← The AI Daily Brief: Artificial Intelligence News and Analysis
The AI Daily Brief: Artificial Intelligence News and Analysis
Surprise Elon Anthropic Team Up Reshapes the AI Race
00:31:25
Loading summary
Transcript1 lines
- [00:01]
A
Today on the AI Daily Brief, a surprise team up between Elon and Anthropic could totally reshape the AI race. And before that in the headlines, sorta. It's kind of all one big episode today. Everything that was announced at the Code with Claude Anthropic event yesterday. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG Granola section and zencoder. To get an ad free version of the show go to patreon.com aidaily brief or you can subscribe on Apple Podcasts. If you want to learn more about sponsoring the show, send us a Note@ SponsorsIDailyBrief AI and just a couple other action items. First, the April AI Usage Pulse survey is live. If you want to see all the results from the previous months, go to Pulse aidailybrief AI. There's a link to the survey from there as well. And then a quick mention of our latest AIDP training programs. The Free Agent OS Self directed program to build an agentic operating system has tons of you building right now and I'm really really loving seeing all the things that you're doing. And then if you are a team that is looking for something similar but more hands on, we've got the third cohort of Enterprise Claw registering and if you need a little primer even from that, a new four week executive Sprint called AI Executive Catch Up. That'll take you from basic prompting all the way to being ready to build your first agents. As always, all the info for that can be found at aidleepbrief AI. Now with that out of the way, let's talk about yesterday's big news. Yesterday was Anthropics Developer Day and so I knew that today's episode was going to be all about whatever it was that they announced. Now there were some really interesting things and in many ways I think you can view their conference as a really interesting indication of where companies are with agents, the problems they're trying to work on, and the significance of harnesses in the AI race. But when it comes to the AI race, everything that was announced at Dev Day was absolutely drowned out by the surprise announcement of an Elon Musk, SpaceX and anthropic team up. So what we're going to do today is we're not going to strictly divide this into headlines in main, but instead the practical tool and tech related announcements from Dev Day are effectively the headlines, while the Elon Anthrop story is going to be the equivalent of the main but let's talk about Dev Day, because I don't want to undersell how interesting it was in even though the other thing is dominating the conversation. Right from the beginning you could tell where the emphasis was going to be in this event, given that they called it Code with Claude. Now at last year's event, anthropic rolled out Sonnet 4 and Opus 4, which was their response to OpenAI's O3. This year there was no big model release. We didn't get a public rollout for Mythos or even a hint of when and how that might happen. Nor did we get any hints about Opus 48 or 49 or anything in the line. Instead, the focus was squarely on agents and the applications built around Anthropic's models, which I think reflects how competition has changed in AI over the last six months. It would be insane to say that models don't matter at this point. Most high value production use cases are of course using either Opus 4.7 or GPT5.5, and the advances specifically of GPT5.5 have helped OpenAI significantly reclaim some narrative space. And yet, if you had to put your finger on the important competition of 2026, it's been way more about Codex versus Claude Code than it has been about Opus versus GPT. And I think what we're starting to see is the next evolution in the process. Increasingly, you see Claude code evolving into an ecosystem of agent harnesses that are tuned for particular workflows. That process began with Cowork, but has been refined into even more specific workflows like CLAUDE design. And while OpenAI is taking a different approach with codecs trying to centralize all the activity in that one app space, you are at least seeing little hints of this harness customization, for example, in the form of Codex's quickstart profiles for various different professions. But when it comes to what was announced at Code with Claude this week, it reads like a map of the key challenges of agents. We have features focused on memory, features focused on quality review, and even some hints towards a future question of continual learning. The central new releases were all around Claude managed agents. Now, for those who don't remember, managed agents are Anthropic's way of allowing users to piggyback on Anthropic's infrastructure for all the ancillary services that make an agent work. The initial launch in April was about providing agents with a sandbox state management and error recovery, with one of the big changes being that user created agents could access a cloud computer rather than needing to operate solely on a local desktop session. That is, instead of giving your agent a Mac Mini to work from, you could just spin up a cloud instance on Anthropic's infrastructure. Now, very clearly, managed agents was one indication of the harness as a service moment that I called out in an episode a couple of weeks ago where the big labs are effectively taking a lot of the capability set of open highly configurable but highly complex tools like OpenClaw or Hermes and bringing that more easily into their native offerings. And interestingly, some of the features that Anthropic announced yesterday continue to take their cues from things that have been experimented with in that open space. The first big new feature is something that Anthropic is calling Dreaming, a quote, scheduled process that reviews your agent sessions and memory stores, extract patterns and curates memories so your agents improve over time. Now, Dreaming is essentially a memory management system of a fashion. It's effectively a scheduled memory review that runs between sessions and which, as Anthropic puts it, surfaces patterns that a single agent can't see on its own, including recurring mistakes, workflows that agents converge on, and preferences shared across a team. They continue, it also restructures memory so it stays high signal as it evolves. The core idea is to allow agents to not only deliver their completed task, but also to report what they learned while doing that task, allowing the system to encode those learnings in the orchestration memory to be preloaded the next time that sub agent or agent is called upon. Memories persist between sessions and should automatically improve agent performance the longer the system is in operation. Jan Kronberg wrote, agents that learn from past sessions and iterate until they hit quality enough is the architecture most teams have been trying to build manually. Dreaming seems to be the missing piece to that puzzle. VC Intern writes, think of it as the agent equivalent of REM sleep. Chirish points out that this is something that has attracted people to Hermes. They write that the Hermes agent reviews past conversations, builds skills from experience, has persistent cross session memory and gets smarter the longer it runs, which is very similar to what Claude Dreaming shipped. Jatin Garg writes, the underrated story of 2026 is that the open source agent ecosystem is leading on primitives noose research with Hermes for orchestration, GBRAIN for personal memory and eval substrates. These projects shipped working production systems before Anthropic shipped a research preview of similar functionality. The closed labs have raw model capability. The open source ecosystem has agent primitives. Those are different layers. The open source side has been further ahead on the second one for nearly a year now. Now in addition to dreaming, that is in addition to dealing with memory, Anthropic has also improved the oversight of managed agents with a feature called outcomes. Outcomes allow the user to write a rubric for what success looks like for a particular task. Once an agent completes a task, that output is scored by a separate grading agent against the rubric. The separation means that the grading agent isn't influenced by the reasoning of the task based agent but but instead looks purely at the output and scores how closely it fits with the provided rubric. If there's a problem, the grading agent can highlight the issues and kick the task back for another run. Anthropic has also added webhooks so users will be automatically notified when the task is complete. Anthropic said that in their testing using Outcomes improved file generation quality by 8.4% for Word documents and 10.1% for PowerPoint slides scored on their internal benchmarks. Now once again we are dealing with pretty core challenges of agents. One of the big shifts is that because agents can output so much now, human review becomes a bottleneck. Because of that, versions of external grading or external grading agents have been a fairly common part of multi agent system design for some time now. Most of these systems so far however have been deployed against coding tasks, with the grading agent doing things like automatically running unit tests. The benefit is that coding rubrics are typically well defined. A PR either works or it doesn't. The idea of subjective rubrics applied to non code knowledge work outputs is a lot less well developed. Like dreaming. One big impact of the feature will be to make the use of an external grading agent part of the default setup. Users won't need to string together a grading agent, they can just let Anthropic handle it behind the scenes. If a non technical user is designing a report generation agent, they'll use outcomes to automatically check and iterate on the output before delivery. A lot of the harness work right now is around systems that don't just shut down after first input, but through some formalized process. Whether it's loops or now, this outcome rubric based system can continue to refine and improve the work without the user having to sit there and manage everything. Finally, the managed agents platform can now handle multi agent orchestration. Anthropic writes that multi agent orchestration quote lets a lead agent break the job into pieces and delegate each one to a specialist with its own model, prompts and tools. Or for example, a lead agent can run an investigation while sub agents fan out through deploy history, error logs, metrics and support tickets. The agents work in parallel on a shared file system, with their work feeding back into the lead agent's overall context, and the lead agent can check in on the subagent's MID workflow to ensure they're still on track. The entire system can be tracked into CLAUDE console, allowing users to see what each subagent did and in what order. In addition, an explanation of the reasoning behind the task execution is auditable. The giving users visibility into the process. In short, write sif CLAUDE can now act like an AI worker. It can take a goal, run tasks on its own, use multiple agents, connect with other tools Anthropic included a few examples of the agentic systems people have built using managed agents. One of the more interesting ones to me came from Every with their Spiral writing. Agent Spiral is a tool that's meant to, in short, make AI writing not suck, which if you've ever tried to use AI for writing other than just generic business stuff, you know is no small task. Every Spiral uses a multi agent system, tapping into a range of different Anthropic models for cost optimization, and now they use this new outcomes feature to enforce writing quality. Every has defined their own rubric based on editorial standards and writer voice to ensure the agentic drafting is up to par, which is kind of the whole ball game for them now. It is also worth noting that prior to their dev day kicking off, Anthropic shipped a big suite of agents for financial services. On Tuesday, the company released a package of 10 predefined agents within CLAUDE Finance. The agents can be used as plugins for cowork CLAUDE code or deployed as managed agents. The suite includes a pitch builder, a meeting preparer, a market researcher evaluation reviewer, and a month end closer, among many others. The idea is to give financial services firms the starter pack of basic agents they need, rather than requiring a custom build alongside the agents. Anthropic released a full cookbook so users can understand how the agents work and go in and make modifications as needed. As part of the release, Anthropic highlighted a feature called Add Ins, which allows CLAUDE to work directly within productivity software. For example, instead of accessing Microsoft Word via MCP or a connector, CLAUDE can work directly in the program. This means that it has the software native context, such as your company's template for drafting docs or linked spreadsheets for building financial models. In Addition, Anthropic rolled out a series of new connectors for industry specific platforms, including Dunn and Bradstreet for business identity, fiscal AI for market analysis, and Verisk for insurance underwriting. Most of the commentary on Twitter was pretty much what you would expect from people trying to win clicks, basically claiming that with one fell swoop, Anthropic had killed another wave of AI startups. But in this case these agents are much more about replacing a bunch of the grunt work that was already semi automated through traditional software or outsourced. None of these are really attacking the high skill knowledge work, instead going after low skill repetitive task type of knowledge work Returning to Dev Day in addition to things that were actually announced, we also got a sneak peek at Anthropic's model training roadmap. During the opening keynote, Research Head of Product Diane Penn discussed what Anthropic is working on, highlighting three key features of future higher judgment and code taste, infinite context windows and multi agent coordination. Going back to this theme, that code with Claude Day was all about Anthropic addressing the big challenges of agents. Infinite context windows was the feature that got the most attention. The discussion was largely about whether this would just be an improved version of compaction, that is the process by which, as the context window fills up, the harness compresses it, leaving only the important details and opening up more space for the next part of the conversation, or whether it was some more fundamental research breakthrough. Penn's precise wording to some felt instructive, given that the word infinite was in quotation marks and that Penn explained that Anthropic is working on context windows that feel infinite. Some remain skeptical. Peter dedenne writes infinite context that must be rag wearing a trench coat, no? But Dan McAdir speculated on the significance. He writes, anthropic is hinting at infinite context windows. That matters because models already learn in context. If you can keep adding to the context window forever, the model can keep learning from experience forever. Some people will say that's not real continual learning, but that sounds a lot like saying reasoning models don't quote unquote really reasonable. At some point the functional distinction collapses. Infinite context means AI systems that continually learn, and when that arrives, it'll be much harder to deny that we haven't arrived at AGI. The two other little things of note from the actual event itself include Claude code creator Boris Czerny disavowing the term vibe coding. In a side interview, Czerny said that the term is starting to annoy him as it no longer describes the way that he and most other developers use AI. In a panel discussion on Wednesday's event, Czerny said that there's literally no manually written code anywhere in the company anymore. Instead, clauds coordinate with each other over slack code in loops and resolve issues across the code base. In that context, Czerny thinks the term Vibe is significantly underselling the system. Anthropic's workflows now include copious automated testing and verification to ensure that their code is ready to ship. Still, one challenge is that Boris doesn't have a replacement term to attach to the new process. While Andrej Karpathy, the corner of the Vibe coding term, has suggested the term agentic engineering, something about that still isn't sitting right with Boris. He says he's fielding suggestions, so if you have a good term for the way we use modern coding agents tweet it at him. The one other mic drop moment from the event itself came when Anthropic CEO Dario Amadei put some actual numbers around Anthropic's insane growth. Discussing the challenges of compute, Dario said, we planned for a world of 10x growth per year. In the first quarter of this year, we saw 80x annualized growth per year in revenue and usage. 80x in a single quarter. Now the context for those comments was a new SpaceX compute deal, through which Dario said, quote, we're working as quickly as possible to provide more compute than we have in the past. One of the most important AI questions right now isn't who's using AI? It's who's using it? Well, KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising the highest impact Users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com us sophisticated that's kpmg.com us sophisticated. Today's episode is brought to you by Granola. Granola is the AI notepad for people in back to back meetings. You've probably heard people raving about granola. It's just one of those products that people love to talk about. I myself have been using granola for well over a year now and honestly, it's one of the tools that changed the way I work. Granola takes meeting notes for you without any intrusive bots joining your calls. During or after the call, you can chat with your notes. Ask Granola to pull out action items, help you negotiate, write a follow up email, or even coach you using recipes which are pre made prompts. Once you try it on a first meeting, it's hard to go without. Head to Granola AI AIDAily and use code AIDAily. New users get 100% off for the first three months. Again, that's Granola AI AIDAily. Here's a harsh truth. Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized. Half of companies have AI tools, but only 12% use them for business value. Most employees are still using AI to summarize meeting notes. If you're the one responsible for AI adoption at your company, you need section Section is a platform that helps you manage AI transformation across your entire organization. It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value. The result? You go from rolling out tools to driving measurable AI value. Your employees move from meeting summaries to solving actual business problems and you can prove the roi. Stop guessing. If your AI investment is working, check out section@sectionai.com that's S E C T I O N A I.com so coding agents are basically solved at this point. They're incredible at writing code. But here's the thing nobody talks about. Coding is maybe a quarter of an engineer's actual day. The rest is stand ups, stakeholder updates, meeting prep, chasing context across six different tools. And it's not just engineers. Sales spends more time assembling proposals than selling. Finance is manually chasing subscription requests. Marketing finds out what shipped. Two weeks after it merged, ZenCoder just launched ZenFlow work. It takes their orchestration engine, the same one already powering coding agents, and connects it to your daily tools. Jira, Gmail, Google Docs, Linear Calendar Notion. It runs goal driven workflows that actually finish your standup Brief is written before you sit down. Review cycle coming up, it pulls six months of tickets and writes the prep doc. Now you might be thinking, didn't openclaw try to do this? It did, but it has come with a whole host of security and functional issues which can take a huge amount of time to resolve. Zencoder took a different approach. SOC 2 type 2 certified curated integrations, tighter security perimeter, enterprise grade from day one model agnostic and works from Slack or Telegram. Try it at ZenFlow free. Wait a second, what you're saying a SpaceX compute deal? Yes. After all of those interesting managed agents and memory and infinite context related announcements after about noon Eastern time yesterday, basically no one was discussing anything other than a new anthropic SpaceX partnership with a tweet sitting at a casual 20 million views at the moment, the Claude AI account wrote, we've agreed to a partnership with SpaceX that will substantially increase our compute capacity. They went on to specify how that would allow them to increase immediate term usage limits. But the TLDR of the deal is that Anthropic now has full use of Xai's Colossus 1 data center. Now, the XAI campus in Memphis consists of two main data centers, Colossus 1 and Colossus 2. Colossus 1, you might remember, was the data center that was built at record speed over a few months in mid to late 2024. It's since been scaled to contain 220,000 Nvidia GPUs, mostly H1 hundreds operating at a 300 megawatt capacity. Colossus 2 is Xai's Blackwell based cluster containing around 550,000 GPUs. The deal begins immediately with Anthropic stating the inference will be available within the month. Now, in terms of specifics, Anthropic is making three changes to deliver this new compute to users. First, Claude Code's five hour rate limit has been doubled for Pro, Max Team and SEAT based enterprise plans. Second, peak hour usage reductions for Claude Code will be eliminated for Pro and Max accounts. Third, Anthropic is raising the API rate limit for Opus models substantially. Output token throughput will be increased between 2x and 10x depending on account tier. Anthropic head of growth Amal Avasare explained the reasoning for these as the first moves and indicated that there was more to come. Amal wrote only a very small percentage hit weekly limits while a much larger portion of users hit the five hour limit. So we fixed that first. As the compute comes online we will look at weekly. Now for most people the first speculation was all about how this could have possibly come together. While of course his bigger ire is saved for Sam Altman in OpenAI. Elon has not been a big fan of Anthropic as well. He has frequently on Twitter referred to them as misanthropic and said that there's no scenario in which they win. So how did things change? In a tweet Elon wrote by way of background for those who care. I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed. Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector. So long as they engage in critical self examination, Claude will probably be good. After that, I was okay leasing Colossus 1 to anthropic as SpaceX AI had already moved training to Colossus 2. Now, believe it or not, as unexpected as this tie up seems, if for no other reason than the personalities involved, some had recently been speculating that it was in fact a perfect mashup. On the all in podcast, Chamath Palihapitiya recently explained how power constraints would give Elon leverage to make AI deals. Referring to all of these massive data center projects, Chamath said, less than half of it is actually being built. Most of it is stuck in red tape. There's no credible strategy to turn any of this stuff on. Who will this hurt? He asked. It will hurt Anthropic and OpenAI the most. Who will this benefit? It'll benefit the hyperscalers, specifically Oracle, Amazon, Meta, Microsoft and Google. And now, chamath continued, what you're going to see is a negotiation and a trade back and forth. How much equity do I have to give up? How much control do I have to give to get access to the compute? How badly will I miss my growth forecast if I don't? And then the money shot? That's a huge lane, chamath said, for Grok and SpaceX to run through because they have a ton of excess capacity. If I were Elon now, I'd be running all over this market because if the models catch up in quality, I think he could also do something really crazy with Anthropic. He and Dario should do a deal tomorrow. And from a business perspective, there's twin contexts that make Anthropic and Xai a pretty great match. Anthropic has of course been straining under a compute crunch for all of this year, massively degrading the user experience. I was literally complaining right before the announcement that there wasn't a day that went by that I could actually just use Claude without interruption. Anthropic has an extremely compelling model and harness combination, but OpenAI has recently been taking advantage of Dario and Anthropic's underinvestment in compute to start reclaiming some of the space that Anthropic has opened up this year on the Elon side of the ledger, something had to give with XAI as well. Even before the SpaceX merger, things were heading south. Model improvement had completely stalled out with the release of Grok 4.2 in February, gathering effectively no real buzz. The company has no meaningful agentic harness product to compete with Claude Code and Codex, and people on X have even stopped asking Grok if this is true. The personnel story was also not optimistic. Each co founder left one after another over the past year, leaving Elon as the last man standing. And around rumors of huge staff turnover, Elon acknowledged the company was not built right the first time and needed a total rebuild. Even the reclamation projects didn't seem to be working out. The cursor deal announced last month was heralded as the saving grace, but the information recently reported that there's no plans to co develop a coding model in a piece that they framed as cursor keeping its distance from Xai. And yet what XAI does have is a warehouse full of GPUs with too little to do. So as Derek Thompson put it, musk has compute capacity but a meh model. And Anthropic has a fantastic model with weak capacity. And thus a new alliance is born. Now if you'll give me a minute to speculate a little bit farther, I think it's interesting to broaden out this conversation even beyond the very obvious and specific reasons for that tie up. I think Elon has two things going on simultaneously. First, in Elon world, it seems to me that he's long wanted the one company to rule them all. It's always felt like over time there would be some inevitable realignment between all the pieces of the Elon Empire. I mean, the man puts X in all the names, at least in part, so they could be easily recombined. Now, for a long time the obvious bet as the leading entity was Tesla. Which is why I predicted last year that if XAI couldn't really break out of its very behind position, I thought Elon would end up folding XAI in with Tesla. Now there was this weird little X factor. See what I did there for a while? Of whether XAI itself could somehow surge and become the one to own them all. The only reason that was even possible was the clear recognition of the significance of AI relative to all the other industries. And for a minute there was a hint that maybe that would be the direction as XAI folded in X. But obviously reality intervened and the one company to rule them all in the Elon Empire was going to be something different now. Second, Elon has for a long time been determined to have an outsized hand in shaping AI, which to him is absolutely not about making more money and much more about him thinking he needs to be involved for the sake of humanity. And there have kind of always been examples of these two goals potentially intersecting. I'm thinking especially of him wanting to fold OpenAI into Tesla early on. So how do we get to SpaceX being the absorber rather than Tesla? And what does it suggest about what SpaceX is actually going to become now? First note, I wouldn't go so far as to predict that Elon will absolutely continue to drive consolidation into one company. He may ultimately be fine with Tesla and SpaceX as big and separate, with smaller things like boring off on the side. But there are some inside AI and outside AI reasons why I think SpaceX has started to make more sense. And as one of the crown jewels first, Tesla has stalled at least a little bit. Fully autonomous driving is really, really hard technologically. And it also has major barriers outside of technology in society and politics and consumers that no matter how good Elon is at building stuff, he can't just force his way through. I also think the fact that Tesla has Optimus creates an easy path for future tie ups, making it less essential to do the tying up now. Now, in terms of SpaceX itself, it's kind of gotten clearer and clearer, especially as the demand for tokens has started to so dramatically outstrip the supply that king making in AI was in many cases going to be about compute. When Elon announced the merger of SpaceX and Xai, one of the big things he talked about was his vision of future orbital data centers. I think they are to him much more than a market narrative. I think he actually sees it as a key part of the future, both of the company SpaceX, but also of the world. And in that light, the fold in of XAI might have been less about giving SpaceX a model in Grok and more about giving SpaceX a footprint in terrestrial compute and supercomputers that it could then build upon up from the earth into the sky. The point is, while everyone was Talking about the SpaceX XAI tieup in either very mechanical, bailout type terms, or seeing it as SpaceX somehow having a connection to a model like Grok, my argument is that maybe it actually had nothing to do with models and was always about compute positioning. Basically, I think that Elon started to realize that his best path to influencing the shape of this most important industry was being akin to Jensen Huang as opposed to being akin to Sam and Dario. And if I'm right and that is the way he started to think about it, there was then literally no question of who he was going to work with. It was Anthropic or bust. Hold aside the leans right, leans left, woke anti woke politics. I do think that Anthropic's extremely disciplined and focused approach, you could argue, is more aligned with how Elon builds things, at least within the context of specific companies. And secondly, and more importantly, obviously the Sam Altman feud is so deep that Anthropic was the only option. Now I don't think that Elon is going to abandon Grok immediately. I think he's going to leave optionality around Grok. GROK will remain an option because a x Twitter has to have something like this integrated and there are benefits to owning it and b it also gives them more options when it comes to Optimus as embodied robotics mature. Still, I think that we are seeing a pretty clear and full pivot as part of this, Elon even tweeted. XAI will be dissolved as a separate company, so it will just be SpaceX AI the AI products from SpaceX. Effectively I think we're seeing Elon's AI play 3.0 1.0 was as OpenAI funder 2.0 was as model builder, 3.0 is as compute czar and a lot of people are really bullish on the shift, rohit writes. Elon's extraordinary hardware genius shows up again. He fumbled the model but built a NEO cloud that's highly competitive and works great for Frontier Labs, rohit added. For what it's worth, I pointed this out four years ago that Elon's unique talent is suited better to some things than others. Getting a NEO cloud up and running is a known but hard thing to do. Getting a model to be as good as the Frontier Labs is an unknown and hard thing to do. In that essay Rohit had written, Elon looks at something he wants to accomplish and as long as existing knowledge is able to create what he wants, theoretically acts as an individual shelling point to coalesce money and talent around to create them. Rohit continued, though things that he has not done for which he gets flak are areas which are not purely dependent on doer energy. These are things that require thinkers and some sort of step change in our ability. Alas, we know of no way to throw resources at one end and get thinkers at the other end. Derek Thompson wrote, I don't think I've seen this take before, but I like it. Musk has been world leading at compressing money, resources and time to make known but hard things at scale. But he's less than world leading at cracking open breakthroughs in more unknown spaces. So it would make sense that XAI is lagging the Frontier Labs on new AI agents, but also that he'd have built a NEO cloud to power those models once they run short of compute. Dean Ball writes, I would be very excited about an XAI SpaceX as an AI infrastructure firm. Elon's great strength, where he is truly goaded, is building things in the real world. Colossus came online faster than anyone expected. Huge asset for America. As Aaron Levy Simply put it, SpaceX as a vertically integrated AI compute company makes an insane amount of sense. Now, to the extent that anyone has concerns, it's that consolidation and fewer players in a market is does have real consequences. But by and large the ghost of David Ricardo is celebrating, as everyone remembers, the incredible power of comparative advantage. Good thing Elon won't be wasting COMPUTE on random dead ends anymore and Anthropic won't be nerfing Claude like it's a hobby. And I think maybe the best summary of how everyone feels comes from Chubby who writes, okay Anthropic, show us what you could do with 220,000 Nvidia GPUs and 310megawatts. Sometimes everyone talks about things because they're interesting, juicy tabloid style things to talk about. And unfortunately a lot of the discourse around Elon Musk can fall into that category. This is not one of those. This is a massive deal that has the potential to significantly reshape the face of the AI battle. So what happens next? Will OpenAI respond with their own deal? Will Mark Zuckerberg swoop in as another Elon style compute capacity kingmaker? No one knows for sure, but boy oh boy, there is never a dull day in AI land. For now, that is going to do it for the AI Daily Brief. Appreciate you listening or watching. As always and until next time, peace.