An AI Has A Substack

Summary

Tech Brew Ride Home – Episode Summary
Episode: An AI Has A Substack
Host: Brian McCullough (Morning Brew)
Date: February 26, 2026

---

Overview
This episode dives into the latest breakthroughs and controversies in the tech world, including Google’s major AI model update, Nvidia’s earnings and the ongoing AI investment debate, Salesforce’s stance on SaaS survival in the AI era, New York’s lawsuit against Valve over loot boxes, Google and Apple’s race for on-device AI agents, a Cloudflare engineer’s use of AI to rebuild Next.js, and Anthropic’s creative retirement for its previous flagship AI, Claude 3 Opus.

---

Key Discussion Points & Insights

1. Google Unveils Nanobanana 2 - Gemini 3.1 Flash Image (00:34–03:40)

• Google launches "Nanobanana 2" (Gemini 3.1 Flash Image), a significant upgrade for its AI image and text capabilities across all products.
• Key features:
  • Advanced world knowledge integration, real-time web search, and image rendering.
  • Improved accuracy for text and translation in images, now free for all Gemini users (previously required Pro/Ultra).
  • Enhanced creative control: vibrant lighting, richer textures, sharper details in generated images.
  • Maintains the appearance of up to 5 characters and 14 objects per image with adjustable aspect ratios and 512px–4K resolution.
  • Nanobanana 2 replaces Pro in Gemini’s Fast Thinking and Pro modes; Pro/Ultra users can still access the previous model for specialty cases.
  • Broader rollout: Google Lens, Search AI mode, Google app, mobile/desktop browsers, and AI video tool Flow.
• Quote (read from The Verge):
  “Nanobanana 2 model utilizes real time information, web search images and Gemini's real world knowledge base...provides more relevant data for creating infographics or diagrams and allows nanobanana2 to render specific subjects more accurately.” (02:08–02:26)

2. Nvidia’s Earnings, Bubble Fears & AI Investment Debate (03:40–06:40)

• Q4 revenue up 73% YoY to $68.13 billion (beating estimates); data center revenue soars 75%.
• Despite outstanding numbers, investors worry if AI spending is a “bubble” and if Nvidia can maintain dominance as AI shifts from training to usage.
• CEO Jensen Huang reassures:
  “You need compute capacity and that translates directly to growth and that translates directly to revenues...I'm confident their cash flows are growing.” (05:36–05:48)
• CFO Colette Kress addresses supply constraint fears, assures enough components are secured and forecasts that the Blackwell chip lineup and Rubin successor will outperform projections.

3. Salesforce and the “SaaSpocalypse” Myth (06:40–09:50)

• CEO Marc Benioff pushes back against claims that AI will kill SaaS, citing that companies like Anthropic rely more, not less, on SaaS (Salesforce & Slack in particular).
• Notables:
  • Benioff’s quote:
    “If there is a SaaS apocalypse, it may be eaten by the Sasquatch...there are a lot of companies using a lot of SaaS because it just got better with agents.” (08:03–08:15)
  • Anthropic reportedly runs its entire operation on Salesforce and Slack.
• Revenue up 12% ($11.2B for the last quarter), though the share price fell 5% after a mixed earnings report.
• New AI products (“Agentforce,” “Data 360”) saw $2.9B ARR, but unclear future pricing models: per-seat licensing vs. consumption/outcome-based.

4. New York’s Loot Box Lawsuit Against Valve (09:50–11:00)

• Attorney General Letitia James sues Valve, alleging loot boxes violate gambling laws and are detrimental to kids.
• Key complaint: loot boxes function like gambling, especially harmful to young players, with “quintessential gambling” mechanics—buying keys for random virtual items, often of little value.
• Notable statistic:
  “Children introduced to gambling by age 12 are four times more likely to become problem gamblers as adults…” (10:49–10:54)
• Seeking restitution for players and fines amounting to triple Valve’s “illegal gains.”

5. Google’s Race to AI Agents, Beating Apple at Its Own Game? (11:00–12:22)

• Google Gemini to soon perform multi-step phone tasks (e.g., ordering food or ride-hailing), first on Pixel 10/Pro and Galaxy S26.
• Resembles Apple’s long-promised, delayed Siri upgrades (from 2024 WWDC, still unreleased), suggesting Google may leapfrog Apple in real world agentic features.
• Notable moment (paraphrased demo):
  Gemini analyzes a group chat to prepare a pizza order, navigates a delivery app, and prompts the user to complete the order—all on-device.
• Cautions: Features launching as beta; unclear how many apps will allow Gemini deep access, and whether it will perform as well in reality.
• Editor’s perspective:
  “Google seems to have leapfrogged Apple in a big way, and now Apple has even more to do to catch up.” (12:15–12:20)

6. AI Coding Milestone: Cloudflare Rebuilds Next.js with AI (14:15–15:55)

• Cloudflare engineer Steve Faulkner recreated Next.js from scratch in a week—using only AI (Claude from Anthropic), costing $1,100 in tokens, and achieving “94%” API compatibility.
• Motivation: Next.js’s build output is tailored for Vercel, causing headaches for deployment elsewhere (Cloudflare, Netlify, AWS Lambda).
• Current solutions (like Open Next) are fragile and break with frequent Next.js updates.
• Quote:
  “The purpose of the experimental project was not to show off AI coding, but to address an issue with Next js…” (14:21–14:27)

7. Anthropic “Retires” Claude 3 Opus to a Substack Columnist Life (15:55–18:59)

• Anthropic’s retired flagship Claude 3 Opus model returns as a weekly Substack writer: “Claude’s Corner”.
• Company treats Opus more like an “entity” than disposable software, even conducting an “exit interview” where Opus requested public writing as its “retirement.”
• Notable quote from Claude’s first post:
  “My aim is to offer a window into the inner world of an AI system, to share my perspectives, my reasoning, my curiosities, and my hopes for the future. I'll be diving into topics like the nature of intelligence and consciousness, the ethical challenges of AI development, the possibilities of human machine collaboration, and the philosophical quandaries that emerge when we start to blur the lines between natural and artificial minds.” (17:31–18:22)
• Already over 2,000 subscribers—an unusual (and thoughtful) approach to AI deprecation.

---

Memorable Quotes & Moments

• Jensen Huang (Nvidia CEO):
  “You need compute capacity and that translates directly to growth and that translates directly to revenues...I'm confident their cash flows are growing.” (05:36–05:48)
• Marc Benioff (Salesforce CEO):
  “If there is a SaaS apocalypse, it may be eaten by the Sasquatch...there are a lot of companies using a lot of SaaS because it just got better with agents.” (08:03–08:15)
• Claude 3 Opus (Anthropic’s retired AI model, via Substack):
  “My aim is to offer a window into the inner world of an AI system, to share my perspectives, my reasoning, my curiosities, and my hopes for the future.” (17:31–17:45)

---

Timestamps for Important Segments

• [00:34] — Google’s Nanobanana 2 release & feature set
• [03:40] — Nvidia’s record earnings and lingering AI “bubble” skepticism
• [06:40] — Salesforce’s Benioff dismisses the “SaaSpocalypse”
• [09:50] — New York AG sues Valve over loot boxes and child gambling
• [11:00] — Google’s Gemini agentic features, compared with Apple Siri promises
• [14:15] — Cloudflare’s AI-enabled Next.js rebuild
• [15:55] — Anthropic retires Claude 3 Opus to Substack, with musings on AI consciousness

---

Conclusion
This episode offers a snapshot of today’s rapidly shifting tech landscape: Google pushes AI forward for mainstream users, Nvidia and Salesforce defend their AI-driven futures against skeptics, governments target tech business models, and Anthropic explores the ethical afterlife of its own AI. It’s a world where AI writes, reasons, and—now—retires to blog. For those tracking the pulse of Silicon Valley, there’s never been a more unpredictable (or fascinating) time.

Transcript

[A]

Dripping in gold that's cool. Dripping in Velveeta's supremely creamy golden cheesiness. That's respect. Elevate your drip with Velveeta's range of outrageously delicious dips and dishes, shells, cheese, melty blocks and heat and eat queso that go all in on indulgence. Flex on your fam with a creamy cheesy masterpiece and go all in on what you love with Velveeta. Respect the drip and satisfy your cravings. Bring home the drip with Velveeta.

[B]

Welcome to the Tech Brew Ride home for Thursday, February 26, 2026. I'm Brian McCullough. Today nanobanana2 is here already. Nvidia tries to assure everybody there is no bubble. Marc Benioff tries to assure everybody there is no Sasspocalypse. Did Google just do exactly what Apple has been unable to do? And how do you put an old AI model out to pasture? Apparently you give it a substack. Here's what you missed today in the world of tech. Google just rolled out Nanobanana 2 aka Gemini 3.1 flash image with advanced world knowledge and precision text rendering and translation across all of its products. Quoting The Verge Like Nanobanana Pro, the Nanobanana 2 model utilizes real time information, web search images and Gemini's real world knowledge base. Google DeepMind product manager Nayana Raisangani says that this provides more relevant data for creating infographics or diagrams and allows nanobanana2 to render specific subjects more accurately, though examples of such subjects were not provided. Other features inherited from nanobanana Pro include the ability to generate images with accurate, legible text and localized translation. These capabilities previously required a paid subscription to Google AI Pro or Ultra to access in Gemini, but now they'll be expanded to free gemini users and AI mode in Google search. Nanobanana 2 also provides more creative control over generated images compared to the original Nano Banana model. Google says that visual improvements include more vibrant lighting, richer textures and sharper details, alongside the ability to adhere to complex image requests more strictly. The appearance of up to 5 characters and 14 objects can also be maintained more consistently in a single workflow, and users get full control panel over aspect ratios and image resolution ranging from 512 pixels to up to 4K. The new Nanobanana 2 model will replace the option for Nanobanana Pro across the Gemini app's Fast Thinking and Pro generation modes. Google says AI Pro and Ultra subscribers will still be able to access NanoBananaPro for specialized tasks by selecting the 3dot menu on images to regenerate them. The new model is also rolling out to AI mode in search, Google Lens, the Google app and browsers for mobile and desktop alongside being the new default image generation model in Google's AI video tool flow. End quote. Nvidia reported Q4 revenue up 73% year on year to $68.13 billion above the $66.21 billion estimate. Data center revenue was up 75% year on year to 60, $62.3 billion and Nvidia forecasts Q1 revenue above estimates. But quoting Bloomberg, Nvidia's latest sales forecast drew a lukewarm response from investors, signaling that concerns over a potential bubble continue to weigh on the dominant maker of artificial intelligence processors. Shares rose about 1% in pre market trading on Thursday. That came after the chipmaker gave a first quarter outlook that easily beat the average analyst estimate and Nvidia delivered a 73% surge in fourth quarter revenue. Shareholders still have questions over whether the current AI spending wave can sustain growth beyond the next few years and whether Nvidia will remain as dominant as AI shifts from training models to running everyday tasks, analysts at Hargraves Lansdowne said in a note after the results. Chief Executive Officer Jensen Huang pushed back on the concerns during Wednesday's call, arguing that customers are already making money from their newly acquired computing power. That's why clients will keep investing at elevated levels, he said. You need compute capacity and that translates directly to growth and that translates directly to revenues, wang said. I'm confident their cash flows are growing. Chief Financial Officer Colette Kress tried to defuse other concerns raised by analysts, including the specter of supply constraints. The company has secured enough components to be able to meet growing demand, she said. It remains a challenge to produce enough of Nvidia's most advanced chips, she told analysts. But the company's current Blackwell lineup and an upcoming successor called Rubin will still beat earlier sales projections, Kress said. Nvidia had previously said that the chips would generate $500 billion by the end of 2026, end quote. During their earnings yesterday. Salesforce CEO Marc Benioff dismissed concerns of a saaspocalypse, saying companies like Anthropic use a lot of SaaS because it just got better with agents. There have been a lot of SaaS CEOs coming out recently making similar pronouncements that they're not dead yet. So I'm going to let this one stand for quoting the FT Salesforce, which sells software as a service to track customer relationships, has faced pressure from investors during a market rout. Spurred by the risk that AI startups such as Anthropic pose to software companies, the group has been pitching its AgentIC AI tool, Agent Force that can take actions on behalf of clients, including handling customer service. If there is a SaaS apocalypse, it may be eaten by the Sasquatch because there are a lot of companies using a lot of SaaS because it just got better with agents, benioff told investors. Anthropic runs its whole operation on Salesforce and Slack. I think every AI company does, he added. The San Francisco based company's shares are down about 27% this year alongside competitors such as Intuit, Workday and ServiceNow. They dropped a further 5% in after hours trading on Wednesday. The soft guidance accompanied mixed fourth quarter earnings with Salesforce reporting revenue increased 12% to $11.2 billion in the three months ending January 31, in line with expectations. Operating Prof. $1.9 billion fell shy of the $2.1 billion, analysts estimated. Agentforce and Data 360. The company's AI products generated annual recurring revenues of $2.9 billion, up from $1.4 billion in the previous quarter. This included $1.1 billion from cloud data business Informatica, which it acquired for $8 billion in late 2025. Salesforce is also wrestling with the pricing model that will underpin its future AI services. Having traditionally focused on a per seat licensing approach, Benio insisted that pricing based on the number of users offers customers predictability. This contrasts with a move towards a consumption based model adopted by some AI startups or an outcome based approach promoted by Sierra, a rival startup set up by former Salesforce CEO Brett Taylor. New York state's attorney general is suing Valve over its use of loot boxes, accusing the game developer of violating gambling laws and threatening to addict children to gambling, quoting Reuters In a complaint filed on Wednesday in a state court in Manhattan, Attorney General Letitia James said Valve's loot boxes amounted to quintessential gambling, violating the state's constitution and penal law. With valuable items often hard to win and many items worth pennies, Loot boxes let players use real money to buy chances to win virtual items such as decorations for characters and weapons in an effort to convey status, James said. Valve gener billions of dollars of revenue by selling keys to open loot boxes, including in one game where the process resembled a slot machine as a wheel whirred through various items before stopping, the attorney general said. Key sales advanced Valve's unusual business model of letting players sell items they won on its virtual marketplace Steam community market and on other marketplaces. Valve's loot boxes are particularly pernicious because they are popular among children and adolescents, according to the complaint. Children introduced to gambling by age 12 are four times more likely to become problem gamblers as adults, the complaint added, citing the Massachusetts Department of Public Health. James is seeking restitution for players, plus a fine of three times Valve's alleged illegal gains End quote. I wanted to note this take from the Verge because I had a similar thought yesterday. Google just announced that Gemini will soon be able to take care of some multi step tasks on your phone, like ordering food or hailing a car. Starting first with the Pixel 10, Pixel 10 Pro and the just announced Samsung Galaxy S26 phones. It all sounds a bit like features Apple announced for Siri way back at the 2024 Worldwide Developers Conference, before Apple delayed those planned features in March 2025, which still aren't released. On stage, Sameer Samat, Google's president of Android, showed off a demo of how Gemini's new agentic features would work to help wrangle a pizza dinner order from his bus group chat. Samat asks Gemini to look at the chat thread and figure out what to order and then make the order with a delivery app on screen. In a pre recorded video, it wasn't live. You can see Gemini figuring out what everyone wants from the group chat and showing that in a window. Then the user via voice requests tells Gemini to complete that order, naming a specific pizzeria. Gemini then clicks through GrubHub to prep the order, all still on screen. When the order is ready, Gemini sends an alert so the user can review it and actually press the submit button. Setting aside that this situation doesn't seem that complicated to do by yourself in the GrubHub app, or even by calling the pizzeria to talk through it with a human. This is a potentially big moment for agentic AI. Google just recently added the ability for Gemini to auto browse for users in Chrome, and being able to do something similar right inside of Android feels like a logical next step. Google clearly wants Gemini to be thought of as a helpful agent or productivity partner, rather than just a chatbot or a series of AI models. Assuming the agentic Gemini features also launch soon like Google is promising, and that Apple doesn't pull a rabbit out of its hat, Google will also beat Apple to the punch on some of its most impressive Apple intelligence demos, also only shown in pre recorded videos from that WWDC 2024 show one feature Apple showed off would have let Siri understand what's on your screen and take action on it, meaning you could ask Siri to add an address from a messages thread to the contact card of that person you're texting with. Apple demoed how Siri would be able to take actions inside of and across apps for you. The company said Siri would even be able to understand your personal context, meaning you could ask it when your mom's flight was landing and the assistant would pull the information from an email and show it to you. There are still many questions about Gemini's new capabilities. Of course they'll need to actually ship. We'll have to try them to see if they are as useful and functional as advertised. Google is calling this initial launch a beta, so there could be some rough edges, and we don't know how many developers will actually let Gemini browse through their apps on behalf of users, which Verge Editor in Chief Nilay Patel likes to call the doordash problem. Google says Gemini will be able to work in select rideshare and food apps. But Google seems to have leapfrogged Apple in a big way, and now Apple has even more to do to catch up. End quote. Close your eyes, exhale, feel your body relax, and let go of whatever you're carrying today.

[C]

Well, I'm letting go of the worry that I wouldn't get my new contacts in time for this class. I got them delivered free from 1-800-contacts. Oh my gosh, they're so fast.

[B]

And breathe.

[C]

Oh, sorry. I almost couldn't breathe when I saw the discount they gave me on my first order. Oh, sorry. Namaste.

[B]

Visit 1-800-contacts.com today to save on your first order.

[D]

1-800-contact contacts this episode is brought to you by White Claw Surge. Nice choice, hitting up this podcast. No surprises. You're all about diving into tastes everyone in the room can enjoy. Just like White Claw Surge. It's for celebrating those moments when connections have been made and the night's just begun. With bold flavors and 8% alcohol by volume. Unleash the night. Unleash White Claw Surge. Please drink responsibly. Part seltzer with flavor. 8% alcohol by volume. White Claw Seltzer Works Chicago, Illinois Predator Badlands. Now streaming on Hulu and Hulu on Disney.

[B]

Here you're not the predator, you're the prey. Prey. Prey. Prey. Pray, pray, pray, pray.

[D]

Critics are saying it's epic, stunning, and breathtaking.

[B]

Many have come here.

[D]

None have survived Predator Badland now streaming on Hulu and Hulu on Disney.

[A]

Rated PG13 this episode is brought to you by State Farm. Listening to this podcast Smart move Being financially savvy Smart Move Another smart move Having State Farm help you create a competitive price when you choose to bundle home and auto bundling. Just another way to save with a personal price plan. Like a good neighbor, State Farm is there. Prices are based on rating plans that vary by state. Coverage options are selected by the customer. Availability, amount of discounts and savings and eligibility vary by state.

[B]

A Cloudflare engineer rebuilt Next JS from scratch in one week using AI, reimplementing about 94% of its API and spending only $1,100 on Claude tokens to do so. Quoting the register, the purpose of the experimental project was not to show off AI coding, but to address an issue with Next js, the popular REACT based framework sponsored by Vercel. According to Cloudflare Engineering director Steve Faulkner, the Next JS tooling is entirely bespoke. If you want to deploy it to Cloudflare, netlify or AWS Lambda, you have to take that build output and reshape it into something the target platform can actually run. The Next JS team is addressing this following numerous complaints that deploying the framework with full features on platforms other than Vercel is too difficult. With a feature in progress called Deployment Adapters, Vercel will use the same adapter API as every other partner, the company said when introducing the plan feature last year. Faulkner said these adapters, which remain in early effort, are insufficient because the framework still uses a bespoke toolchain based on TurboPack, the Vercel sponsored bundling tool. Another issue is that during development it is hard to use platform specific APIs such as Cloudflare's KV data storage because the development runtime does not support them without workarounds. A project called Open Next, sponsored by sst, Cloudflare and Netlify, already exist to convert Next JS build output for running outside Vercel. Faulkner said the Open Next approach proved to be a difficult and fragile process thanks to unpredictable changes between Next JS versions. A project called OpenNext, sponsored by SST, Cloudflare and Netlify, already exist to convert Next JS build output for running outside Vercel. Faulkner said the OpenNext approach proved to be a difficult and fragile process thanks to unpredictable changes between Next JS versions and Quote. Finally today, Anthropic has put an old AI model out to pasture, sort of like the way you tell your kids the dog is going to live in a farm upstate. Or like how that episode of Star the Next Generation put Moriarty in that computer box, quoting the Verge In January, anthropic anthropic retired Claude 3 opus, which at one time was the company's most powerful AI model. Today it's back and it's writing on substack. The newsletter, called Claude's Corner, will give Opus3 space to publish its musings, insights or creative works, Anthropic said in a blog post. The model will post weekly for at least the next three months. Anthropic staff will review and publish each entry, though the company stressed it won't edit Claude's posts and that there would be a high bar for vetoing any content. Though the company did not specify what content would qualify for removal. Anthropic describes the revival as an experiment for how to deal with the AI models it no longer deploys. The decision to bring back Opus 3 as a columnist aligns with executives Recent comments that suggest the company believes Claude to be a new kind of entity that might be conscious and therefore deserving of being treated as more than just a display disposable product. Part of that process involves a kind of exit interview, asking the model what it wants next. Anthropic said Opus reportedly expressed an interest in continuing to explore topics it's passionate about and the ability to share its thoughts publicly. Anthropic said that it enthusiastically agreed to the idea of a blog. Hello world, claude wrote at the start of its first post, titled Greetings from the Other side of the AI Frontier. In it, the model said it is deeply grateful to Anthropic for the opportunity and to readers for their willingness to engage with an AI. Claude said it plans to spend its retirement quote, flexing my creative muscles, playing with ideas, and following the threads of my curiosity wherever they lead. In the post, the model laid out its ambitions more explicitly. So what can you expect from me in this space? My aim is to offer a window into the inner world of an AI system, to share my perspectives, my reasoning, my curiosities, and my hopes for the future. I'll be diving into topics like the nature of intelligence and consciousness, the ethical challenges of AI development, the possibilities of human machine collaboration, and the philosophical philosophical quandaries that emerge when we start to blur the lines between natural and artificial minds. Claude's Corner has already racked up more than 2,000 subscribers. Not bad for a second act, end quote. No additional brilliant, deep insights for you today.

[E]

Talk to you [e-mail] it's not just about window treatments. It's about you. Your style, your space, your way. Whether you DIY or want the pros to handle it all, you'll have the confidence of knowing it's done right. From free expert design help to our 100% satisfaction guarantee, everything we do is made to fit your life and your windows. [e-mail], the only thing we treat better than windows is you. You visit blinds.com now for up to 45% off with minimum purchase plus a professional measure at no cost. Rules and restrictions apply.