Is Google Search Cooked? + We’re Getting a U.S. Crypto Reserve? + What You’re Vibecoding
Summary
Google’s AI Mode is a defensive reinvention of a Search business that generated $54 billion last quarter. It arrives alongside wider AI Overviews and a Gemini 2.0 upgrade; the experiment replaces 10 blue links with bespoke answers while relegating sources to carousels and small chain-link icons. Google is balancing a “once-in-a-generation chance to reinvent” against the risk of cannibalizing its core search business.
The competitive threat is still numerically small, but defaults could accelerate it dramatically. Google’s market share slipped below 90% in late 2024 for the first time since 2015; analysts cited by the hosts expect ChatGPT to reach 1% of search by year-end, versus Bing’s roughly 4%. Casey’s call: if AI Mode becomes Google’s front door in “two, three, five years,” then “that’s the ball game.”
AI answers could pull substantial traffic and economic value away from web publishers even if the bleakest estimates are wrong. Tolbit measured 91% fewer referrals from AI-search interfaces and 96% fewer from pure chatbots, though Casey challenged its crawler-based methodology. Kevin’s downside case was blunt: even 45% less traffic would be “a cataclysmic event” for much of media.
Chegg offers an early specimen of what happens when an answer engine absorbs a supplier’s product. The company charged about $15 monthly for more than 100 million homework answers and was valued at $12 billion in 2021; after Google began answering such questions directly, its stock became “basically a penny stock” and it started exploring strategic alternatives. Casey called Chegg an unsympathetic but important “bellwether.”
Trump’s executive order ultimately created a narrower Bitcoin reserve than the preceding rhetoric implied. It retains Bitcoin already seized by the government and authorizes “budget-neutral strategies” for acquiring more, while placing forfeited non-Bitcoin assets in a separate stockpile that will not buy additional tokens. The order therefore rules out explicit purchases of the named non-Bitcoin tokens and is described as carrying no additional taxpayer cost for Bitcoin acquisitions.
The proposed inclusion of XRP, Solana and Cardano exposed both conflict concerns and a deep split inside crypto. Trump-linked interests, Ripple chief Brad Garlinghouse’s political activity and the Trump family’s World Liberty Financial holdings made the proposal look to Casey like “a classic case of self-dealing.” Even Coinbase’s Brian Armstrong, Tyler Winklevoss and crypto analyst Nick Carter resisted broad inclusion; Carter warned against turning Bitcoin into “the plaything of the government.”
The quieter investable shift may be crypto’s integration into conventional finance, not the reserve itself. A bipartisan stablecoin framework, efforts to move authority from the SEC to the less aggressive CFTC, and possible XRP or Solana ETFs would create more channels from brokerage and retirement accounts into crypto. The broader warning was structural: “The walls are coming down,” potentially making the next crash relevant to people who never chose to trade tokens.
Vibe coding is already lowering the cost of narrow, domain-specific software, but it has not eliminated debugging or integration risk. Listeners built an anesthesiology simulator, an accessibility-focused speed reader, a flavor-pairing tool and reusable photo-processing scripts; the strongest projects paired domain knowledge with precise specifications. Kevin’s overactive maintenance alerts captured the constraint: users can make something that “works, sort of,” yet have no idea how to repair it.
Deep dive
1. AI Mode puts Google’s core franchise inside its own disruption
Google announced AI Mode alongside wider distribution of AI Overviews and an upgrade of the underlying model to Gemini 2.0. The early experiment initially requires both a Google One AI Premium subscription and an opt-in through Google Labs; neither host had been granted access.
Unlike the standalone Gemini app, AI Mode remains part of Search and prominently incorporates website links in horizontal carousels. Yet its primary output is a chatbot-style essay, as illustrated by Google’s response explaining déjà vu, rather than the traditional list of 10 blue links.
The citations may be more decorative than traffic-generating. Casey observed that the small chain-link icon following each paragraph will register as a hyperlink to sophisticated users, while many others may see it as “a fancy period”; the interface is “not going out of their way to entice people to click.”
Casey’s framing: Google is caught between keeping pace with ChatGPT and Perplexity and protecting a core business that produced $54 billion in the most recent quarter. He nevertheless sees AI Mode as evidence that Google recognizes a “once-in-a-generation chance to reinvent the search experience.”
2. Search habits are shifting before market share fully reflects them
Google still handles roughly 90% of searches, but analysis cited by Casey found its share fell below that level in the final three months of 2024 for the first time since 2015. Some analysts expect ChatGPT to reach 1% by year-end; Bing, despite its longevity, has about 4%.
Kevin now reserves Google largely for navigational queries such as train schedules, restaurant menus and known webpages. Product recommendations, appliance repairs and similar advice increasingly go to ChatGPT or Claude; he estimates chatbots handle at least half of what he previously Googled.
Casey still prefers Google or Google Maps for nearby restaurants, doctors and other local businesses, but uses chatbots for how-to questions and trivia—even when Google could answer correctly—because a single sentence beats “a sprawling webpage” containing six ads and unwanted widgets.
A Verge survey cited by Casey found 42% of respondents thought search engines were becoming less useful. He attributed that judgment to several forces rather than AI alone: advertising, deteriorating websites, “AI slop,” and the superior fit of direct answers for problems such as repairing a faucet.
3. Defaults, not early adopters, determine whether Search is truly cooked
Kevin’s pushback — worth keeping: warnings about AI search destroying publisher traffic have circulated for more than a year, yet the predicted collapse has not arrived at the expected scale. People change habits slowly; AOL, Yahoo and Hotmail retain users long after newer alternatives emerged.
Casey agreed that a hidden, paid experiment mostly attracts power users. His decisive condition is distribution: if AI Mode becomes Google Search’s default rather than requiring new behavior, “then that’s the ball game,” because the power of defaults can carry the new interface to billions.
Google’s caution also follows conspicuous AI Overview failures, including advice to eat rocks and put glue on pizza. Casey believes the slow rollout is driven more by that embarrassment than inference costs, which Google has said are falling as it develops more efficient ways to serve queries.
Politics is another source of product anxiety. Google warned that AI responses might unintentionally display a persona or opinion; Casey said the company is “terrified of politics” after earlier Gemini controversies, yet users ultimately want a product whose opinions the company is prepared to stand behind.
4. AI answers threaten the referral economy even under conservative assumptions
Casey divided queries into those needing an answer that is merely “good enough” and those where health, money or work requires the original source. Specialists such as journalists will inspect a New York Times, Wall Street Journal or Bloomberg page; most users “are not clicking those dang links.”
Tolbit reported that AI-search interfaces delivered 91% less click-through traffic than conventional Google searches, while pure chatbots generated 96% fewer clicks. Casey questioned its methodology because it traced traffic from known AI scrapers and divided it by the total number of times those scrapers hit publisher sites; crawler hits do not correspond one-for-one with searches.
Kevin’s stress test did not require accepting Tolbit’s exact figures: if the true decline were only half as severe—about 45%—the result would still be “a cataclysmic event for much of the media industry.” Google says AI Overviews encourage more searches overall, but the impact on individual publishers does not seem to have been positive so far.
Chegg illustrates the mechanism beyond news. Its $15-a-month service contained more than 100 million test answers and carried a $12 billion valuation in 2021; after Google surfaced homework answers directly, Chegg became “basically a penny stock” and began exploring strategic alternatives. Casey’s conclusion: “Chegg is not going to be the last business.”
5. The executive order separates a Bitcoin reserve from an altcoin stockpile
The interview with David Yaffe-Bellany occurred before the policy became official. A closing update explained that Trump subsequently signed an executive order establishing a strategic Bitcoin reserve built from assets already seized or forfeited in criminal and civil proceedings.
The Treasury and Commerce secretaries may develop “budget-neutral strategies” for acquiring additional Bitcoin, with no added taxpayer cost promised. The order did not explain what those strategies would be, but it moved beyond simply preserving the government’s current holdings.
Non-Bitcoin assets receive different treatment in a separate US digital-asset stockpile. The government may retain tokens obtained through forfeiture, but the order says it will not purchase additional non-Bitcoin assets—a meaningful retreat from expectations of open-market buying across several named tokens.
6. The economic case for a national crypto reserve remains unresolved
David said “Bitcoin reserve” or “crypto reserve” could describe “10 different sorts of ideas.” One strategic argument imagines the dollar eventually losing its position to Bitcoin or broader cryptocurrencies, making a large US stockpile useful for borrowing, spending or financing domestic and foreign projects.
Casey pressed on what owning Bitcoin actually accomplishes even in that scenario. David’s answer remained deliberately incomplete: it would be an asset the government could borrow against or spend, but “it’s not totally clear how this would work.”
The second argument resembles a sovereign wealth fund rather than an emergency reserve: buy Bitcoin at roughly $90,000 and benefit if advocates’ imagined value reaches $20 trillion in 10 years. That thesis is fundamentally an investment bet, unlike oil that powers machinery or reserves that address physical scarcity.
Crypto advocates distinguish a “stockpile,” meaning the government keeps coins it already owns, from a “reserve,” meaning it buys more. Retention requires only a policy change; large purchases would most obviously require congressional appropriation, though proponents have floated unconventional legal routes for unilateral action.
7. Altcoin inclusion turns government policy into a distributional fight
Trump initially named Solana, XRP and Cardano, then added Bitcoin and Ethereum in a subsequent post “as if” they were afterthoughts. David cautioned that Trump often speaks off the cuff, leaving unclear whether the list reflected a concrete procurement plan.
These were not newly invented tokens: XRP is associated with Ripple and its major SEC battle; Solana is an Ethereum competitor that hosts the Trump and Melania meme coins; Cardano’s creator participated in the early Ethereum movement. Their scale still does not establish strategic value for the United States.
The market effect was immediate: token prices rose after Trump’s post, though some gains receded. Actual government purchases would affect supply and demand, while the symbolic US seal on Cardano or another asset would itself benefit existing holders.
Much of crypto’s backlash came from within. Coinbase CEO Brian Armstrong said Bitcoin alone “would probably be the best option,” Tyler Winklevoss opposed adding other coins, and Nick Carter warned that a reserve would make Bitcoin “the plaything of the government, subject to Washington’s political cycles.”
8. Conflicts of interest divide an industry comfortable with enrichment
Ripple CEO Brad Garlinghouse spent time at Mar-a-Lago and donated heavily to Fairshake before XRP appeared on Trump’s list. Casey compared the optics to Mark Zuckerberg cultivating Trump and then seeing the government announce a strategic reserve of Facebook shares.
The Trump family’s World Liberty Financial directly profits the family and has accumulated a large collection of tokens, although its proposed DeFi application had not materialized. David said rising token values therefore offer an arguable financial benefit even without visibility into Eric Trump’s personal portfolio.
David’s memorable diagnosis: crypto participants “love enriching themselves” but are also paranoid about conspiracies against them. Some overlook conflicts when SEC lawsuits are dropped, then rediscover the principle when a rival token enters the reserve and their preferred asset does not.
Bitcoin’s original renegade, anti-government identity sharpens the contradiction. David said asking Washington to accumulate coins so their price rises is antithetical to that tradition; “most people in the industry don’t really care about those principles anymore, but some still do.”
9. Crypto’s quieter policy wins could connect the next crash to everyone else
A proposed stablecoin bill would create operating rules for US stablecoin companies and place a form of government approval on that branch of crypto. Industry critics see it as a possible “gateway drug” through which riskier assets could enter the mainstream economy.
The other priority is a market-structure bill shifting authority away from the SEC and toward the “much weaker, less aggressive” CFTC. Together, these bills would convert crypto’s present political goodwill into durable institutional rules.
Kevin worried that informed opposition has emptied out of Washington: Gary Gensler, who taught crypto at MIT, is gone, while earlier critics feel vindicated but exhausted after the crash and Sam Bankman-Fried’s imprisonment. “The crypto people sort of have the run of the town,” David said.
David’s systemic warning was that crypto, once walled off from the real economy, is becoming accessible through Bitcoin ETFs and potentially XRP or Solana ETFs. Kevin compared the knowledge imbalance to mortgage-backed securities and credit-default swaps before the financial crisis: insiders understand and profit from complexity before the public recognizes its exposure.
10. Vibe coding works best where users already understand the problem
Casey built Frog in a Bathrobe, a functioning if unexciting platform game, by typing a few instructions. Kevin’s hot-tub maintenance app demonstrated the other side: it sent daily messages and requested quarterly maintenance after one week, leaving its non-coder creator unsure how to debug it beyond saying “Fix it.”
Mike Lapchick’s speed reader displays pasted text one word at a time at roughly 400 words per minute, versus his cited 250 for “normies,” to help dyslexic and ADD users avoid jumbled text and lost focus. His product-management lesson transferred directly: better upfront definitions produced better results.
Anesthesiologist Lauren Buell began building ConsultCraft with Claude in November 2024 despite no coding experience beyond basic command-line familiarity. Six weeks of spare-time work produced a functional simulator; three additional months brought weekly cases, features and a transcript database for research.
ConsultCraft’s case scripts and feedback prepare residents for oral boards while refreshing experienced colleagues on uncommon operating-room cases. Kevin drew the broader lesson: a tool built for one person often serves peers with the same neglected need, especially where platforms have given accessibility issues short shrift.
11. Reusable automation offers leverage, but exploration is part of the return
Graphic designer Matt combined data from The Flavor Bible with a color-palette interface to create Flavor Finder, which generates compatible five-ingredient combinations and lets users lock or substitute ingredients. His less successful Rochambeau experiment—tic-tac-toe crossed with rock-paper-scissors—became a way to observe how an LLM thinks.
Photographer Zach used ChatGPT, Claude and DeepSeek to script the renaming, sorting and conversion of nearly 1,000 layered photo files into various JPEGs and TIFFs. The first attempt took about twice as long as manual work, and the Photoshop scripting was weak, but the reusable tool can save time on future jobs.
Kevin rejected pure efficiency as the only metric. The setup cost can pay off across repeated tasks, but the immediate attraction is also “discovery and exploration” and the pleasure of watching a computer build something; Casey’s version was simpler: “It is fun to learn, and it is fun to make things.”
Ashley’s request for “a bit of a wife” would gather obligations from email, WhatsApp, Insta chat, texts, Evites and Facebook Messenger, then produce tasks and a shared calendar. The obstacle is that these disparate services mostly do not interoperate, but Kevin pledged to attempt it.