Tech Brew Ride Home - Wed. 06/04 – State Of The AI Coding Landgrab

Episode Date: June 4, 2025

Windsurf accuses Anthropic of blocking it from AI models in a sign that competition in the AI coding space is fierce. But will the real battle happen when the likes of Microsoft and Google really go after the startups? NotebookLM is now sharable. Pump.fun is raising big money. And Nintendo didn’t send out any Switch 2 review units. What are they afraid of? Sponsors: 1Password.com/ride Links: Windsurf says Anthropic is limiting its direct access to Claude AI models (TechCrunch) AI startups revolutionize coding industry, leading to sky-high valuations (Reuters) Google’s NotebookLM now lets you share your notebook — and AI podcasts — publicly (The Verge) Pump.fun plans $1B token sale at $4B valuation: Sources (Blockworks) How Morgan Stanley Tackled One of Coding’s Toughest Problems (WSJ) It’s official: There are no Nintendo Switch 2 reviews. Here’s what that means for us, and you (VGC) Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Well, come to the Tech meme right home for Wednesday, June 4, 2025. I'm Brian McCullough today. WindSurf accuses Anthropic of blocking it from AI models in a sign that competition in the AI coding space is fierce. But will the real battle happen when the likes of Microsoft and Google really go after

the startups? Notebook LM is now shareable. Pump. Fun is raising big money, and Nintendo didn't send out any Switch to Review Units. What are they afraid of? Here's what you miss today in the world of tech. WindSurf says Anthropic is limiting its direct access to its AI models, including the newest Claude 4 family and the highly popular Claude 3 models, which is a big deal because Claude is popular with coders, I'm led to believe, quoting TechCrunch. Windsurf CEO Varroon Mohan said in a post on X Tuesday that Anthropic gave WinSurf little notice for this change, and the startup now has to find other third-party compute providers to run several of Anthropics' most popular AI models on its platform.

We have been very clear to Anthropic that this is not our desire. We wanted to pay them for the full capacity, said Mohan on X. We are disappointed by this decision and short notice, end quote. In a blog post, WinSurf said it has some capacity with third-party inference providers, but not enough, so this change may create short-term availability issues for WinServe users trying to access Claude. The decision comes just a few weeks after Anthropic seemed to pass over WinSurf during the launch of Claude 4, the company's new family of models. which offer industry-leading performance on software engineering tasks.

On launch day, Winsurf said it did not receive direct access from Anthropic to run Cloud4 on its platform and still hasn't. This forced the company to rely on a workaround that's more expensive and complicated for developers to access Cloud4. Meanwhile, other popular AI coding tools, including AnySphere's cursor, Cognitions Devon, and Microsoft's GitHub co-pilot seem to have direct access to Cloud4 models at launch. The AI-assisted coding sector, also known as Vibe coding, has heated up in recent months. OpenAI reportedly closed on a deal to acquire WindSurf in April. At the same time, Anthropic,

whose AI models are a favorite among developers, has invested more in its own AI coding applications. In February, Anthropic launched its own AI coding application Claude Code, and in May, the startup held its first code with Claude Developer Conference, end quote. Yeah, the competition with OpenAI and the whole excitement around the coding space makes that interesting, but I led with this story so that I could talk about something I've been thinking about. Sure, AI coding startups are hot now, but couldn't all these startups be potential roadkill? Not just OpenAI, but Google and Microsoft have popular AI coding tools and platforms. A source says Microsoft's GitHub co-pilot grew to more than $500 million in revenue last year.

Quoting Reuters, founders, founders of code gen startups and their investors believe they are in a land grab situation with a shrinking window to gain a critical mass of users and establish their AI coding tools as the industry standard. But because most are built on AI foundation models developed elsewhere, such as OpenAI, Anthropic, or Deepseek, their costs per query are also growing and none are yet profitable. They're also at risk of being disrupted by Google, Microsoft, and OpenAI, which all announce new code gen products in May, and Anthropic is also working on one as well, two sources familiar with the matter told Reuters. The rapid growth of these startups is coming despite competing on big tech's home turf. Microsoft's GitHub co-pilot launches

in 2021 and considered CodeGen's dominant player grew to over $500 million in revenue last year, according to a source familiar with the matter. Microsoft declined to comment on GitHub co-pilot's revenue. On Microsoft's earnings call in April, though, the company said the product has over 15 million users. Google's CEO also said in April that well over 30% of Google's code is now AI generated, and Amazon CEO, Andy Jassy, said last year that the company had saved the equivalent of 4,500 developer years by using AI. In May, Microsoft CEO Satchin Nadella said at a conference that approximately 20 to 30% of their code is now AI generated, the same month the company announced layoffs of 6,000 workers globally, with over 40% of those,

being software developers in Microsoft's home state of Washington. Some vibe coding platforms already boasts substantial annualized revenues. Cursor, with just 60 employees, went from zero to 100 million in recurring revenue by January 2025, less than two years since its launch. Winsurf, founded in 2021, launched its code, generation product in November 2024 and is already bringing in $50 million in annualized revenue according to a source familiar with the company. Both startups operate with negative gross margins, meaning they spend more than they make, according to four investor sources familiar with their

operations. The prices people are paying for coding assistants are going to get more expensive. Quinn Slack CEO at Coding Startup Source Graph told Reuters. Both cursor and Windsurf are led by recent MIT graduates in their 20s and exemplify the Gold Rush era of the AI startup scene. I haven't seen people working this hard since the first internet boom, said Martin Casato, a general partner at Andresen Horowitz, an investor in any sphere, the company behind Cursor. What's less clear is whether the dozen or so co-gen companies will be able to hang on to their customers as big tech moves in. In many cases, it's less about who's got the best technology. It's about who's going to make the best use of that technology and who's going to be able to sell their products better than others, said Scott Rainey, managing director at Redpoint Ventures,

whose firm invested in Sourcegraph and Poolside, a software development startup that's building its own AI foundation model. Most of the AI coding startups currently rely on the Claude AI model from Anthropic, which crossed $3 billion in annualized revenue in May, in part due to fees paid by CodeGen companies. But some startups are attempting to build their own models. In May, Winsurf announced its first in-house AI models that are optimized for software engineering in a bid to control the user experience.

Cursor has also hired a team of researchers to pre-train its own large frontier level models, which could enable the company to not have to pay foundation model companies so much money, according to two sources familiar with the matter, end quote. Google's notebook LM now lets users share notebooks publicly. People can interact with a public notebook by asking questions or exploring the generated content, quoting the verge. Though viewers can't edit what's in your notebook, they can still use it to ask questions and interact with AI-generated content like audio overviews, briefings, and FAQs.

First launched as an experiment in 2023, Notebook LM has become a breakout hit for Google. The app is designed to help you understand material from a variety of sources such as notes, documents, presentation slides, and even YouTube videos. It can provide AI-generated summaries of the content, generate AI podcast-style discussions, chat with you about the material and more. Google launched a mobile Notebook LM app last month. The steps to making your notebook available publicly are pretty soon. similar to the way you share something in Google Drive, docks, sheets, and slides. You just select

the share button in the top right corner of the notebook and then change the access to anyone with a link. From there, hit the copy link button and then paste the notebook link into a text, email, or even on social media if you want more people to interact with the information. Google also lets you share your notebooks with others by entering their email addresses. Unlike with public link sharing, you can give individual users the ability to edit your notebook. You can share audio overviews from within the Gemini app as well, end quote. Sort of an interesting raise, sort of an IPO in a way. Sources say pump.

Fund plans to raise $1 billion via a token sale at a $4 billion valuation. Pump. Fund has generated over $700 million in cumulative revenue since its launch in early 2024, quoting Blockworks. The token will be sold to both public and private investors. Sources added, Blockworks could not immediately confirm when the token is set to go live or whether it will be issued on the Pump. Dot Fund platform. However, a post on X tease that it could come in the next two weeks.

The potential $4 billion token valuation would mint Pump. Dot Fun as crypto's latest unicorn startup after a year in which it helped ignite a meme coin frenzy. Pump.combe launchpad that allows anyone to spin up their own Solana token instantly and for free was a breakout hit with crypto users after launching in early 2024. The app has already generated over $700 million in cumulative revenue, according to Blockworks research data. Users on the platform have minted nearly 11 million new tokens with a current cumulative market cap of roughly $4.5 billion. As competition heats up among token launchpads, Pump.comfunk fun has made some updates to its core business, most notably when it launched

an AMM that ended its unofficial partnership with radium. Pump.combe. Fun has also recently launched a mobile app and re-released a live streaming feature it had temporarily suspended after content moderation complaints late last year, end quote. The journal has a piece taking a look at how Morgan Stanley is using AI to tackle a major challenge common to a lot of big enterprises, converting outdated legacy code into modern languages. We've been talking about this a lot recently. In January, Morgan Stanley launched devgen.a.i, a tool built in-house using OpenAI's GBT models. Devgen.a.i translates legacy code like Cobol into plain English specifications, making it easier for developers to rewrite. According to Mike Pizzi, the firm's global head of technology and operations, the tool has already reviewed 9 million lines of code and saved developers 280,000 hours. While modernizing old software is crucial for security and innovation, it's long been a tough

problem for AI tools until now, apparently. Quote, the latest AI coding tools are excellent at writing new modern code, but they don't necessarily have as much expertise in less popular or older programming languages or in those customized for a given company, Pizzi said. It's an area many tech companies are working on, but at the moment their offerings don't have the flexibility enterprises need, he added. That's why Morgan Stanley opted not to wait. We found that building it ourselves gave us certain capabilities that we're not really seeing in some of the commercial products, Pizzi said. The off-the-shelf tools might yet evolve to deliver those

capabilities, he said, but we saw the opportunity to get the jump early. Morgan Stanley, he said, was able to train the tool on its own codebase, including languages that are no longer or never were in widespread use. Now the company's roughly 15,000 developers, around the world, can use it for a range of tasks including translating legacy code into plain English specs, isolating sections of existing code for regulatory inquiries, and other asks, and even fully translating smaller sections of legacy code into modern code. But when it comes to full translation, that technology still has some room to mature, he said. It can technically rewrite code from an old language like Pearl into a new one like Python, but it wouldn't necessarily know

how to write it as efficient code that takes advantage of all Python's capabilities, he said. and that's one big reason humans are staying in the loop, he said. Where the tool really shines is in translating legacy code into English specs, basically a map of what the code does, according to Pizzy. It's something an ever-dwindling pool of developers trained on super old or specific coding languages knows how to do. With those specs, any developer can then write the old code as new code in a modern programming language, he said.

Pizzi said, you're not going to see fewer heads in software engineering, just more code, including more AI apps that will help Morgan Stanley deliver on its business goals. Currently, the company has hundreds of AI use cases in production aimed at growing the business, automating workflows, and doing it more efficiently. But none of that is possible without a modern standardized, well-thought-out architecture, Pizzi said. You're always modernizing in tech, he said, today, with AI, this becomes even more important, end quote. Finally, today, this is odd. Nintendo says it didn't send any switch to review units to outlets as, quote, important features and updates will only be available via an update on its June

5th launch. Quoting Video Games Chronicle, it's official for the first time in more than 20 years, there will be no pre-release reviews for a new Nintendo console. According to a company spokesperson, Nintendo decided not to send Switch to Review units to press ahead of launch because important features and updates will only be available via a system update on the day of the console's official release June 5th. It did not say what these updates entail or explain why the media wasn't offered the opportunity to review its launch games like Mario Kart World and Switch to Welcome Tour at in-person

events instead, like it did with previous console launches such as the Wii. Neither did it explain why critics can't be trusted to cover products with knowledge that additional functionality will be patched later as they're frequently used to. How and when game creators decide to distribute their review products is, of course, their prerogative, and the media has no right to early samples. Game coverage has also changed drastically over the past decade, with more priority given to post-release coverage of today's increasingly complex video games. But a lack of coverage at-or-before-release is bad for you are readers. When Switch 2 arrives, there will have been no reviews of its software, no analysis of its hardware performance or battery life or any testing of its

key features such as Game Chat or how backwards-compatible Switch-1 games perform. No hidden gems will be unearthed and no potential dodgy ports highlighted before you have a chance of spending your hard-earned money. And when reviews do arrive, well after launch day, many critics will have been incentivized to rush through their games as quickly as possible with Google and social media rewarding speed over quality. But it's not just the media that misses out. More than 20 Switch 2 launch games from third parties have now effectively been prevented from generating any day one review coverage for their titles. Many publishers have reached out to sites like VGC asking for clarification around when the press would receive review units, and some even said they wished

they'd chosen to release their games later after it became clear they would not be able to secure authentic editorial coverage at all. Perhaps it really was a technical nightmare for Nintendo to provide pre-launch review access to press. We're already seeing, for example, that those who have somehow gained access to the console pre-launch cannot use them at all until a future system update. And there's certainly no precedent of Nintendo being adversarial with press reviews. In my experience, in fact, it's historically been one of the most critic-friendly companies, often supplying software many weeks in advance. And I don't personally believe the decision to hold back reviews comes from a lack of confidence in its launch lineup either. My recent time playing Mario Kart

World and Switch to Welcome Tour at Nintendo's UK office was great fun and had me excited to spend more time with the games. But consoles have launched with day one patches for over a decade, and the media are well used to covering unfinished products, so in my opinion, this decision is more likely related to Nintendo's paranoia around leaks, as it is with any missing functionality before launch day. Those leaked videos of early consoles that can't run games certainly strengthen that hypothesis, end quote. Yeah, this sort of feels like when movie studios won't let reviewers see the movie before opening day, it smells fishy. I'm going to make a tiny point that will be unimportant for most of you, but controversial to a very specific subset of this audience. I think that new Mario Kart game looks bad. It looks like they might have broken the game's classic formula.

Classic Mario Kart is about chaos for sure, bananas, red shells, bombs, and the like. but it's still a racing game at heart. You're still racing on actual tracks, actual courses, where the driving environment has actually been thought out. The videos of the gameplay of the new Mario Kart don't really show courses. I mean, they're kind of courses, but they're wide open,

and you just kind of go forward and just hope to survive the chaos erupting all around you. It's like they de-emphasize the actual driving, the actual racing part of the game. Instead of driving amid chaos, it's now just all chaos. I don't know. I haven't played it yet, obviously, but I have watched an hour or two of gameplay videos on YouTube, and it doesn't look fun to me. I hope I'm wrong. And I say this as someone who has played and plays Mario Kart once or twice a week with my college buddies have done so since the switch came out, what was it, seven years ago. So I'm worried. Talk to you tomorrow.