Nate Roth

my wife just asked me how much ethernet we own ngmi

ChatGPT can't read Consumer Reports, but it can read the listicle where a brand ranked itself number one. Consumer Reports has tested products since 1936, and thier robots.txt tells GPTBot to stay out. same for Google's AI crawler and Common Crawl. 60% of high-credibility news sites now block AI crawlers, and i get why. bots passed human traffic on the web this June. the sites with real testing labs and real reporters locked up. so browsing mode reads whatever left the door unlocked. researchers checked 26,000 sources cited by AI search engines and one in six was written by another AI. 44% of ChatGPT's product citations come from best-of listicles, some published by the brand sitting at number one on its own list. there's also post hoc citation. the model writes the answer first, then goes hunting for pages that agree with it. so click the link before you buy anything a model recommends. read the actual page.

zero to one taught a decade of founders to perform crazy for the term sheet. the page that escaped into the culture is the founder one. thiel and blake masters observed taht the borderline, narcissistic, edge-of-society types build the outliers. in 2014 that was a real signal. the weirdness was a byproduct of conviction nobody was paying for yet. then the book sold 3 million copies and the signal became a costume. adam neumann went barefoot through the wework office trailing a $47 billion mark. sbf played a video game on the call where sequoia decided to invest, and the partners loved it enough to celebrate it in a glowing profile on their own site. sequoia later wrote that $214 million position down to zero. a pattern stops predicting the moment it starts paying. thiel spotted authentic strangeness back when acting strange cost a founder money. the book itself did the repricing. every seed round since has carried a little neumann premium.

Hermès just hired 50 human illustrators to draw its social posts by hand. the internet is having its organic-food moment. processed food ran this play a century ago. food scientists built cheap calories that hit harder than the real thing, real food couldn't compete, and it slid off most plates. a farmers market tomato now costs four times the canned version. fresh food became a class marker. AI content is cheap calories. machine output already outweighs human output online, and it's built for the scroll the way fructose is built for the tongue. so luxury is doing what luxury always does. it prices the scarce thing. Dior prints classic novels on totes. Bottega shoots a campaign about hands. Hermès pays 50 artists. the product is proof a human spent time. the part that bothers me is the split underneath. Jack Clark cofounded Anthropic and limits his own kids' AI time becuase the algorithm freaks him out. the people building the feed are buying their children out of it. everyone else gets the cognitive dollar menu. "organic" took thirty years to go from hippie label to a $60 billion grocery aisle. "made by humans" will get there faster.

asking a VC to sign an NDA tells them the only thing you own is the idea. the exceptions are narrow. multi-exit founders, or deep IP plays waiting on FDA approval. everyone else gets a polite pass and a quiet downgrade. it's only natural. you spent two years on this thing and a stranger with a checkbook wants the whole playbook in a thirty-minute call. the paper feels like protection. but a fund partner is holding three decks with your exact thesis this quarter, becuase founders keep hitting the same pain point at the same moment. no agreement legislates that coincidence away, and no fund accepts lawsuit exposure on a deal it will pass on 99 times out of 100. what protects you is being three months further ahead than the other two decks. a live customer number. a shipping cadence a copycat can't match without your two years of scar tissue. an idea that dies from being described was already dead.

a luxury brand going viral is spending down the only asset it owns. every luxury boardroom is hearing the same brief right now. get younger, democratize, meet Gen Z in the feed. Bain counted 50 million customers walking away in two years, so the panic is earned. watch what the panic produces though. Bentley shipped a chalet edition with logos sized for a ski rental shop while sales sit under 10,000 cars a year, down from almost 15,000 at peak. Audemars Piguet tried to introduce a new generation to the house and turned a status symbol into streamer bait instead. it took five days for the watch forums to call it the NFT of watches. standing still feels like death when the chart points down, and Aston Martin has tried nothing beyond an F1 collab that smells like a licensing brand in the making. but scarcity is the inventory. every skit and every blown-up logo sells off a piece of a name taht took a century to build. Patek Philippe posted nothing this week and won it.

the $750 billion question in AI is where the money exits the circle. everyone is pointing at Cisco right now. fair play. in 1999 Cisco lent telecoms cash so they could buy Cisco routers, then booked the sales as revenue while the stock went vertical. when the carriers went under, Cisco recovered pennies on the dollar. NVIDIA's version is bigger. $100 billion into OpenAI, which turns around and buys NVIDIA chips. a backstop under discussion so OpenAI can lease compute it can't yet pay for. the latest round of deals runs past $750 billion. Jensen Huang calls the circularity concern ridiculous, which is exactly what you'd say in either scenario. the telecom loop collapsed at its exit. dial-up subscribers never earned back the fiber. 47 carriers went bankrupt between 2000 and 2003 while thier networks sat almost empty. a loop only becomes a bubble when nobody outside it pays. and right now most of the paying happens inside.

Anthony Bourdain called elite Yelpers the lowest form of human being and got labeled a snob for it. Socrates made the same argument 2,400 years earlier with the ship of state, a crew fighting over the helm, none of whom ever studied navigation. Athens called him an elitist too. neither man was protecting the powerful. Bourdain ate on plastic stools in Hanoi and deferred to street vendors and grandmothers. the claim was narrower. judgment is a trained skill. what he couldn't stand was the owner who scrapes money together for a year, works her heart out, and gets "worst dinner ever" posted ten minutes after her doors open. we credential pilots and electricians. a stranger with a ring light can end a family's livelihood before the ovens cool. the fix is weighting. platforms already know which reviewer has eaten at 600 places and which signed up angry last Monday. show that history next to the stars adn let diners decide what a one-star from a first-week account is worth.

the research shows influencers are hated as a group, but loved one at a time. the category is vapid and the world would improve if it vanished tomorrow. but the foodie who surfaces hidden local gems stays. the numbers hold the same split. 5% of consumers fully trust influencer endorsements. 58% have bought something because of one. we like being marketed to. we buy things that confirm who we are or who we're trying to become, and we only resent the ad when the messenger is wrong for us. the influencers everyone hates are the ones who feel too visible. good marketing, delivered by the right avatar, feels invisible. enjoyable, even. before signing the next creator, ask whether their followers would defend this person by name in a group chat.

the empire state building went up in 410 days. the millennium tower in San Francisco took five years and multiple planning committees. it now leans 29 inches and the engineers are still arguing about the fix. the good-or-fast trade-off is posed by people who can't do both at once. i get why the objection exists. shipping sloppy in healthcare breaks real things, and slow can look like diligence from the inside. but Bret Taylor rebuilt Google Maps in a weekend, it loaded 10x faster at a third of the size. Tony Fadell shipped the first iPod seven months after kickoff. the same heads-down sprint taht makes the speed makes the rigor. slop comes out of a calendar full of alignment meetings.

the fastest risers i've seen skip the ladder by volunteering for the ugliest ticket in the queue. Tanay Tandon calls these employees "heat seeking missiles for pain." the missile reads 100 gnarly support tickets in a morning, solves 30, traces the root cause, and has engineers committed to a fix before dinner. at most companies that same loop takes six months. support escalates. a PM schedules discovery. a board of tickets appears, slated for Q1 release. ask your PM how those are going. The worst folks hide. They do the easy "strategic" work and send clean status emails while the real problem sits there getting bigger. The second-worst group, people who raise the alarm adn offer no path to killing it. noticing is free now. an LLM can notice. the scarce skill is the appetite to sit inside the pain until it's dead.

bots passed humans on the web in June, and your homepage is still dressed for the minority. everyone is still running the old playbook. redesign the hero, tighten the copy, test the button. i run an agency that makes those things, and the humans who do land still deserve them. but a person shopping for a camera checks five sites. the agent doing the same job checks 5,000, reads them as raw text, and hands its user one answer. Monday.com just shipped a signup flow only an AI can complete, a reverse CAPTCHA, becuase they want to be the option the agent hands back. one web for people, one for machines, and the machine web is already the bigger room.

$2.5 trillion went into AI this year, but only 2% of companies let it run anything that matters. Notion surveyed 6,000 professionals and mapped four levels of AI adoption: level one is ChatGPT open in a separate tab, drafting emails, never touching company data. most companies live there. level two is where AI is embedded in your tools, code, docs, and CRM. 31% are here. level three has agents running entire workflows end to end where a human simply reviews the output. a paltry 10% are here. level four has agents run complex business-critical processes autonomously. only 2% are here. 60% of executives say their company is ready to deploy agents. 36% of teh people who would actually run those agents agree. the same dataset shows senior leaders are the heaviest agent users in the whole sample. so the exec is reporting his own usage as the company's. one heavy user in the corner office doesn't count.

the last three people a company promoted tell you more about its culture than anything the CEO says on stage. past a thousand employees, the reward system flips. impact stops paying. protecting an exec's ego pays. keeping the hierarchy undisturbed pays even better. in Gary Hamel's survey, 96 percent of respondents at companies over 1,000 employees said a front-line person couldn't easily launch a new initiative. i get why it happens. at that size you need process or the machine eats itself, and the process was the right call when it got written. but process defends itself. any pushback gets logged as a character flaw. so the best operators learn to nod. the sharpest ones leave. a bad quarter hits and leadership looks around at a room trained to agree, and nobody in that room has taken a real risk in years, becuase risk was priced out of every promotion decision. the tell is charisma. a company promoting its smoothest people is running a political culture, and political cultures cannot execute hard pivots.

the most expensive rooms in Los Angeles now start by putting a sticker over your phone camera. at Versailles, nobles went into debt to stay visible. Louis XIV made presence at court the whole game. retreat to your estate and your power evaporated. for three hundred years wealth and visibility moved together. at San Vicente Bungalows the sticker goes on before your first drink, and posting from inside gets your membership pulled by morning. Jeff Klein built a waiting list out of the promise that nothing leaves the room. when anyone can broadcast, the broadcast stops carrying status. in HBO's I Love LA, Tallulah wears a WI SPA shirt you cannot buy. the only way to own one is to steal it. proof you were there, no post required. anti-merch works becuase nobody who wasn't in the building can fake it. documenting an experience now marks you as the person who needed the documentation.

building in public in 2026 is free market research for the studio that will clone you. every app founder gets the same playbook. document the journey, post daily, build an audience, launch to your own crowd. i handed out that advice myself about a decade back. meanwhile the watchers got faster. Witch Beam shipped Unpacking and a clone topped Apple's free charts within three months of release. that studio worked from a finished game. a daily public roadmap makes the job even easier. keep the useful half of the old advice. learn to talk into a camera without cringing. then skip the influencer arc adn buy the distribution. run those same videos as Meta and TikTok ads pointed straight at the product. build in silence until showtime.

the AI label in your first ten words now prices your deck at a discount. every founder is still being told to lead with the model. meanwhile the buyers already ran the experiment. S&P Global surveyed over a thousand enterprises and 42% abandoned most of their AI initiatives before production, up from 17% the year before. those are the exact customers seed decks promised would show up with budget. Marshall Sandman at Animal Capital put it bluntly. yank AI out of the opening and sell a service clients already pay for. he's right. and he's early to say it out loud. put the model in the back office and the line item on the slide. a bookkeeping firm that closes the month in two days instead of ten never has to say AI. the CFO signing the contract is buying the eight days.

Chinese labs gave their frontier models away and took 30% of global AI usage in under a year. everyone keeps framing US vs China as one race for AGI wiht one finish line. watch the revenue instead. for OpenAI and Anthropic, the model is the business. billions in training costs, paid back token by token. releasing the weights would hand the product to anyone with a GPU, so closed is the only rational call. Alibaba runs different plumbing. Qwen ships open under Apache 2.0, and 70% of new derivative open models worldwide now build on it. Singapore picked Qwen over Llama for its sovereign AI stack. the return shows up as defaults. Google already ran this play. Android was free because the OS was bait. search and the app-store cut were the business. i get the closed logic. venture money wants revenue this quarter, and weights are the only asset on the balance sheet. but a lab that monetizes indirectly can price the model at zero forever, and zero is a hard number to compete with.

everyone who can type is a software engineer now, yet almost nobody outside Silicon Valley acts like it. the whole industry is grinding on AI productivity. better emails, cheaper video. i run an agency, i use these tools every day, the speed is real. it's also the small story. the Pets.com read on the mid-2020s. the printing press turned scripture into a mass product. industrial machinery pulled people into cities. for the last 20 years the tool was software, and you needed a programmer to type the magic words. that gate came down. Kevin Roose, a NYT columnist who can't code, shipped working apps with a chat box and some patience. and yet the people building with this are the same people who could already build. everyone else is asking the model to punch up an email while sitting on the biggest transfer of capability in decades. a software engineer is anyone who can make a computer do a real thing. taht includes you, today, on the subscription you already pay for.

the best venture investor of the last decade doesn't run a fund. Alphabet just posted its biggest quarter ever, $62.6 billion in profit. nearly half of it, $28.7 billion, came from marking up private stakes. mostly Anthropic. Search and YouTube are still printing. teh portfolio out-earned them anyway. in one quarter. corporate venture is supposed to be where returns go to die. i've watched strategy memos strangle good checks, every deal has to serve the mothership or it dies in committee. Google wrote the SpaceX check straight off the balance sheet. $900 million in 2015 at a $12 billion valuation. eleven years later that stake trades north of $100 billion. a traditional fund holding the same position sells in year seven to show DPI, then watches the last 10x from the sidelines. no LPs, no fund clock. the core business prints cash, so nothing ever forces a sale.