The Calm Before the AGI Storm

0:00Today, we are catching up on the most important recent AI news and talking about the calm before the AGIi storm. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:36Now, one other quick announcement. For those of you who are looking for ways to get your team up to speed with building custom agents and agent teams, we are launching our second cohort of Enterprise Claw. The program is led by Nufar Gaspar, who you've seen on this show numerous times, including last week. And you can find out more at EnterpriseClaw.ai. I expect it will fill up quite quickly. So again, if you want to check it out, it's EnterpriseClaw.ai. Now, as for us here today, I am back from traveling. We had a great spring break with the kids. got to see some giraffes and fireworks out our window at Disney, and my fears that some crazy thing would happen in AI that demanded that I pull my head up from the parks and put it back into AI land did not materialize.

1:16And yet, although there was not any one huge massive story like some new model coming out, the last week did have many stories that fell along the themes of our episode from yesterday, which was the six questions shaping AI. Even more than that, though, I think when you take the sum total of the news from the last week or so, there is a very distinct picture emerging. I'm calling it the calm before the AGI storm. And what it feels like to me is that even in the quiet moments for AI, the big labs are all jostling and positioning for a very different and fast-moving future. You almost have that feeling of the electric charge that's in the air before a thunderstorm.

1:55So let's talk about the most important stories from last week and why they all add up to something that is even bigger than they might at first seem. As they are wont to do, OpenAI was in the news throughout the week. And the company actually began the week on a pretty high note as they closed a record-breaking fundraising round. Now this is the round we've been hearing about for a while, with the company having already announced the first$110 billion of the funding back in February. You might remember that that round was sourced from Amazon, NVIDIA, and SoftBank, but they've added an additional$12 billion to the round.

2:24this time from largely financial rather than strategic investors, meaning that the total size of the round is$122 billion, closed at an$852 billion valuation. Now, we had heard that the additional number was going to be around$10 billion, so the fact that it was$12 billion is a good sign that investor demand is still strong. For the first time, OpenAI took capital from individual investors through wealth management channels. That accounted for about$3 billion of this overall total. OpenAI also announced that their stock would be included in multiple ETFs managed by ARK Invest. Alongside the raise, OpenAI disclosed that they're now generating$2 billion in revenue per month, up from around$1.6 billion at the end of last year.

3:01The company said that they are currently growing revenue at four times the pace of the companies that defined the internet and mobile eras, including Google and Meta. That said, the picture wasn't all rosy. Although the fundraising was strong, Bloomberg reports that OpenAI stock is struggling to find buyers in secondary markets. Ken Smythe, who operates secondary marketplace Next Round Capital, said he's seen hundreds of millions of dollars worth of OpenAI stock come to market in recent weeks with no one biting. He said, we literally couldn't find anyone in our pool of hundreds of institutional investors to take these shares.

3:32Meanwhile, he added that his buyers have suggested that they have about $2 billion in cash ready to deploy into Anthropic. Now, some are suggesting that the lack of demand for OpenAI and the apparent interest in Anthropic is simply about the gulf in valuations. Anthropic last raised at$380 billion, but shares are currently changing hands at up to a$600 billion implied valuation. And for many, that still seems cheap compared to the official$852 billion number for OpenAI. Adam Crawley of Augment Capital said, It's just better risk-reward right now. People are betting that Anthropic's valuation will catch up with OpenAI's.

4:06But if you buy OpenAI shares, it's less clear what the return will be in the near term. Now, this market dynamic has huge implications as the two companies race to go public. It is very important to note that it is not a one-to-one comparison between public markets and private market for secondaries, but it still isn't a great sign if you are on the OpenAI team that there is some amount of exhaustion in those secondary markets. The dynamic could also mean that OpenAI doesn't really have a choice to keep raising money from private markets and has to go to IPO. Anyways, we'll come back to IPO discussions because that would be a big part of OpenAI's story later in the week.

4:40Before that, though, we got upheaval in the C-suite as there were several executive shuffles. CEO of AGI deployment, Fiji Simo, announced that she would be taking several weeks of medical leave. Simo suffers from a chronic neuroimmune condition and had experience to relapse shortly before beginning at OpenAI. In a memo to staff, Simo said she had deferred medical tests and new treatments in order to commit full-time to her role, but after having caught up on some of those tests recently, she said, it's now clear that I've pushed a little too far and I really need to try new interventions to stabilize my health.

5:09With her taking a step back. President Greg Brockman will run the product organization, while Chief Strategy Officer Jason Kwan, Chief Financial Officer Sarah Fryer, and Chief Revenue Officer Denise Dresser will take over the business and operations functions. In addition to this, there were several other leadership changes. Brad Lightcap will be stepping down as COO to begin a new role focused on special projects, which it sounds like will include leading the effort to form joint ventures with private equity firms as part of this larger sales and consulting play that we've been tracking. In addition to her function as Chief Revenue Officer, Denise Dresser will take on the COO role, and Chief Marketing Officer Katie Rausch has stepped down to focus on her cancer recovery.

5:46She will return in a more narrow role as her health allows, with former Meta Chief Marketing Officer Gary Briggs filling in for Rausch until OpenAI can find a permanent replacement. Now this to me very much does not read. As the type of leadership shakeup we've seen from other labs as they really try to get their ship in order, it just kind of feels like a bunch of things that have really bad timing. Still, an executive reshuffle is basically the last thing that OpenAI needs this year as the competition with Anthropic heats up and they shift their focus to preparing to IPO. Indeed, on that front, over the weekend, we got reporting that IPO strategy is a controversial topic among the leadership team.

6:19The information reported that Sam Altman and CFO Sarah Fryer are at odds over IPO timing and spending. Altman reportedly wants to take the company public in the fourth quarter, with reports suggesting he may even push to IPO ahead of Anthropic, who are targeting October to go public. Now, this is exactly what we talked about in our prediction post, where admittedly I said that I thought that when push came to shove, no one would actually go public in 2026, but that if that changed, it would be because OpenAI wanted to get out ahead of Anthropic. CFO Sarah Fryer, for her part, apparently doesn't believe that OpenAI will be ready this year due to the procedural and organizational work required ahead of the IPO.

6:55Information sources also said that Fryer is concerned about the risks from OpenAI spending commitments. Reportedly, she has expressed skepticism that OpenAI will need to pour so much money into data centers, and also apparently has doubts about OpenAI's revenue growth's ability to support those spending commitments. According to the most recent figures, Altman is committed to spend$600 billion on infrastructure over the next five years, with OpenAI's forecasts suggesting that they will burn$200 billion before turning a profit sometime towards the end of the decade. The information sources also suggest that this has gone beyond just a simple disagreement, suggesting that Altman has excluded Fryer from some conversations about financial plans, including a recent conversation about data center spending with a leading investor.

7:35The sources commented that her absence was noticeable and awkward given that she was present for previous discussions. Part of what makes this notable is that Fryer was explicitly brought in to have these types of tough conversations and bolster investor confidence. She had previously taken on a similar role at Square, getting the company's books in order and shepherding them to a successful IPO back in 2015. One source said that Fryer has a hard job, saying, quote, she's working with a founder with big ambitions who wants to push the envelope as hard as he can on spend. I think it's always worth taking these stories about executive disagreements with a grain of salt.

8:09The information sourcing is generally good, but it sounds like we're talking about a single missed meeting here, and there are about a million reasons why Fryer might have not been at that meeting that don't come back to some Machiavellian psychodrama. That said, to the extent that the meta story that we're exploring is the ratcheting up of stakes and the feeling of anticipation as the acceleration gets closer, this does feel like it could be an example of that growing intensity. Now, of course, the other big piece of OpenAI news from last week was their acquisition of tech talk show TBPN. Now, many of you listening to this show have probably come across TBPN, but for those who haven't, the show is a video podcast in the format of a daily talk show.

8:47They stream for three hours a day, and have become a very desirable place for tech executives, startup founders, and other commentators to show up talking about whatever the most recent news is. Part of why the story got so much traction is that it came hot on the heels of Fiji SEMO mandating the cutting down of side quests, and to some seemed like exactly that, despite the fact that it was reportedly SEMO herself who was pushing for the acquisition. Reactions were wildly varied. Any take that you have, you can probably go find someone who had it and who had it publicly on X. There was a lot of confusion, especially among more traditional commentators.

9:22Wharton professor Paul Neri wrote, Open AI acquiring TBPN makes zero sense to me, an M &A professor. New York Times reporter Mike Isaac argued that this was the culmination of tech's frustration with mainstream media coverage, a divide that has been growing for years now. He wrote, The Open AI buying TBPN story is, to me, the biggest proof point yet of CEO frustration with mainstream media coverage of tech at a time when consumers are growing increasingly skeptical of the effects of AI on society. I see this as a marketing expense. And indeed, that's what a lot of people thought about this. That this was a way to try to present a better face to open AI for the world.

9:58Slow Ventures investor Jack Rains writes, Everyone putting out the gigabrain future of marketing and media or whatever takes on TBPN and trying to apply that to other media brands is completely missing the point. This isn't a pattern matching thing. They aren't buying distribution. Sam's Twitter has bigger distribution than any tech media platform. The real takeaway is that if you build the right relationships with the right parties that find your skill set useful, they'll pay a gargantuan premium to work with you. Same as ML researchers copying fat compensation packages. One thing that's interesting, though, about this is that if the goal really is to bring John and Jordy, the founders and hosts of TBPN, closer into the open AI fold, the deal isn't really structured to do that.

10:37Indeed, they wrote very strict and clear editorial independence into the terms, leading to attention pointed out by Simon Smith. He writes, Here's the TBPN issue for me from a focused narrative standpoint. Either it maximally supports OpenAI's current focus on productivity, in which case it can't have full editorial independence, or it has full editorial independence, in which case it's a side quest. Basically, what Simon is arguing is that it can't both actually stay editorially independent and at arm's length if it is supporting this new, more focused OpenAI. There are also a few, like Robert Scoble, who thought that this was some big new play for the future of media, He writes,

11:32And frankly, I don't think this has anything to do with it. I think all it comes down to is this. For more than a year now, OpenAI has felt the sting of being the company most associated with a technology that is, to say the least, very publicly controversial. And for a couple years, the challenges inside the company have made coverage even more difficult. Meanwhile, they look over at TBPN, and it seems to have good media juju. And so the logic is buy them and see if you can't get some of that lightning in a bottle for yourself. Ultimately, I think it's a tough play. I think it creates some challenges for TBPN.

12:04On the one hand, I don't think that people care overly who owns different media properties. But at the same time, at the very least, I don't think OpenAI direct competitors like Anthropic are going to be particularly keen to break news with them anymore. To the extent that this is VG Simo and OpenAI wanting to leverage the host talent outside of the show because of their marketing instincts, which is part of the logic that she shared internally, I think that could certainly make sense for OpenAI, but also might not be best for the TBPN audience. But I think that the single biggest challenge is that TBPN indexes extremely highly in a demographic that I don't think is OpenAI's main problem.

12:38TBPN has become the preferred insider conversation space for tech, but at least so far hasn't broken super far outside of tech circles. I think going forward, not just OpenAI, but every AI's company's biggest challenge from a communications perspective is not going to be winning the narrative battle inside tech. It's going to be with everyone else. At the same time, ultimately, we are in uncharted territory, and you kind of just have to make big moves and see how they work.

13:09All right, folks, quick pause. Here's the uncomfortable truth. If your enterprise AI strategy is we bought some tools, you don't actually have a strategy. KPMG took the harder route and became their own client zero. They embedded AI and agents across the enterprise, how work gets done, how teams collaborate, how decisions move, not as a tech initiative, but as a total operating model shift. And here's the real unlock. That shift raised the ceiling on what people could do. Humans stayed firmly at the center while AI reduced friction, surfaced insight, and accelerated momentum. The outcome was a more capable, more empowered workforce.

13:43If you want to understand what that actually looks like in the real world, go to www.kpmg.us.ai. That's www.kpmg.us.ai. With the emergence of AI code generation in 2022, NVIDIA master inventor and Harvard engineer Sid Paresci took a contrarian stance. Inference time compute and agent orchestration, not pre-training, would be the key to unlocking high-quality AI-driven software development in the enterprise. He believed the real breakthrough wasn't in how fast AI could generate code, but in how deeply it could reason to build enterprise-grade applications. While the rest of the world focused on co-pilots, he architected something fundamentally different.

14:21Blitzy, the first autonomous software development platform leveraging thousands of agents that is purpose-built for enterprise-scale codebases. Fortune 500 leaders are unlocking 5x engineering velocity and delivering months of engineering work in a matter of days with Blitzy. Transform the way you develop software. Discover how at blitzy.com. That's B-L-I-T-Z-Y dot com. Today's episode is brought to you by Robots and Pencils, a company that is growing fast. Their work as a high-growth AWS and Databricks partner means that they're looking for elite talent ready to create real impact at velocity.

14:53Their teams are made up of AI-native engineers, strategists, and designers who love solving hard problems and pushing how AI shows up in real products. They move quickly using RoboWorks, their agentic acceleration platform, so teams can deliver meaningful outcomes in weeks, not months. They don't build big teams. They build high-impact, nimble ones. The people there are wicked smart with patents, published research, and work that's helped shape entire categories. They work in velocity pods and studios that stay focused and move with intent. If you're ready for career-defining work with peers who challenge you and have your back, Robots and Pencils is the place.

15:28Explore open roles at robotsandpencils.com slash careers. That's robotsandpencils.com slash careers. You've heard me talk about Assembly AI and their insanely accurate voice AI models, but they just ship something big. Universal 3 Pro is a first-of-its-kind class of speech-language model that lets you prompt speech recognition with your own domain context and vocabulary, instead of fixing transcripts and post-processing. It's more flexible than traditional ASR and more deterministic than LLMs, so you get accurate output at the source and can capture the emotion behind human speech that transcripts often miss, all without custom models or post-processing hacks.

16:03And to celebrate the launch, they're making it free to try for all of February. If you're building anything with voice, this one's worth a look. Head to assemblyai.com slash free offer to check it out.

16:19Moving away from OpenAI, though, their chief competitor, Anthropic, had one hell of a week themselves. On Tuesday, an update to Claude Code included 512 ,000 lines of source code for the platform. Anthropic quickly removed the code, but not fast enough for the internet, where people began hosting the code across various platforms. Anthropic's lawyers spent the following day taking down over 8 ,000 GitHub repos using copyright claims. But as it turned out, most of these repos weren't infringing and were actually forks of publicly released versions of Claude Code. Anthropic's Boris Cherney later apologized for the mistake and retracted all but one of the takedown notices.

16:53Anthropic blamed the code release on human error and did not imply a security breach. The leak allowed us to learn about a few unreleased features that may be planned for Claude Code. Perhaps most notably, the repo included an always-on agent called Kairos, which allows Claude Code to work in the background and send periodic updates to a user's phone. Kairos includes a Dream Mode, which allows it to autonomously consolidate memory across sessions, and it can also work in a Proactive Mode, allowing it to take initiative and make progress without needing instructions. On the other end of the spectrum of seriousness, the code included a virtual pet feature called Buddies.

17:25The Tamagotchi-like features used duck avatars, and the source code implied that Anthropic hoped to generate what they called sustained Twitter buzz from the feature. Others took a deep dive into the code and found that Claude Code is way more complex than it appears. It contains five different compaction strategies to compress context, dozens of tools, caching optimizations for sub-agents, and highly configurable system prompts. Yuchen Jin wrote,

18:15Many suspected this was a bug in the way that prompt caching was handled, and the chatter grew so loud that Anthropic was forced to investigate. On Thursday, Cloud Code developer Lydia Halley delivered the results, posting, Peak hour limits are tighter and million token context sessions got bigger. That's most of what you're feeling. We fixed a few bugs along the way, but none were overcharging you. We also rolled out efficiency fixes and added pop-ups in product to help avoid large prompt cache misses. Halley suggested that users should switch to Sonnet, lower effort levels, and start fresh sessions instead of resuming after an hour.

18:45Alex Volkoff, host of the Thursday AI podcast, represented the voices of many when he found this response deeply unsatisfying. He said, uh-oh, Anthropik's official response to everyone burning through their sessions is, you're holding it wrong? Come on. I'm sure that this won't go well with the thousands of folks who experienced a significant decrease in their ability to use their pro and max plans and are canceling in favor of other solutions. Now to cap it all off, on Friday, Anthropic announced changes to charge more for people using Clawed to manage their OpenClaw. As of Saturday, users will no longer be able to use their subscription for third-party tools.

19:19Now to be clear, you can still use Clawed models to drive OpenClaw, but you have to do it on a paper token basis via the API. OpenClaw creator Peter Steinberger said he had tried to talk OpenClaw out of this move, but only managed to delay the decision for a week. Author Daniel Jeffries suggested that this is about more than just the competitive dynamic between OpenClaw and AnthropicsClawedCode. He writes,

20:02is not cheap. Anyone who thinks we will be running super-intelligent agents around the clock on the most expensive chips ever made, chips that depreciate to worthless in three years, while running in data centers on nuclear power is not doing the math. Things do get cheaper over time, but the key is over time. The best models are more and more costly to build and run, and will be for a long time, barring some kind of revolutionary architecture that replaces the transformer with something much more memory efficient, or wetware-style chips that sip power or both. Intelligence going up into the right keeps eating the bleeding edge of the best chips in memory as fast as we can make them.

20:33Older tasks will get cheaper and easier, and on-device models will be able to do cool things. But those machines are not cheap either, unless you think four Mac studios networked together is cheap. This smashes the AI does all the jobs theory to little bitty pieces. Good. It's terrible PR for the industry to keep babbling on about it anyway. True intelligence will be like paying for a full-time salary to people and then the calculation of whether it's just cheaper and less error-prone to throw more people at the problem comes into play. Hint, it's usually not cheaper to use the machine. All this is not to say that agents are not incredible and valuable, but the subsidy era is coming to an end.

21:06It always was. And everything just happens faster in the age of AI. I think this is a super, super salient point and one that we are going to be watching much more closely. Because if Daniel is right, and that if we're actually about to see what it really costs to run these highly intelligent models, and that that cost looks a lot closer to human salaries than we think it does, obviously that has pretty dramatic implications for the whole jobs conversation. On the other end of the spectrum, Google actually increased its open-source capabilities with the release of Gemma 4. The new model is the latest in Google's open-source family and represents a significant jump in capabilities.

21:43Google claims the model family delivers state-of-the-art performance across four different sizes. The lineup includes 2B and 4B models for small-edge devices, as well as a$26 billion mixture of experts model and a$31 billion dense model. The 31B model is currently ranked at number 3 on the Arena AI text leaderboard for open-source models, behind Kimi K2.5 Thinking and ZAI's GLM-5. The models are optimized to deliver strong coding and agentic performance with Google writing, This new level of intelligence per parameter means achieving frontier-level capabilities with significantly less hardware overhead.

22:15The models are built on the same architecture that underpins Gemini models, meaning you can expect a similar feature set. Now, this is the first Western open-source model competing at this level in years and could have big implications. Greg Eisenberg wrote, Thinking about Google's Gemma 4 and what it means. A few months ago, running something this capable locally meant serious hardware and serious trade-offs on quality. Now it runs on your laptop, works offline on your phone, speaks 140 languages natively, 256k context window, costs nothing, lol, performs better than models 20x its size, and you can swap it in as your model in Claude Code, Cursor, Hermes, or OpenClaw right now.

22:51Okay, here we go. It's a good time too, because in China, Alibaba continues its shift away from open source. Alibaba released three proprietary models in three days, culminating in the release of Quen 3.6 Plus on Thursday. Even though this is closed, it's still in the good enough performance plus better cost camp. For example, it lags Opus 4.5 by a few points on Sweebench Verified, has full multimodal capabilities and utilizes a million token context window. But cost is massively reduced at around one-eighth of Opus. This release reinforces Alibaba's new strategy of proprietary models to capture more revenue from their models.