chips — actually show up.
Why do it as a lease instead of just borrowing? Because it diversifies the funding away from traditional debt markets, which — and this is the backdrop — have gotten squeamish about AI infrastructure. Investors are nervous about runaway capital expenditure and thin profits. SpaceX's CFO, Bret Johnsen, said they're seeing "less than one-year payback periods" on their AI infrastructure. Less than a year to pay back a GPU. If that's true, you lease all the GPUs you can get your hands on.
And the scale here is wild. SpaceX's AI segment — which includes Grok, the Cursor coding tool, and their cloud compute business — brought in seven-point-six billion of SpaceX's fourteen-point-six billion in Q3 revenue. More than Starlink and rocket launches combined. Let that sit for a second. The AI business inside the rocket company is now bigger than the rockets. Capital expenditure surged to nearly twenty billion in the quarter, up from under seven billion a year earlier, and Johnsen said they expect to spend thirty-five to forty billion in Q4 alone.
Now the victory lap. Musk used the forty-billion announcement to dunk on his competitors. Pointing to SpaceX's investment-grade status and the steady cash flow from Starlink and launches, he argued on X that the debt markets have basically slammed the door on companies that don't have major revenue outside of AI data centers. And then the quote: "That means the 200-plus neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many."
Guppy-level death. Only Elon.
Now, I'd take the gloating with a grain of salt — the man has every incentive to spook the capital markets for his pure-play GPU-rental competitors while he sits pretty on Starlink cash. But the underlying point is real and it's a founder lesson: if your entire business is renting out GPUs and you have no independent cash flow, the debt markets are getting cold on you fast. The companies that survive this are the ones with a second engine — Starlink revenue, launch revenue, something that isn't just "we bought chips and rent them out." If you're building in infrastructure, diversified revenue isn't a nice-to-have anymore. It's the difference between making it through and guppy-level death.
Alright, let me come back down to earth — Microsoft offered buyouts to around sixty thousand employees. Now, be careful with this one, because the headline sounds scarier than the reality, and I want to be precise.
This is not sixty thousand layoffs. It's a Voluntary Exit Offer — Microsoft confirmed it to TechCrunch. Sixty thousand employees are eligible, meaning they're being given the option to take a buyout and leave. The chief people officer, Amy Coleman, sent the note, which is up on Microsoft's own website. Eligible folks are at the vice president level and below, on select teams, in the US, Canada, India, and the UK. Bloomberg reports the participating divisions include Xbox, Windows, Copilot, and Finance. Some folks are excluded — data center workers, sales and marketing.
The framing from Microsoft is all soft edges — "greater flexibility," "more ways for employees to shape their next chapter," giving people "who may be considering a future outside the company a new choice." And the terms are similar to standard severance. The honest read: this is a way to shrink headcount without the brand damage and morale hit of a mass layoff. You offer the door, you see who walks, you save the cost of forcing it. They did a smaller version earlier this year with retirement-age employees — about eight thousand seven hundred eligible, and over thirty percent took it. Microsoft's total headcount at the end of June was two hundred twenty-three thousand.
Why does this matter to you as a builder? Because it's the quiet version of the workforce story. The loud version is "AI replaces jobs." The quiet version is the biggest, most profitable software company on earth gently nudging sixty thousand people toward the exit while it "reworks its workforce for the AI age" — their phrasing. No drama, no mass-layoff headlines, just attrition engineered on purpose. That's probably the template a lot of big companies copy. Watch for "voluntary exit offer" to become the polite phrase of 2027.
Now let me get to the one that's genuinely a conscience story, and it's close to home for the industry: Boston Dynamics. Hundreds of their own employees are trying to block a sale of robot dogs to ICE.
Quick framing before I go further — this is a story that's been building since late September, so I'm treating it as the current state of an ongoing fight, not breaking news. Here's where it stands, per Jeremy Hsu's reporting at Ars Technica. On September 25th, a Department of Homeland Security contract awarded one-point-three million dollars to a Connecticut company called Radeco to provide ICE with modified versions of Boston Dynamics' quadruped robots — the Spot robots, the famous robot dogs. Radeco upgrades them into a platform called Cerberus, kitted out with a thirty-times zoom camera, 360-degree cameras, infrared thermal imaging, an oxygen sensor for confined spaces like tunnels, and an optional robotic arm that can open doors and manipulate objects. The contract says they're meant to detect radiation, toxic gas, and explosive threats, and to support counter-tunnel operations against smuggling and trafficking networks.
Now, the workers revolted. More than three hundred employees — out of a workforce of about fourteen hundred, so better than one in five — signed a petition to executives asking them to kill the deal. Half the signatures were anonymous, which tells you something about how safe people felt putting their names on it. One line from the petition: "Many of us have gotten messages from friends and family, asking us in disbelief and horror if this news is true, and how we can live with ourselves if it is."
And they weren't abstract about it. The petition specifically referenced an ICE and Homeland Security Investigations raid last September that detained four hundred seventy-five mostly South Korean workers at a Hyundai facility in Georgia. And here's the knot at the center of this: Hyundai became the parent company of Boston Dynamics back in 2021. So the employees wrote, "We cannot be proud of working at Boston Dynamics when that work is being used by a government agency that brutalizes civilians, including some of our own colleagues at Hyundai." Their own corporate family got caught up in a raid. That's personal.
The company's response? They're standing by the deal. The spokesperson said, "Spot robots are tools that help keep people out of harm's way. Regardless of any political implications or personal feelings about the organizations involved, if our robot can help keep people safe, we believe it should." And the former interim CEO, Amanda McMaster, told workers in an email that the decision is "in line with our purpose of ethically advancing intelligent machines in service of humanity," and urged them to "respect and accept this decision, even if you do not agree with it."
Boston Dynamics is leaning hard on two things. One, the robots won't be weaponized — their terms prohibit using Spot as a weapon or to harm or intimidate anyone, and they've previously signed an open letter against weaponizing general-purpose robots. Two, they say the gear will mainly go to the Homeland Security Investigations unit that chases transnational crime. But the Brennan Center for Justice has documented that same unit being heavily involved in workplace raids and deportations, and under the current administration thousands of its agents got diverted to immigration enforcement. So "it's just for the crime-fighting unit" is doing a lot of load-bearing work in that defense.
Here's why I'm covering this as a builder story and not just a culture-war flashpoint. The employees made an argument that's pure business risk: they said this deal "makes Boston Dynamics a less attractive employer" and will hurt hiring and retention, right as the company is standing up a hundred-million-dollar robotics and AI facility that's supposed to create over a thousand new jobs. That's the tension every hardware and AI founder is going to face. Your most lucrative customer might be a government agency. Your best engineers might refuse to build for that customer. And "we don't weaponize it" is a thin line when a camera with thirty-times zoom and autonomous navigation gets pointed at a workplace. A tool's capabilities don't care about your terms of service. The worker backlash is the real cost here, and leadership is betting the contract is worth it. We'll see if the talent agrees.
Okay, let me move from the heavy stuff to a couple quick hits, because there's genuinely useful founder material in here.
Databricks' co-founder, Arsalan Tavakoli-Shiraji, got up at TechCrunch Disrupt and shared his secrets for landing giant enterprise customers, and honestly it's some of the most practical advice I've seen in a while. Databricks is a hundred-ninety-billion-dollar company now with over seventy percent of the Fortune 500 as customers — but it was a tiny startup once, and he was the guy bringing in the first big deals. His core lesson: everybody thinks it's about having the best tech. It's not. "People and processes matter the most." He told a story about finally landing a meeting with a big prospect, only to realize it was the same guy who'd signed a mega-deal with their competitor — a guy who had, in his words, "bet his career on that competitor." Dead on arrival. No tech demo was ever going to beat that man's career incentive.
His actual tactics are gold. One: sell under a bigger company's umbrella. Databricks couldn't get a meeting with UBS, the Swiss bank, on its own. Then it became an Azure cloud partner with Microsoft, and suddenly Microsoft's salespeople were booking the meetings and vouching for them. That meeting that was impossible? Easy. Distribution beats product. And notice — this is exactly the "joint go-to-market" move Anthropic described in its S-1, riding AWS, Google Cloud, and Azure into customer accounts. The pattern repeats at every scale.
Two: when selling AI, address what he calls the four C's — context, choice, cost, and control. Context, meaning the AI can learn your company like a new employee. Choice, meaning you're not locked to one model provider. Cost, meaning predictable spend. Control, meaning safeguards and data protection. If you're pitching AI into an enterprise, those are the four objections. Have an answer for each.
And three — this is the big opportunity — he says go attack the expensive legacy systems enterprises think they're stuck with. He asks CIOs to rate themselves one to five on AI. Everybody says "three leaning toward four." But he says ninety-nine percent are actually bucket one or two — meaning their people just have ChatGPT access and that's it. The real value isn't there yet. Where it is: every bank still runs ancient IBM AS/400 systems and mainframes that nobody wants to touch because nobody knows what the code does anymore. And now AI agents can go in, read that code, and actually decompose what it does. That opens the door to products that either put a modern face on those dinosaurs or replace them entirely. If I'm a founder looking for a wedge right now, "AI that finally untangles the mainframe the bank's been afraid to touch for thirty years" is about as real a problem as it gets.
Couple more fast ones.
Bedrock Robotics raised six hundred million at a four-point-one-billion valuation for self-driving excavators. Two-year-old company, Series C led by CapitalG and Emergence, and the pitch is dead simple and I kind of love it: there's a labor shortage of skilled operators. Takes five years to master running an excavator, and the trade group says more than forty percent of construction workers retire in the next five years. So Bedrock makes a retrofit kit — you bolt it onto existing construction vehicles, and they drive themselves. Their machines are already working commercially at a Nevada water treatment plant and big sitework projects in Texas. CEO Boris Sofman used to run Waymo's shuttered self-driving trucks program. The lesson here: autonomy aimed at a desperate, unsexy labor gap is a much cleaner business than autonomy aimed at consumers who have opinions. Nobody's writing think-pieces about whether an excavator should feel like a friend.
And Instagram is going after YouTube — they're testing "Features," videos up to ninety minutes long. After years of everybody, even LinkedIn, chasing short-form vertical video, Instagram's pivoting to long-form. And Adam Mosseri was refreshingly honest about why. The reason is money. YouTube pays creators fifty-five percent of ad revenue through its Partner Program — paid out over a hundred billion dollars to creators in four years. Instagram, by Mosseri's own admission, mostly pays creators through brand deals, runs a small performance program, and "we spend more money than we make back by a long shot." Short-form is just hard to monetize — there's no room to jam ads into a nine-second clip. So Instagram wants long videos so it can eventually run ads and finally compete for the YouTube creator's paycheck. The honest read for anyone building in the creator economy: the platform that pays per-view wins the serious creators, full stop. Instagram knows it's been losing that fight, and this is the counterpunch. Whether it lands depends entirely on whether Meta actually shares the ad money. The spokesperson's answer on that was, and I quote, "it's too early to share a long-term monetization model." That's not a no. It's also not a yes.
And since we're on Apple — quick two-parter. Apple unveiled a new iPad mini with an OLED display and a faster A20 Pro chip, starting at six ninety-nine, dropped just days before their "Pro and Tell" event Wednesday. Fine, it's a nice tablet. But the story right next to it is the one that stings: the iPhone Duo, Apple's two-thousand-dollar foldable, is cracking. Bloomberg reports early users complaining about screens cracking "after just hours or days of use" — some hearing a loud crunching noise, then a crack appears. 404 Media reported big fat creases down the middle of the folding screen. Apple told Bloomberg it's "aware of an issue impacting the inner display on a small number" of units. Apple promised it had finally cracked the foldable form factor — and, well, it cracked something. The lesson's old but eternal: shipping a notoriously hard hardware problem on a flagship timeline is how you turn your most exciting launch in years into a durability headline. Ask anyone who shipped a foldable before them.
Alright, let me land the plane with one palate cleanser, because I can't send you off on cracked screens and robot dogs.
Ars ran a lovely interview by Jennifer Ouellette with the medical historian Lindsey Fitzharris about her new book, Sleuth-Hound: The Case of the Real Sherlock Holmes. And the hook is this: Sherlock Holmes was real — or at least, he was inspired by a real guy. A surgeon at the University of Edinburgh named Joseph Bell, who taught Arthur Conan Doyle when Doyle was a young medical student. Bell was famous for diagnosing a patient's whole life story just by looking at them — he once figured out the exact walking route a patient took to the infirmary by identifying the type of clay on their boots. Doyle lifted that straight into The Sign of the Four.
But here's the part I loved, because it's a quiet lesson about how we see people. Fitzharris says Conan Doyle took the flashiest, most recognizable trait of Bell — the cold deductive parlor trick — and transposed it onto Holmes. But the real Bell was a deeply empathetic family man who raised three kids alone after his wife died young and never remarried. Fitzharris's husband is a caricaturist, and she makes this beautiful point: a caricaturist can only caricature people they don't know well, because caricature is about first impressions. Doyle knew Bell as a student knows a professor — from the outside. So he caricatured the cold genius and missed the warm man. The question she poses: would Holmes even exist if Doyle had actually known Bell? Probably not — or not like that.
There's a line in there for all of us. We caricature the people we only half-know. The flashy trait becomes the whole person. Something to chew on while you're updating your pitch deck.
And on an optimist's note, because I liked it — Fitzharris, who's been through cancer herself, said her superpower is making people feel good about living today. Her words: "Today is the best day to be alive and to get that diagnosis, and tomorrow will be a little bit better." Progress isn't a straight line, she says, but if you had your medical crisis a hundred years ago, it would've been a whole lot worse. In a week full of truck heists and cracking phones and companies admitting they might end the world, I'll take that.
That's the menu, kiddos. Vet your drivers, watch your net revenue retention, and don't caricature the people you only half-know. I'm Tony DeLuca, this has been Barely Possible, and I'll catch you on the next one.