Claude Code Killed the AI Bubble

Today on the AI Daily Brief, how Claude Code Killed the AI Bubble. The AI Daily Brief 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, Assembly, Robots and Pencils, Blitzy, and Super Intelligent. To get an ad-free version of the show, go to patreon.com slash AIDLebrief or you can subscribe at Apple Podcasts.

Ad free is just three dollars a month. If you are interested in sponsoring the show, send us a note at sponsors at AIDAybrief.ai. And finally, while you are at aidailybrief.ai, you can find out about all the various projects that we have going on, including one which as I am recording this, I have not pushed live yet, but which I am clearly committing to by Sunday.

It's the follow-up to our AI New Year's Self-Directed Learning Program, and it's going to be a new program to match OpenAI's internal objective of agent-first work by March 31st. The safest thing is to go to aidailybrief.ai to look for the link, but I assume it will also be on march31.ai or adbtraining.com. And with that announcement out of the way, let's move on to today's episode.

So this is a weekend episode, which as you guys know is a long reads and or big think episode. And there is a really interesting theme that has taken hold that I think is so fascinating and a perfect encapsulation and capstone to everything we've been talking about throughout 2026 so far. On Thursday, of course, we got two frontier models within 20 minutes of each other: Anthropics Opus 4.6 and OpenAI's ChatGPT 5.3 codec.

Something about this clicked for people. Prominent thinker Tyler Cowan wrote, Today we'll go down as some kind of turning point, somewhat arbitrarily, but it is okay if journalists and historians have to present things in that manner. Nathan Young wrote, If you're walking around SF, does it feel like the early days of COVID where it's clear what's on everyone's mind?

Wayne on Twitter said, Can someone explain to me what concrete thing happened in the last forty eight hours that explains the fact that I've seen fifty seven thousand two hundred and forty six vague posts like this one? Andy Masley wrote, I know everyone's saying it's feeling a lot like February twenty twenty, but it is feeling a lot like February twenty twenty.

So what is going on? Investor Chow Wang put it simply. He wrote, I think AI is much less of a bubble than I thought two months ago. And pretty much everyone I know who used Claude and Codex in the last two months feels that way. In short, what we have experienced so far in 2026 is a set of cascading recognitions.

As we've discussed ad nauseum, it took even the most enfranchised and technical AI users going home over the holidays and having some time and space to really understand just how different the capabilities of the models, including Opus 4.5 and Codex 5.2, really were. Claude Code, of course, became the harness encapsulation of using those things to transform what you can do.

When people came back, they started talking about how they had pushed more code in the last two weeks than they had done in the year before. You started to see a shift in the narrative. Where even the folks who had previously said vibe coding is just for prototyping were now recognizing that agenda coding was kind of for everything. Claude Cowork came out, and the team behind it revealed that they had put it together in ten days, and it was basically exclusively coded using Claude Code.

Now Claude Cowork was interesting as an inflection point in the story, because it came out around the middle of the month, and that's when the mainstream started picking up on this story as well. It wasn't just that they were using Claude Cowork, although many of them were. Some were even finding their way into Claude Code, even though it's technically more challenging.

You started to see think pieces show up in business and finance publications away from technology, about how different the agentic capability set was with Claude Code. And you started to see it have a market impact as well. The new concern started to be less about an AI bubble, and more about what some dubbed the SAS Pocalypse, a broad-based plunge specifically in software, but not other types of technology stock.

Where the rise of these agentic coding tools had people really questioning how valuable and how durable the positioning of those SaaS companies was. That is the environment into which 4.6 and 5.3 codecs came. And that's the environment in which Semi-Analysis wrote their recent post, Clawed Code is the inflection point.

So this will be the long reads portion of this episode, and we'll read not the whole thing, but a number of excerpts from the great team at semi-analysis that starts with a fairly profound stat. Claude Code, which was released less than a year ago in March of 2025, as a research preview, mind you, just about one month after Andre Carpathy coined the term vibecoding, now represents 4% of GitHub public commit.

And you can see in this chart that this is accelerating. There started to be viral growth around October. Then at the beginning of January, things really started to heat up. It came in part around Boris, the creator of Cloud Code, introducing himself on Twitter and starting to talk about how he used it. But obviously there has been a lot going on this month that has significantly increased the engagement with Claude Code.

OpenClaw Multebook. This has been the story of 2026 so far. Semi-analysis' Dylan Patel continues, at the current trajectory, we believe that Claude code will be 20% plus of all daily commits by the end of 2026.

While you blinked, AI consumed all of software development. Let's continue on into the larger piece. The semi-analysis team writes, We believe that clawed code is the inflection point for AI agents and is a glimpse into the future of how AI will function. It's set to drive exceptional revenue growth for Anthropic at twenty twenty six, enabling the lab to dramatically outgrow OpenAI.

Anthropic, they argue, is on track to add as much power as OpenAI in the next three years. They then share a building-by-building tracker of Anthropic and OpenAI, and write Sam's AI lab is notably suffering from multiple data center delays. And since more compute means more revenue, we can forecast ARR growth and compare anthropic to open AI directly.

Notably, they continue, our forecast shows that Anthropic's quarterly ARR additions have overtaken OpenAIs. Anthropic is adding more revenue every month than OpenAI. We believe Anthropic's growth will be constrained by compute. The next section they call Claude Code in the Agentic Future. Agents, they write, will be the primary method of how organic intelligence, humans, interacts with artificial intelligence.

But Claude Code is also a demonstration of the reverse, showing how agents interact with humans. We believe the future of AI will be about the orchestration of tokens, not just selling tokens at base cost. With history as a guide, we view the OpenAI ChatGPT API as the call and response of tokens akin to Web 1.0 with TCP/IP, connecting users to static websites hosted on the Internet.

While TCPIP is a foundational technology, this communication protocol became just the means to the end of enabling the Internet during Web 2.0 and the shift to dynamic web pages. Today the Internet uses TCPIP packets to organize much larger sets of information than a static website. The protocol matters, but it was the applications built on top of this protocol that created trillions in value.

This is why semi-analysis believes we are yet again at another critical moment in AI, one that matches, if not exceeds, the ChatGPT moment in early 2023. Each moment expanded what AI could do. GPT-3 proved scale workable. Stable Diffusion showed AI could make images. ChatGPT proved demand for intelligence. Deep Seek proved that it could be done on a smaller scale. And O1 showed you that you could scale models to even better performance.

The viral moments of Studio Ghibli are just adoption points, while Claude Code is a new breakthrough in the agentic layer of organizing model outputs into something more. Now in describing Claude Code, they continue, it might be incorrect to think of Claude Code only as focused on code, but rather as claude computer. With full access to your computer, Claude can understand its environment, make a plan, and iteratively complete this plan, the whole time taking direction from the user.

Claude Code does more than just code and is the best example of an AI agent. You can interact with a computer with natural language to describe objectives and outcomes rather than implementation details. Provide Claude and input such as a spreadsheet, a code base, a link to a web page, and then ask it to achieve an objective. It then makes a plan, verifies details, and then executes it.

It's a glimpse of the future, but it is also here today in software already. Your favorite engineers are vibe coding. Andre Carpathy, who coined the term vibe coding one year ago, is openly discussing the phase shift. And specifically says, I've already noticed that I am slowly starting to atrophy my ability to write code manually. Generation, writing code, and discrimination reading code are different capabilities in the brain.

Maltai Ubel, the CTO of Vercell, claims that his new primary job is to tell AI what it did wrong. Ryan Dahl, creator of Node.js, says the era of humans writing code is over. David Henimer Hansen, creator of Ruby on Rails, is having some sort of anticipated nostalgia, reminiscing about writing code by hand while writing code by hand.

Boris Cherney, creator of Claude Code, says that pretty much one hundred percent of our code is written by Claude Code and Opus four five. Even Linus Torvalds is vibe coding. But it isn't just coders. from which semi-analysis describes how the different members of their team all use this tool in different ways. They write that the data center model team needs to review hundreds of documents every week. The AI supply chain team needs to inspect BOMs with thousands of line items.

The memory model team needs to build forecasts in near-real time as spot market prices explode. Technical staff needs to maintain a live dashboard, meaning in total as they write, from regulatory filing to permits, spec sheets to documentation, config to code, the way we interact with our computers has changed. Coders will stop doing code and rather request jobs to be done on their behalf. And the magic of Claude Code is that it just works.

Many famous coders are finally giving in to the new wave of vibe coding and now realizing that coding is effectively close to a solved problem that is better off supported by agents than humans. The locus of competition is shifting. Obsessions over linear benchmarks as to what model is quote unquote best will look quaint, akin to how fast your dial-up is compared to DSL.

Speed and performance matter, and the models are what power agents, but performance will be measured as the net output of packets to make a website, not the packet quality itself. The website features of tomorrow is going to be the orchestration through tools, memory, subagents, and verification loops to create outcomes and not responses. And all information work is finally addressable by models.

If you're building anything with voice AI, you need to know about assembly AI. They've built the best speech-to-text and speech understanding models in the industry, the quiet infrastructure behind products like granola, dovetail, ashby, and clule. Now, as I've said before, voice is one of the most important modalities of AI. It's the most natural human interface, and I think it's a key part of where the next wave of innovation is going to happen.

Assembly AI's models lead the field in accuracy and quality so you can actually trust the data your product is built on. And their speech understanding models help you go beyond transcription, uncovering insights, identifying speakers, and surfacing key moments automatically. It's developer first, no contracts, pay only for what you use, and scales effortlessly. Go to assemblyai.com slash brief, grab$50 in free credits, and start building your voice AI product today.

Most companies don't struggle with ideas. They struggle with turning them into real AI systems that deliver value. Robots and Pencils is a company built to close that gap. They design and deliver intelligent cloud-native systems powered by generative and agentic AI, with focus, speed, and clear outcomes. Robots Pencils works in small high-impact pods.

Engineers, strategists, designers, and applied AI specialists working together to move from idea to production without unnecessary friction. Powered by RoboWorks, their agentic acceleration platform, teams deliver meaningful results, including initial launches, in as little as 45 days, depending on scope.

If your organization is ready to move faster, reduce complexity, and turn AI ambition into real results, Robots and Pencils is built for that moment. Start the conversation at robotsandpencils.com slash AI Daily Brief. That's robotsandpencils.com slash AI Daily Brief. Robots and pencils impact at velocity. Weekends are for vibe coding. It has never been easier to bring a passion project to life, so go ahead and fire up your favorite vibe coding tool.

But Monday is coming, and before you know it, you'll be staring down a maze of microservices, a legacy COBOL system from the 1970s, and an engineering roadmap that will exist well past your retirement party. That's why you need Blitzy, the first autonomous software development platform designed for enterprise scale codebases. Deploy the beginning of every sprint and tackle your roadmap 500% faster. Blitzy's agents ingest your entire code base, plan the work.

And deliver over 80% autonomously. Validated end-to-end tested premium quality code at the speed of compute. Months of engineering compressed into days. Vivecode your passion projects on the weekend, bring Blitzy to work on Monday. See why Fortune 500s trust Blitzy for the code that matters at Blitzy.com. That's B L I T Z Y.com.

Today's episode is brought to you by my company, Super Intelligent. In 2026, one of the key themes in enterprise AI, if not the key theme, is going to be how good is the infrastructure into which you are putting AI and agents? Superintelligence agent readiness audits are specifically designed to help you figure out one, where and how AI and agents can maximize business impact for you, and two, what you need to do to set up your organization to be best able to leverage those new gains.

If you wanna truly take advantage of how AI and agents can not only enhance productivity, but actually fundamentally change outcomes in measurable ways in your business this year, go to bsuper.ai.

And this is really the big theme they pick up from there. That the reason that this is an inflection point moment is not just about coding capability, but about what that leads to. They continue. Coding was once the most valuable work of all, with programmers in hot demand during the 2020 era of software engineering. Coding is now a beachhead in terms of the disruption that agentic information processing has, and the larger$15 trillion information work economy is now at risk.

There are one billion plus information workers or roughly a third of the global workforce. Every single workflow in the information work category is often similar, and shares a workflow that Claude Code proves works for software. Read, ingest unstructured information, think, apply domain knowledge, write, produce structured output, and then verify, check against standards.

This is large swaths of most information workers, including research. And if agents can eat software, what labor pool can they not touch? Our view is quite a few, and with the rise of Claude Code and Cowork, the total addressable market of agents is much larger than LLMs. Given the killer use case in coding and the clear generalizability of Claude Code and Cowork, this justifies a completely different calculus.

Automating most call and response and information fetching is likely doable, and this opens the absolute dollars possible. And what they say really makes larger parts of the Pi available for disruption is longer task horizon. How long can an agent work before it fails its task? Meter data shows autonomous task horizons doubling every four to seven months, accelerating to around every four months in twenty four and twenty five.

Each doubling unlocks more of the total pie. At 30 minutes you can autocomplete code snippets. At 4.8 hours you can refactor a module. Multi-day tasks you can automate an entire audit. And it's clear Anthropic sees this too. On January 12, 2026, Anthropic launched Cowork, Cloud Code for General Computing. Four engineers built it in ten days. Most of the code was written by Cloud Code itself.

Same architecture, Claude Agent STK, MCP subagents. It creates spreadsheets from receipts, organizes files by content, and drafts reports from scattered notes. It's Claude Code minus the terminal plus a desktop. This is the Glimpse of the Future, a harness that understands the context of your day-to-day job or work and can build and generate information processing as needed.

Instead of creating images from reports you download from your database, an agent will generate a report with better formatting than you could do yourself within Excel for you. Whenever you need to gather information about, say, a sales quota, your agent will extract the information from a UI or API, and generate the report for you on your behalf.

Information work itself is going to be automated like Cloud Code has automated software engineering. And while it's not perfect today, it clearly can generally process, synthesize, and format data faster than most humans can. This all comes at higher fidelity and lower costs than the average worker in some areas. While there will be hallucinations, most systems already exist with many human led errors in the process.

If the information is processed at a viable level of fidelity and then passed to the next step, this itself will massively increase the supply of work. We are literally at the point where any individual could type into one of these agent workflows to run a multivariable regression that would have taken a lifetime of training in the 2000s. The Stack Overflow 2025 Developer Survey has 84% of coders using AI, and that is the bleeding edge of adoption. Only 31% use coding agents.

And that means that this penetration curve is early for broader waves of information work. Just like the blink for coding agent penetration, broader information work will quickly see AI adoption. Now the last section of this piece that we're going to read is about cost and market impact. They have a whole secondary section on competitive race and who's winning, but that's less the point at least for this show.

Moving back to where we left off, they write, Now engineering has and always will be the gold standard information work, but as the quality has finally crossed over a critical threshold, the relationship between coders and their tools have flipped. Coders are effectively just harnessing a black box to achieve outcomes. And that was all possible because not only the quality but the cost of the intelligence of tokens has fallen an amazing amount.

One developer with Cloud Code can now do what took a team a month. An enterprise is already starting to move. The massive deflationary cost of intelligence is going to reprice every information company's margin for repeatable work.

Accenture just signed a deal to train thirty thousand professionals on Claude, the largest Claude code deployments to date. Accenture will focus on financial services, life sciences, healthcare, and the public sector. Those are all huge untapped markets for information automation. OpenAI just announced Frontier focused on enterprise adoption. Enterprise software has easily been the first casualty of the great cost decline of intelligence.

SAS itself is just crystallized information processing of workflows into code. The three motes of SAS, switching costs of data, i.e., data is trapped. Workflow lock-in, i.e., learning the UI, and integration complexity, how Slack works with JIRA, have all been partially eroded at the margins.

The 75% gross margin of SAS looks like a huge opportunity. As agents migrate data between systems with lessened migration costs, agents themselves do not rely on human-oriented workflows, and MCP integrations make integration much easier. Every aspect of SAS is cheapening and the margins have become the first opportunity of AI. In our view, anything that has a human click buttons, gather information, reformat it into another medium is a huge risk.

So okay, that's the part of this essay that we're going to read. And when push comes to shove, the key phrase here is inflection point. What's important about the last month is not just that en masse, the most enfranchised and highly technically literate AI users realized that we had reached an inflection point. It's that that perception has now cascaded into the wider world.

What really crystallized this for me, and what basically prompted me to want to do this show, was when former Atlantic author and co-author of Abundance, Derek Thompson, tweeted out on Thursday, For me, the odds that AI is a bubble declined significantly in the last three weeks. And the odds that were actually quite underbuilt for the necessary levels of inference and usage went significantly up in that period.

Basically, I think AI is going to become the homescreen of a ludicrously high percentage of white collar workers in the next two years. And parallel agents will be deployed in the battlefield of knowledge work at downright Soviet level.

The New York Times Kevin Roos reposted it and said, This is why everyone was freaking out about clawed code over winter break. Once you see an agent autonomously doing stuff for you, it's so instantly clear that roughly all computer based work will be done this way. Kevin continued, this is why my serious AI policy proposal is to sit every member of Congress down in a room with laptops for thirty minutes and have them all build websites.

Deedra Bassa, who you might remember in preparation for a show about the SAS pocalypse as a reporter for CNBC, tried to code herself up a version of Monday dot com, not expecting to actually do anything. About an hour later she had a fully working version and kind of became a convert. The way that she described this shift, which I thought was quite crisp, was that over the last couple of months, in her words, AI went from talking to doing.

Now, not everyone fully agrees. Mike Catone reposted Derek and said, I agree mostly with this. However, there's a big assumption contained within that the organizations these white-collar workers are employed by actually have the appetite to integrate the tools. Lots of process and system change will need to be made with current capabilities.

I think it goes even farther than that. To put it bluntly, the value of using AI well has gone way up. But the difficulty of learning how to use AI well has also gone way up. That makes the natural enterprise inertia barriers even more pronounced. There's also plenty of reactions like this one from Van Jackson, who writes the AI bubble is about lack of profitability in firms being overleveraged, not about usage.

Everyone already uses AI unprofitably, destroying most of the workforce and press ganging those still clinging to jobs into using AI changes nothing. But this at least on the market side is kind of what's shifted.

The interesting wrinkle that this adds to the bubble conversation, and the reason that folks like Derek and Chow are talking about why an AI bubble is likely, is that for the average person, the AI bubble argument was that we were overbuilding AI infrastructure that maybe we weren't even going to need, or that maybe these companies couldn't even pay for.

You kind of want it running all the time. In fact, you want multiple agents running all the time to do more things. Multiple agents running all the time means more tokens consumed. And that, as Ethan Moloch puts it, we are going to need more compute now that agents can complete long term economically viable tasks.

Ethan clarifies this does not mean that there couldn't be some sort of financial issue with financing the compute, but does point to the idea that compute is not being overbuilt. And that is what is at least starting to shift. Now it would be way overblown to argue that this has fully found its way into public market. But you're starting to see it happen and it's kind of headspinning as no one knows what all these signals taken together should mean in aggregate. Seb K sums up the confusion.

Sudden smart consensus today is that the AI takeoff is rapidly and surprisingly accelerating, but stocks for Google, Microsoft, Amazon, Facebook, Palantir, Broadcom, and NVIDIA are all down around ten percent over the last five days. SMCI's down ten percent today. This by the way was from Thursday. Only Apple's up and it's the least AI. Strange in my opinion. All I can say is buckle up friends, because I think we are in for an interesting and confusing period.

Back in October, OpenAI's rune wrote, Not enough people are emotionally prepared for if it's not a bubble. And I kind of think that's part of what we're seeing here. AI, as Deirdre put it, over the last couple months has entered the show-not tell phase.

It's doing things, not talking about them. Agents have turned the corner from a thing that would be really cool to a thing that is doing real work right now. And everywhere around us, the signals that the way that work is done has changed are profound.

To take an example that we shared the other day, OpenAI President Greg Brockman says that by March 31st, for any technical task that happens inside that company, the tool of first resort for humans is interacting with an agent rather than using an editor or a terminal. Agent First Work by March 31st.

Now, as I mentioned in the intro, if you want to get on that timeline as well, I decided to throw together another free self-directed learning experience like the New Year's resolution, because heck yeah, if Greg is going to challenge his team to meet that goal, why shouldn't the rest of us figure it out too? In any case, whether Tyler Cowan is right, and last Thursday when Opus 4-6 and 5-3 codecs were released goes down in history as some kind of turning point.

What's clear is that a shift has happened. It has in fact been happening for two months, but now it is fully working its way through the system, and everyone is grappling with the implications. I wish all of you listeners nothing but the best navigating this period, and I will of course continue to do my best trying to help you make the most of it. For now, that is gonna do it for today's AI Daily Brief. Appreciate you listening or watching as always, and until next time, peace.