The AI Daily Brief: Artificial Intelligence News and Analysis - 5 Debates Shaping AI

Episode Date: September 9, 2025

AI is at the center of five big debates: is massive AI spending fueling real growth or just a bubble, will entry-level jobs vanish, does AI truly boost productivity, is vibe coding overhyped, and should we accelerate or slow down with open-source and policy? These questions capture the mix of hype, hope, and anxiety shaping AI today.Brought to you by:KPMG – Discover how AI is transforming possibility into reality. Tune into the new KPMG 'You Can with AI' podcast and unlock insights that will inform smarter decisions inside your enterprise. Listen now and start shaping your future with every episode. ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://www.kpmg.us/AIpodcasts⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Blitzy.com - Go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://blitzy.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ to build enterprise software in days, not months Robots & Pencils - Cloud-native AI solutions that power results ⁠⁠⁠⁠https://robotsandpencils.com/⁠⁠⁠⁠Vanta - Simplify compliance - ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://vanta.com/nlw⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠The Agent Readiness Audit from Superintelligent - Go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://besuper.ai/ ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Interested in sponsoring the show? [e-mail]

Today on the AI Daily Brief, five debates shaping AI. Before that in the headlines, Anthropic announces a major copyright settlement, but who is it actually good for? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, super intelligent, robots and pencils and Vanta. To get an ad-free version of the show, go to patreon.com. slash AI Daily Brief, and to get information about sponsoring the show, shoot us a note at sponsors at AIdailybreep.a.a. Welcome back to the AI Daily Brief Headlines edition, all the daily AI news you need

in around five minutes. We kick off today with a story from last week, a big update around one of the copyright cases swirling around the AI industry. Anthropic has agreed to pay $1.5 billion to authors in a landmark AI copyright settlement, making it the largest settlement in the history of U.S. copyright law. Now, back in June, a judge ruled that Anthropical, Anthropic's use of around 500,000 books in training data was fair use. However, the judge found that Anthropic would still be liable for copyright infringement for pirating the books rather than purchasing them. That claim comes with a hefty per-infringement penalty,

so Anthropic couldn't really risk going to trial and ending up with a ruling in the tens of billions. A settlement fund will be used to make cash payments to impacted authors and to cover legal costs. Now, as you might imagine, the debate very quickly became, is this good for Anthropic, is this good for authors, is this good for both, is this good for no one? TechCrunch took the view, and it's right there in the title that the settlement, quote, sucks for writers. They noted that each of the half a million authors affected would only receive $3,000 settlement checks, and frankly, that assumes that legal costs won't reduce the number, so that's probably even a little bit generous. By way of comparison in February, Microsoft offered authors working for Harper Collins,

$5,000 per book to be included in training data, split evenly between publishers and authors. It's also noteworthy that under this ruling, AI companies can train on any data they obtain legally, throwing into question the need to license books rather than just buying them off the shelf. Simon Willisson tweeted, Am I the only person who thinks this $1.5 billion anthropic book settlement counts as a win for Anthropic? Later, he said, to clarify, it appears it is legal, at least in the USA. To buy a used copy of a physical book, chop the spine off, scan the pages, discard the paper

copy, and then train on the scan data. The transformation from paper to scan is fair use. Fairly trained CEO Ed Newton Rex said that part is a win for them agreed, Although pointed out the judges in other instances of AI lawsuits weren't necessarily taking the same approach and said that, quote, the $1.5 billion to rights holders, though, is a win for rights holders, particularly as many other companies are suspected to have used pirated content. Maybe the most coherent, if nuanced take came from Aaron Moss of Copyright Lately, who wrote, Anthropics' $1.5 billion copyright settlement is simultaneously groundbreaking and trivial,

a paradox that reveals how AI has fundamentally altered the economics of copyright infringement. Now, speaking of Anthropic, shifting off a big macro-type story to a very positive feature update, the company has rolled out a version of long-term memory to pro users. Previously only available through Mac's team and enterprise plans, the feature allows Claude to reference previous conversations. Users can prompt Claude to go back and look at previous chats to recover context. Anthropic demonstrated the feature using the prompt, I'm back from vacation, what were we working on last week?

Claude then searches through conversations from the previous week and pulls out the topics the user might want to load into context. They wrote, the feature does not invisibly bring in context from past chats. When turned on, you can specifically ask Claude to search past chats or Claude will infer based on context. You will always see a tool call when this happens. Now, take this alongside OpenAI's introduction of branching conversations last week, and it is very clear that there is a conscious effort among the big labs to give users more control over context within their sessions. I'm telling you, it's starting already, but 2026 is absolutely the year of context management, engineering, and orchestration.

Moving over to the infrastructure part of the industry, OpenAI is set to begin mass production of their own chips next year. Last Thursday, during their earnings call, Broadcom's CEO said that they had secured a $10 billion order for custom AI chips from an unnamed customer. The Financial Times now reports that the mystery customer is OpenAI, with multiple sources telling the FT that the custom silicon would ship next year. The report states that OpenAI is planning to deploy the chips internally rather than sell them to external customers. Now, 10 billion is enough to buy more more than 160,000 Nvidia H-100s, so OpenAI clearly has some very large-scale plans here. Reds Mahidar, the AI chip wars are on. Google, TPU, AWS, and Anthropic Traneum, and OpenAI

Plus Broadcom custom A6, everyone's pushing to reduce dependence on Nvidia. Who wins will come down to software stacks, fab capacity, and power efficiency. Now, speaking of OpenAI and infrastructure, the company is also projecting vastly more spending than previously anticipated. The information reports, that investors were recently informed that OpenAI could spend $115 billion by the end of 2029. That's more than three times as much as the previous forecast of $35 billion, with OpenAI increasing their estimates by $80 billion. The report states that enhanced spending is due to OpenAI developing their own chips and data centers, as well as rising computer costs as the company scales.

OpenAI now expects to spend $150 billion on compute between 2025 and 2030. Now, of course, all of this is contributing to some of the big debates that we're going to be talking about in the main episode, but OpenAI is not the only company upping their estimates around how much they're going to spend on infrastructure. According to comments from meta CEO Mark Zuckerberg and a dinner with President Donald Trump and other tech execs last week, Zuck said that META plans to spend something like at least 600 billion, his terms, on data centers and other U.S.-based infrastructure. That seems to be an increase from estimates back in July, where META suggested that its 2025

capital expenditures would be between 66 and 72 billion, which was itself up 68 percent from CAPX in 2024. Lastly, today, a pair of very similar feature updates from some image generators. At the end of last week, Mid Journey announced its first Style Explorer. They tweeted, go-to-explore on our website and press Styles. Click Try Style to quickly test them with whatever you have in your prompt bar. Fuzzy search works, too. Type photo or anime in the search bar to narrow styles to just those domains.

Now, this is basically a U.X upgrade to allow people to get a generation in a specific stylistic zone, without having to redescribe it every time, and with potentially more precision than their ability to articulate it via text. Now, on the same day that Mid Journey announced their style explorer, Ideogram also announced Ideogram styles. They launched with a huge array of styles, promising, quote, sophisticated consistent aesthetics, and even matching stylized typography. Indeed, they wrote, we worked hard to support text rendering in ideogram styles, bring

creative stylized typography to your designs in seconds. You can also upload a set of reference images to create custom styles that can be used in generations as well. Now, as someone who uses ideogram as their daily driver design tool, and who has a very specific set of a couple different bundles of aesthetics, this is a potentially super powerful and time-saving update and one that I'm excited to check out. For now that that's going to do it for today's AI Daily Brief Headlines edition. Next up, the main episode. What if AI wasn't just a buzzword, but a business imperative? On You Can with AI, we take you inside the boardrooms and strategy sessions of the world's most forward-thinking enterprises. Hosted by me, Nathaniel Wittamore, and powered by KPMG, this seven-part series delivers real-world insights from leaders who are scaling AI with purpose,

from aligning culture and leadership to building trust, data readiness, and deploying AI agents. Whether you're a C-suite executive, strategist, or innovator, this podcast is your front-row seat to the future of enterprise AI. So go check it out at www.kpmg.org.us slash AI podcasts or search you can with AI on Spotify, Apple Podcasts, or wherever you get your podcasts. If you are a regular listener, you will have heard about Super Intelligence Agent Readiness Audits at this point. But I wanted to tell you today about the full suite of Agent Readiness products that go beyond just the initial readiness report. Over the last six months, Super Intelligence has built out an entire Agent Planning suite. We help you move from discovery to planning to implementation. After you've completed your agent readiness audits,

we help you double-click on your most important use cases with what we call our use case planning reports. These reports are going to help you understand what sort of technical preparation you need to do to be ready for a use case, what challenges you might face in implementation, and whether you should be thinking about building, buying, partnering, or some combination. After that, you can even get a spec document

in what we call our technical blueprint that gives either your developers or the developers of the partner you work with what they need to build exactly the agent that you're looking for. If you want to learn more about superintelligence agent planning suite, we've built a custom GPT to answer your questions. Just go to bit.ly slash super super agent. That's bit.l.ly slash super super agent, all one word. And if you have any questions, the agent can even help you book an appointment with our team.

AI isn't a one-off project. It's a partnership that has to evolve as the technology does. Robots and Pencils work side by side with clients to bring practical AI into every phase, automation, personalization, decision support, and optimization. They prove what works through applied experimentation and build systems that amplify human potential. As an AWS-certified partner with Global Delivery Centers, robots and pencils combines reach with high-touch service, where others hand off, they stay engaged, because partnership isn't a project plan. It's a commitment.

As AI advances, so will their solutions. That's long-term value. Progress starts with the right partner. Start with robots and pencils at robots and pencils.com. As a founder, you're moving fast towards product market fit, your next round, or your first big enterprise deal. But with AI accelerating how quickly startups build and ship, security expectations are higher earlier than ever. Getting security and compliance right can unlock growth or stall it if you wait too long. With deep integrations and automated workflows built for fast-moving teams, Vanta gets you audit-ready fast and keeps you secure with continuous monitoring as your models, infra, and customers evolve. Fast-growing customers like Langchain, writer and cursor, trusted Vanta to build a scalable

foundation from the start. And look, as someone who lives in the world of enterprise procurement, I love how Vanta makes it easy to get compliance right. The last thing you need when you're trying to win that big deal is to have it scuttled by something that Vanta has solved for over 10,000 companies. Go to Vanta.com slash NLW to save 1,000. today through the Vanta for Startups program and join over 10,000 ambitious companies already scaling with Vanta. That's VANTA.com slash NLW to save $1,000 for a limited time. Welcome back to the AI Daily Brief. Today we are talking about five debates that are shaping AI. Right now we are in an interesting and frankly pretty anxious feeling time period when it comes

to artificial intelligence. There are questions of its relationship with the economy. There is job-related anxiety, there are questions of productivity and how good it actually is. We're seeing a ton of skepticism people wondering if we're entering the trough of disillusionment, that famous period in the Gardner hype cycle, when the hype of a new technology finally dissipates. So where are things really, and what is the state of the conversation? Now, a quick shout-out. If you're not reading the neuron yet, you should. You can get it at the neurondaily.com. This post was inspired by their morning newsletter today, which was also about five major AI debates, although I went in a slightly different direction than some of them. I thought it was a great way to tee off some conversations,

even though exactly what they talked about is a little bit different than what I'm going to get into here. First and foremost, one of the two that I think is causing the most anxiety is the question of to what extent AI is propping up the economy versus just in a bubble versus both of those at the same time. And there is certainly a sense that AI is very much at the center of whatever strength there actually is in the economy. That's both from a stock perspective as well as a real spending perspective, which is what's causing some people even more concern. Now, if you've been listening to this show for a while, you'll know that basically ever since the rate hiking cycle began in 2022, it has really been AI enthusiasm versus the world when it comes to stock market performance. In other words,

in my estimation, a good chunk of the bull market that we've experienced in stocks is not broad-based economic success, but instead simply incredible enthusiasm around AI, that is of course backed by the incredible amounts of spending from the big tech companies building out infrastructure. And that was the point of this piece in the New York Times from a couple of weeks ago called the AI spending frenzy is propping up the real economy too. They write, it's no secret by now that optimism around the windfall that artificial intelligence may generate is pumping up the stock market. But in recent months, it's become clear that AI spending is lifting the real economy too. And they write that that's not because of massive enterprise adoption, but because of, as they put it,

the sheer amount of investment in data centers, semiconductor factories, and power supply, needed to build the computing power that AI demands is creating enough business activity to brighten readings on the entire domestic economy. In 2025, UBS estimates that companies will spend $375 billion on AI infrastructure, a number that will rise to a half trillion next year. Commerce Department suggested that investment in software and computer equipment, not counting data center buildings, accounted for a quarter of all. all economic growth in the past quarter. You might have seen that chart that's been flowing around

that shows that the amount of spending on traditional offices has converged with the amount of spending on data centers, and that data centers are in fact forecast to exceed spending on traditional offices over the course of this year. While all of this is good news, in the sense that this sort of infrastructure investment has a variety of positive residual effects in the short and medium term, there is a lurking question expressed in the section of the times piece, how long can the spending last. So this is the setup to this part of the debate, and there are so many places where you can see some version of this. Earlier in August, Axios called this the AI superstimulant. AI in its blood and oxygen, chips, data, and energy are producing an economic superstimulant strong

enough to prop up the entire industry. They write, a wild cycle is unfolding. The biggest companies in history are spending a stunning amount of money to fuel the AI revolution, driving demand for more chips, more energy, and more capital. The dynamic is basically powerful enough to cover for an economy that otherwise looks like it's weakening by the day. And this is, of course, the other side of the story and why this debate exists. It's one thing to ask whether AI spending is over-exuberant. But the context of it potentially covering up an economy that otherwise looks like it's weakening, to use Axios' terminology, is where things for some people start to get dicey. Business Insider writes, the AI spending boom is boosting US GDP and potentially hiding looming problems. They point

to research from Pantheon macroeconomics who found that without AI spending, the U.S. economy would have grown at less than 1%, a quote, sign that tech companies are propping up a not-so-great economy. All of this has led to a new type of discourse asking, how bad would it be if this was a bubble and it actually burst? That was the subject of a recent Atlantic piece called Just How Bad Would an AI Bubble be? Now, that piece gets into an argument which we're going to talk about in just a minute, which you can see expressed here in the sub-headline, the entire U.S. economy is being propped up by the promise of productivity gains that seem very far from materializing. The core of that piece is the meter study about software developers that we have talked about extensively on this show, and we will recap in just a minute

here. Now, without getting into the bubble or not debate, if you are just looking at narratives and discussions, this is getting louder right now. The conversation is on an upswing. A piece in courts from just this morning reads, the AI bubble fears are getting worse. Wall Street doesn't want to talk about it. Wall Street finds itself caught between optimism and anxiety, deploying different vocabularies depending on which side of the ledger it's addressing. In fact, the bubble conversation is now getting so large that you are once again seeing a counter-narrative. It's not just humble podcasters and entrepreneurs like me, who are taking the other side of that bet, but prominent tech investors like Ross Gerber, who is quoted in Business Insider today

saying that AI is nothing like the dot-com bubble. There's CIO magazine publishing why the AI bubble is good for business. But again, whatever position you take on this, it is one of the conversations that is absolutely shaping the discourse right now. And I think, frankly, in many ways, capturing a larger economically anxious zeitgeist that doesn't even be a very important. necessarily have to do anything with AI, but just where AI is the main recipient. Now, part of why AI is the main recipient of this economic anxiety is the whole jobs dimension. And the interesting debate here is that simultaneously, you have a growing chorus of people who are pointed to evidence that AI is taking our jobs, but then at the same time, and often from some of the same people,

a discourse about it being overhyped and not actually improving things. Now, on the job side, we've been talking a lot about that Stanford study, suggesting that the young employees who are seeing 13% fewer job opportunities than their peers who are not in areas that are as exposed to AI disruption are the canaries in the coal mine to use their terminology for the entire industry. Remember the way that that study worked was that they divided all the jobs into five categories of how exposed to AI disruption they were and found that those in the fifth category the most exposed saw a 13% relative decline for the youngest workers in the US. Recently another paper looked at resume and job posting data on around 285,000

in U.S. firms over the last four years and found that while senior roles didn't seem to be affected, Gen. AI did seem to be reducing the number of junior roles. Now, as we discussed, I think, on Friday, there is a difference between acknowledging a decrease in hiring in general for young people in junior roles and a clear determination that it's because of artificial intelligence. As Professor Ethan Mollock put it, the fact that junior hiring in AI intensive fields has slowed down somewhat in the U.S. seems pretty solid. The evidence linking it to AI is not yet established. We've seen a a couple solid attempts that suggest a connection, but it's really hard to tell for sure, given the data. Stating the obvious but important thing, he says, a lot of macroeconomic things

are happening at once and a short time frame making teasing apart the details challenging. Still again, from a narrative perspective, this is getting louder and louder. And what's more, there are some prominent voices in the industry that are doubling down on their calls that this is going to be an issue. Earlier in the year, Anthropic CEO Dario Amadehadeh, said that he thought that about half of entry-level white-collar jobs could be wiped out by AI, and in a recent podcast appearance, he doubled down on that prediction. Some are extending this even farther. With CNBC publishing a piece, AI is not just ending entry-level jobs. It's the end of the career ladder as we know it. Take that, though, and compare it to this other discourse about AI not being all that good.

From theconversation.com, does AI actually boost productivity? The evidence is murky. And then, of course, we have that much Ballyhoo'd study from meter, summed up in this Reuters piece, AI slows down some experience software developer study finds. I remember seeing some comedian years and years ago in a very different political climate, mind you, who had a joke that ran something along the lines of, are the immigrants lazy or are they taking our jobs? Because it can't be both guys. And I sort of a little bit feel like that's the nature of this discourse right now around AI jobs and productivity. On the one hand, we have all of this pointing to evidence that AI is undoing.

economic opportunity for people. But on the other hand, this other discourse suggesting that productivity gains aren't all that they're cracked up to be. And it's kind of like, is AI not actually all that good or is it taking our jobs? Because it's really hard to see how it could be both. Now, of course, if you take a step away, you can find a study right now that supports any position you want to take on this. Goldman Sachs investment research recently found, for example, that the cumulative evidence that they looked at across academic studies and company anecdotes put labor productivity boost around 27 to 31% for AI. And the point is that what I think is going on more than anything else is this broader pattern of economic anxiety. I think that that economic anxiety is

bleeding into everything, including the AI discourse, making it all very, very murky. Now, one area where there is some specific debate around that productivity question is how good vibe coding really is. Now, I've made the argument that we are on a mainstreaming trajectory where vibe coding is increasingly just agentic coding, but along the lines of this broader anxiety that we keep discussing, there is again a growing discourse that is trying to claim that VibeCoding 2 is in fact overhyped, and not all it's cracked up to be. Now, this isn't all that surprising to me that VibeCoding specifically would see some amount of this skepticism. Basically, I think that agentic coding is quite clearly the breakout AI use case, or agentic use case at least. For 2025, it is everywhere from

consumer to enterprises. Every survey that people do shows that it is the thing that is happening the most inside organizations, but it's also something that has been very, very quickly evolving. This time last year, vibe coding wasn't a thing. It wasn't even a term that was coined until February of this year, and it really took the reasoning models, and even more specifically, the launch of Claude 3.5 Sonnet, to really enter this new agentic coding era. Some of the skepticism have come from people that you might generously call builders themselves, although you could quibble with that in this case. In August, for example,

Spack King Chimath Pala Hapitia tweeted, it's unpopular to say, but it's true. You can't vibe anything useful right now. You can barely vibe a working product, and there are still virtually zero examples of anything even moderately useful or successful that was vived, especially considering how much money has been spent so far

on LLM calls to try and do so. This won't be true forever, but is true today. As with other tech cycles, a trough of disillusionment may soon come as folks get frustrated and give up. So, is this just Chimoth trying to sound clever and ahead of the narrative, or is there something more real here? More recently, people have been revisiting a prediction from Anthropic CEO Dario Amadeh, where he suggested that

around this time, AI could be writing as much as 90% of all code. The information last week wrote a piece called Why Anthropics coding prediction hasn't panned out. In that same piece, they actually asked Claude, the chatbot of course, of Anthropic, to grade the prediction with Claude giving it an F. The information said that Claude cited various industry surveys, estimating that about 40% of all code is AI generated, but only 20 to 25% of code that actually makes it into production is AI generated. Now, there was nuance even in this information piece. For example, Rahul, the CEO of Julius, said that he, quote, wouldn't be surprised if something like 90% of front-end code is being written or initially drafted by AI,

but he thinks there's probably a gap in terms of what makes it to production, and he thinks that it's probably lower when it comes to back-end code. Now, this conversation to me was just absolutely baddie. It was quickly seized upon by the AI skeptics, as another example of why we were getting out over our skis when it came to AI hype, yet I'm sitting here looking at the idea that inside of a year or so, somewhere between 20 and 40% of code is being written by a thing that couldn't write code around this time last year, and we're somehow viewing that as things being overhyped?

It just seems insane to me. Ethan Malik wrote, The funny thing about the prediction that AI would be writing 90% of all code by now is that the prediction's failure distracts from the fact that AI adoption and code writing is actually extremely high. Now, he points to over 30% according to one measure all the way back in December 2024. You also have folks like Coinbase CEO Brian Armstrong, claiming that around 40% of daily code written at that company is currently AI generated

and that he wants to get it to 50% by October. That's up from under 20% back in April, meaning that in six months, the percentage of code that's written by AI at Coinbase has doubled. Flipping back on the other side, you have people who are pointing to Google Trends, suggesting that searches for lovable, Claude, Code, Replit, Curser, are all down from their peak at the beginning of August. But then to me, maybe the more interesting story is what evidence we have that vibe coding

is producing things that people are actually using. Olivia Moore from A16Z writes, the biggest untold story around vibe coding is the explosion of the accompanying stack. One example is Superbase, which many vibe coding apps recommend as a backend database. It's been around for five years, but traffic has skyrocketed with the rise of lovable reflet, etc.

We also recently discussed, Lovable CEO, Antonio Sika, suggesting that the way that the company wants to measure its success from now on is how many visits are going to apps built with Lovable rather than just how many people are using the tool. Last week, Antoine tweeted, Loveable Apps have officially been visited 100 million times in the last two months. Would that milestone we're changing Lovable Success metric be proportional to the traffic users' products get? Because if our users succeed, we succeed. Finally, we come back to Dario, who just did another interview last week with a BBC reporter,

who doubled down and said that 90% of code ad Anthropic is written or at least suggested by AI. Prins on Twitter writes, says 90% of code at Anthropic is written or suggested by AI, what he means is, and if that's not the case for your company yet, that's a skill issue. We've gone through three debates so far in some amount of detail. We'll talk about the last two a little bit more quickly. The fourth is one that I'm kind of bundling together, which is open source and soft power versus the push to stop China, which is more broadly, I think, a part of an acceleration versus deceleration meta debate that has been with us ever since the beginning of chat GPT. On the China front, of course, a lot of 2025

has been racing to readjust our priors in the wake of deep-seek coming out and totally changing how far behind or rather how not far behind China is when it comes to their models. When the White House dropped America's AI action plan this summer, one of the things that was interesting was a shift on open source. Inside the China chip export regime, some of the same logic had applied to open source software, where basically American policymakers were concerned that if we encouraged open source, it would allow Chinese companies to copy American models. And yet, in the new official policy, They wrote, we need to ensure America has leading open models founded on American values.

Open source and open weight models could become global standards in some areas of business and an academic research worldwide. For that reason, they also have geostrategic value. While the decision of whether and how to release an open or closed model is fundamentally up to the developer, the federal government should create a supportive environment for open models. Now, of course, since then, we've gotten the first open AI open model in a number of years, but it's still pretty touch and go about broader questions like chip access. We've gone into a lot of detail about this particular debate and other shows, and there's nothing hugely new here, so I'll leave it there for now.

I just think that you can't really talk about the debate shaping AI without talking about this one, which is such a big driver of policy right now. Now, one story that has gotten a little bit of resonance in the past few days that's about the broader acceleration versus deceleration theme, are these new hunger strikes going on outside the offices of some major AI labs. On September 4th, Guido Reichstader wrote, I'm on hunger strike outside of the offices of the AI company Anthropic Rite. now because we are in an emergency. Anthropic and other AI companies are racing to create ever more powerful AI systems. These AIs are being used to inflict serious harm on our society today and threatened to inflict increasingly greater damage tomorrow. Experts are warning us that this race to ever more powerful AGI puts our lives and well-being at risk as well as the lives and well-being of our loved ones. Now, so far at least, this hasn't gotten a huge amount of resonance or traction. Chabion

X writes, the most foolish thing about his hunger strike against AI is that he does not provide a single argument as to what the danger of AI would be and how it would pose a threat. It is just wild resentment mixed with a feeling of fear without any profound criticism. This prompted another hunger strike, this time from someone named Michael Trazzi outside of Google's Deep Mind office in London. It's basically just a copy of the same post from Guido and got even less resonance. Some are claiming that Michael is a failed AI entrepreneur who's just looking for a way to get some attention. I don't know enough to comment strongly on that. What I am watching is to see whether this actually picks up any sort of attention

in mainstream press or whether it stays confined to Twitter slash X. The last debate that I want to just mention, and I'm going to go extremely quickly on this because I think it's going to be a full show later this week, is that if all of these are big sort of macro debates around the state of AI, how good it is, whether we're in a bubble, are we heading towards a trough of disillusionment, there are also debates happening inside the AI industry about a variety of interesting technical topics. Now, to some extent, these are even more interesting because they have direct resonance in terms of how AI is getting built.

The one that I've been seeing a lot of recently is a debate around e-vows. Entrepreneur Justin Torre wrote, literally F-Evals. We wasted so much time at Helicon building evals because, quote-unquote, customers wanted it, but this is a classic example of the mom test. We needed to distinguish the core customer problem with what they are asking for. Alex Reibman wrote a post called evals are a scam and were being gaslit into believing they aren't. Now, some people are trying to bring nuance to the conversation. Eugene Yan wrote, I've seen teams try to apply evals and get the same outcome and gain reluctance.

to eval. This is because they used off-the-shelf evals by e-vall platforms like faithfulness, cohesion, and helpful. Generic evals aren't useful. Your evils must be aligned with your user problem, and if they're not, you're just measuring noise. In any case, like I said, this is interesting enough that I think it's going to be an entire episode. So I'm flagging it in the context of this show just as a stand-in for all of the interesting technical and builder debates that are happening under the surface as well. So guys, those are five debates shaping AI right now. The field as ever remains dynamic, interesting, and increasingly a receptacle for society's broader anxiety. How much challenge that creates for us in the next few months, we'll just have to wait and see.

For now, that's going to do it for today's AI Daily Brief. Until next time, peace.