AI's Battle for Your Context

0:00Today on the AI Daily Brief, why the biggest battle in AI is the battle for your personal context. Before that, in the headlines, could this be the biggest year for IPOs in history? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:21All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, Assembly, Landfall, Zencoder, and Superintelligent. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts. If you're interested in sponsoring the show, you can send us a note at sponsors at AIDailyBrief.ai. And finally, also at AIDailyBrief.ai, you can find out about all the other goings on in the world of AIDB, including AIDB Intel, our new operators community, our AIDB New Year's self-guided skills course, and the new AIDB Intel service, which will have much more info coming in the next week or so.

0:56But with that out of the way, let's dive in. Welcome back to the AIDaily Brief Headlines Edition, all the daily AI news you need in around five minutes. We kick off today with two stories that I discussed in my 2026 predictions, one where it looks like I might be wrong, one where it looks like I might be right. The one where I might be wrong is that contra to my base case that neither OpenAI or Anthropic ultimately go public in 2026, the New York Times finds a lot of evidence that they are getting ready. They wrote, Anthropic and OpenAI have taken early steps to go public, people familiar with the companies said.

1:27And SpaceX, Elon Musk's rocket company, has interviewed banks to lead an IPO, according to two people with knowledge of the situation. Now, these three companies are already valued between$350 and$800 billion each. Add in a premium for the public offering, and we could easily see multiple trillion-dollar IPOs this year. That is extraordinarily rare. The only real comparison at those levels are the$1.7 trillion valuation for the Saudi Aramco IPO in 2019. No tech startup has ever come close. Morgan Stanley's Eddie Malloy said, We're going to get into a period of potentially unprecedented IPO deal sizes, but we are confident they're executable given the scale of these companies and the investor interest.

2:05Now, Malloy in this case is referring to concerns that the public markets can't handle deals of this size. As part of all of the AI bubble chatter, there's been talk that investment banks might force existing shareholders into a rolling unlock rather than the more usual six-month cliff to stagger the selling. Others expect demand to be so hot that selling won't be an issue. Certainly, it is the case that despite claims that the current stock market is starting to resemble the dot-com boom, we haven't seen anything comparable to the thousands of IPOs for emerging tech companies that mark that era.

2:32What that means is that for most investors, this would be the first time that they got access to pure play companies developing AI. Jeremy Abelson, a VC at Irving Investors, said, In two decades, I haven't seen private companies that are this meaningful and are this impactful. Not only are they bigger and more relevant, but they're incredible companies with numbers that we've never seen before. You can expect analysis of these big three to dominate finance discussions over the coming year, with very loud opinions on both sides of the argument. Some compared this moment to the record-breaking Facebook IPO in 2012.

3:01That listing both solidified social media as a major category and was also seen as an utter catastrophe in the market, with the Wall Street Journal calling it a fiasco. The stock was down 15 % after a week and took 14 months to trade above its IPO price. At the time, it was the third largest IPO in history. Although in the current frame of reference, its$90 billion valuation was exceedingly modest. Jeff Thomas, the head of listings at Nasdaq, said, When these mega deals happen, it takes some of the air out of the room. You want to try to get ahead of it. Others noted that transparent information on the leading AI companies could diffuse the bubble chatter.

3:33Said Notable Capital's Jeff Richards, There is such a big information gap right now. The biggest positive for this entire market would be if a bunch of these companies went public and people could actually see the numbers. Now, my argument for why I didn't think that OpenAI and Anthropic would ultimately go public this year is A, public market reporting is a total pain in the butt, especially when they're in the category that is already the most under scrutiny, and B, I think there is plenty of capital still available in private markets for their financing needs. Now, B might be more dubious that I'm giving it credit for, not because there isn't private capital, but just because the scale of the need might be so huge.

4:06And there is also, of course, a competitive dynamic thing. There are lots of good reasons, for example, for Anthropic to want to scoot in ahead of OpenAI, and lots of good reasons for OpenAI to not want to let that happen. For that reason alone, we may be headed into a very big year when it comes to public markets. Now the prediction I did a little bit better on then is one where I said that Anthropic was going to continue to be hard to displace when it came to its lead in coding, and that I thought that Microsoft was likely to get much closer to Anthropic over the course of the year. According to the information, Microsoft has indeed quietly become one of Anthropic's largest customers over recent months.

4:38As of July last year, Microsoft began using Anthropic to power coding agents in GitHub Copilot. But the big shift began in September, when OpenAI and Microsoft agreed to the amicable end of their exclusive partnership. Microsoft quickly announced they would add support for Anthropics models within their Copilot products. The new multi-model version of Copilot routes each task to the most appropriate model. And for many of the tasks in Microsoft's productivity suite, the right choice seems to be an Anthropic model. The report stated that Claude Sonnet 4.5 has a 15 % performance advantage over GPT-4.0 in agent mode for complex Excel tasks, although why the hell anyone's using 4.0 is beyond me, while the super long context window of Claude Opus 4.1 is being used for mass summarization and analysis tasks.

5:17Haiku 4.5 is also seeing heavy use due to its cost and speed advantage for smaller tasks. Business customers didn't have to upgrade or change anything in their plans, but as of last week, they're now receiving access to Anthropic models by default. The information reports that Microsoft has spent more than$40 million per month with Anthropic starting in July, a$500 million annualized pace that is likely a lot higher now, given that the models are seeing more use. In addition, the report states that Microsoft Cloud staff have been incentivized to sell Anthropic products. And finally, deepening the new ties, Anthropic will reportedly work with Microsoft to develop new cloud power features for Copilot over the coming months.

5:51A liquid on Twitter said what I think a lot of people feel, Anthropic and Microsoft was the partnership that made sense all along. Speaking of partnerships and OpenAI, chip startup Cerebrus has landed a$10 billion compute deal with the company. The three-year deal will see Cerebrus provide OpenAI with 750 megawatts worth of AI inference compute. A press release said that OpenAI planned to integrate Cerebrus' chips into their broader computing network to provide faster response time. Cerebrus CEO Andrew Feldman posted, This has been a decade in the making. Deployment begins in early 2026, and when fully rolled out, it will be the largest high-speed AI inference deployment in the world.

6:26He claimed that Cerebrus' uniquely designed chips are now able to deliver 15x faster inference without sacrificing model size or accuracy. He added, as models grow more capable, speed becomes the bottleneck. Slow systems limit what users can do, how often they engage, and whether AI becomes infrastructure or remains a novelty. Matthew Berman wrote, I've always wondered why OpenAI didn't use Grok or Cerebris. They are so fast. Now we know why Grok was bought by NVIDIA. Everything is moving to specialized chips. Revenue is made at inference. ChatGPT is about to be a hundred times faster. Now, lastly today, we stay on OpenAI, but move to some serious industry psychodrama, a trio of leading AI researchers are returning to OpenAI amid allegations of corporate espionage?

7:07On Wednesday night, we had dueling tweets. Former OpenAI CTO and now CEO of Thinking Machines Labs, Mira Murati, wrote, We have parted ways with Barrett Zoff. Sumit Jantala will be the new CTO of Thinking Machines. He's a brilliant and seasoned leader who has made important contributions to the field of AI for over a decade, and he's been a major contributor to our team. We could not be more excited to have them take on this new responsibility. Meanwhile, about an hour later, OpenAI CEO of Applications Fiji Simo tweeted, excited to welcome Barrett Zoff, Luke Metz, and Sam Schoenholtz back to OpenAI.

7:35This has been in the works for several weeks and we're thrilled to have them join the team. Barrett will report to me, Luke and Sam will report into Barrett. More to come on what they'll focus on soon. Now, by way of background, these three left OpenAI in late 2024 alongside CTO Mira Mirati as part of a mass exodus of talent. They were pivotal in the subsequent founding of Marati's Thinking Machines Lab. Zoff and Metz were in fact listed as co-founders of TML, with Zoff also receiving the CTO title. This is a significant personnel move that could have implications for the course of both companies.

8:05Zoff first joined OpenAI in 2022 to serve as their VP of research. Prior to that, he was at Google DeepMind for six years. At OpenAI, he built the post-training team from scratch with John Shulman, who also left to co-found TML. That team yielded the O1 model and helped kickstart the new reasoning paradigm. Metz and Schoenholtz are also leading experts on post-training and reinforcement learning. Now, for TML, it's very difficult to know how bad a sign this is. The departure of two co-founders and another senior researcher obviously isn't a great indication of how things are going. Anonymous Twitter poster Signal wrote, So like, thinking machines completely imploded today?

8:36Someone DM me the tea, please. And yet, at the same time, adding to the intrigue, Kylie Robison of Core Memory reported the story with a different twist, writing, Thinking Machines has terminated its CTO, Barrett Zolf, due to unethical conduct, according to two sources familiar with the matter. CEO Mira Mirati announced the news that had all hands with employees today. Max Zeff followed up with this angle for Wired, writing that his sources at TML said that Zoff had shared confidential company information with competitors. The timeline was laid out in a memo written by Fiji Simo on Wednesday and shared with Wired.

9:05Zeff wrote, According to the memo from Simo, Zoff told Thinking Machine CEO Mira Mirati on Monday he was considering leaving. He was then fired on Wednesday. Simo went on to write that OpenAI doesn't share the same concerns about Zoff as Mirati. I don't know, man. Obviously, from outside, it's hard to tell exactly what's going on. But from a sheer talent perspective, you got to think it was a good day for OpenAI. That, however, is going to do it for today's headlines. Next up, the main episode.

9:31If 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 Cluey. 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.

10:07It'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.

10:41Let's AI execute the repeatable while attorneys elevate to create the exceptional. Landfall isn't adapting to AI. They were built for it. Have a new idea? Try the Discovery Agent for free. It's a confidential tool that helps innovators synthesize their inventions and instantly see patentable insight. Visit landfallip.com to learn more. That's landfallip.com. If you're using AI to code, ask yourself, are you building software or are you just playing prompt roulette? We know that unstructured prompting works at first, but eventually it leads to AI slop and technical debt. Enter Zenflow. Zenflow takes you from vibe coding to AI-first engineering.

11:19It's the first AI orchestration layer that brings discipline to the chaos. It transforms freeform prompting into spec-driven workflows and multi-agent verification, where agents actually cross-check each other to prevent drift. You can even command a fleet of parallel agents to implement features and fix bugs simultaneously. We've seen teams accelerate delivery 2x to 10x. Stop gambling with prompts. Start orchestrating your AI. Turn raw speed into reliable production-grade output at zenflow.free. Today's episode is brought to you by Superintelligent. Superintelligent is a platform that, very simply put, is all about helping your company figure out how to use AI better.

11:57We deploy voice agents to interview people across your company, combine that with proprietary intelligence about what's working for other companies, and give you a set of recommendations around use cases, change management initiatives that add up to an AI roadmap that can help you get value out of AI for your company. But now we want to empower the folks inside your team who are responsible for that transformation with an even more direct platform. Our forthcoming AI Strategy Compass tool is ready to start to be tested. This is a power tool for anyone who is responsible for AI adoption or AI transformation inside their companies.

12:27It's going to allow you to do a lot of the things that we do at Superintelligent, but in a much more automated, self-managed way and with a totally different cost structure. If you are interested in checking it out, go to aidailybrief.ai slash compass, fill out the form, and we will be in touch soon.

12:45Welcome back to the AI Daily Brief. Today, we are talking about Google Gemini's big upgrade that they are calling personal intelligence. Yesterday, they announced the very obvious and yet very useful ability to connect Gemini with all the information from other Google apps that you interact with like Gmail, Photos, Search, YouTube, all of course in an effort to help make Gemini more personalized for the individual user. What's interesting is that I believe that at core, you can view almost every single move being made in and around consumer AI as in some way a battle for personal context. So let's look at what I mean.

13:19Big news from earlier this week was the announcement of Claude Cowork. It's basically Claude Code, but simplified in a way that it's designed for non-technical users. You don't have to deal with the terminal anymore. It lives right inside your Claude desktop app, and it allows you to do the types of things that people have been using Claude code for outside of coding. Now, the big thing that has made Claude code and now Claude co-work powerful is that it has access to a unique set of context, which is the stuff on your desktop. What makes it different than just the Claude chat window or the chat GPT window or the Gemini window is that instead of having to upload the context that's relevant for any particular thing you're to do, you just point it at the relevant part of the computer.

13:56Now, of course, in addition to having that better context, coworking cloud code can also do things and interact with your desktop, making it more agentic, but that power comes from its ability to access everything on your computer. And yet, even with that, a lot of the issues that people have discussed when it comes to cloud cowork over the last few days, which admittedly are more likely having to do with the fact that it was built in the 10 days previous, isn't around connecting other types of context. While Claude Cowork and Claude Code have access to what's on your machine, if you live in the modern world, there's going to be lots of other data sources and places where your data lives that are not just on your desktop.

14:31And for that, Claude gives you access to things via what they call connectors. Connectors are ways to link things like Google Drive, obviously powered by the Model Context Protocol. And in the first couple of days after Claude Cowork went live, a lot of people's challenges have been in and around making those connectors work. The point being in some ways that we are so hungry for personal context that just having access to our full computers isn't enough. We still need access to everything that exists on the web as well. So, okay, we've repositioned CloudCode work in CloudCode as powerful because of the way they give you unique access to your desktop context.

15:01How are the other things that AI companies are launching right now also in some way about this battle for personal context? I would argue that when it comes to ChatGPT, a huge anchor to their strategy has always been to try to leverage the fact that because ChatGPT was many people's default, it has a huge amount of personal context in the form of past chats. And when you view everything in the battle for personal context, all of a sudden OpenAI's strategy to add more and more applications all of the time with an incredible shipping velocity starts to make a little more sense. They are trying with each new app release to get more personal context, which makes the switching costs of leaving and going to another AI service more and more costly in terms of that lost context.

15:40For months now, folks have been talking about how memory is the next big moat, and I think that that's dead on. Now, bringing it back to things that have been released recently, so far from OpenAI, the biggest product that we've got in January is the introduction of ChatGPT Health. It's a dedicated health experience inside the app whose entire purpose is to collect a huge amount of personal health context and organize it in a single place that makes it accessible to the AI. In their announcement post, they wrote, Today, health information is often scattered across portals, apps, wearables, PDFs, and medical notes, So it's hard to see the full picture, and people are left to navigate a complex healthcare system on their own.

16:15Now, as they point out, people are already using ChatGPT to help navigate all this, but now they're allowing you to port all of that context in. And they are really trying to pull that health context from everywhere it lives. Just a few days later, we got Anthropik's answer to that in their Claude for Healthcare. A big part of that Claude for Healthcare announcement was about connecting personal health data. The announcement came with a bunch of new connectors rolling out specifically for that type of personal context. I would even argue that Grok's big play, outside of having Elon for fundraising and for building the biggest supercomputers in the world, is once again around personal context.

16:50The unique personal context that Grok has access to is everything that happens in and around X slash Twitter. Which for those of you who aren't on X slash Twitter might not seem like it matters, but for those of us who are and who have been for a very long time, is a very significant part of personal context. Okay, so now you're starting to see all of these different moves through the lens of personal context, but Google's latest announcement is in some ways the clearest yet. Yesterday, CEO Sundar Pichai tweeted, Answering a top request from our users, we're introducing personal intelligence in the Gemini app.

17:20You can now securely connect to Google Apps for an even more helpful experience. Personal intelligence combines two core strengths, reasoning across complex sources and retrieving specific details, e.g. from an email or photo to provide uniquely tailored answers. It's built with privacy at the center, and you choose exactly which apps to connect, with the connected app settings off by default. Some of the examples that Google gives about how this might be useful are really concentrated on day-to-day life. This is not about work. In their announcement thread, they wrote, Ever need to buy parts for your car but don't have the info handy?

17:48Ask Gemini to recommend tires for my car. By referencing connected apps like Gmail and Photos, it can understand your car's make and model and even the types of trips you take, to give recommendations of tires and info like your license plate number, to make your visit to the auto shop go more smoothly. when a user asks for recommendations around travel. Instead of it being generic lists, the specific travel dates that can be found in Gmail, plus other evidence like in their example, your love for nature photography found in Google Photos, lead to more personalized recommendations. People's first instinct was that this was a big deal and that in many ways it was inevitable, but kind of a killing blow play from Google.

18:20AI YouTuber Matthew Berman writes, Gemini will now be my daily driver AI for the next few weeks, all because of personal intelligence. Google would have never allowed this kind of feature to release just 18 months ago. They would have been too nervous, too much red tape. But now they got out of their own way and allowed users to choose. Google is so well positioned to win AI. Apple, where you at? Akash Gupta writes, Google just revealed the AI moat nobody can replicate. Every AI company is racing to build memory and personalization. Google connects to a decade of your Gmail threads, every photo you've ever taken, your complete YouTube watch history, and every search query you've made since 2005.

18:54The question for every other AI company, how do you compete on personalization when your competitor has the user's entire digital life and you're starting from a blank conversation. I think there are a couple answers to this. First of all, I do think it's important to note that while it does seem obvious that this would make AI better for a variety of use cases, I don't think we yet have enough evidence to know exactly the full complexity of the way that AI gets used over time. To be clear, I am far from the average consumer and user of AI. And yet, I do represent a type of user of AI, and I couldn't care less about this if I tried.

19:27For my work-related use cases, I care about the quality of AI strategic thinking, its ability to process and articulate multiple angles around the same decisions, how good it is at accessing other types of data, how good it is at analyzing types of data I give it access to, how good it is at building the things that I need. There's not a universe in which I'm switching models because I can get better travel recommendations or need a shortcut way to figure out what my license plate is. And to be clear, this is not at all a knock on these new features from Google, nor an argument that I'm anywhere near the normal consumer.

19:55My point to solely that when it comes to these big, bold claims that Gemini is killing everyone because of this, I think there's going to be a lot of types of different AI users, all of whom have different types of priorities. Still, let's assume that this type of personalization is really valuable for many, if not most, consumers. Well, another path is to ask who else has access to that data, which brings us back to Matthew Berman's question, Apple, where are you at? When Apple announced Apple Intelligence way back when, it was all with this same argument. The pitch was simple, helpful day-to-day use cases that took advantage of the context that Apple had about you because it powers all of your devices.

20:31Now, obviously, it has not delivered on that promise. One of the big takeaways after Google I.O. last year, in fact, was that Google had basically shipped everything that Apple's AI wanted to do. And now, of course, Gemini is going to actually power Apple intelligence. And yet, Apple still does have an enormous amount of personal context that others don't have. Google, for example, does not have your iMessages. And for iPhone users, iMessages tend to represent dozens of gigabytes of personal context that is extraordinarily valuable. And frankly, when it comes to a personal level, more valuable for many use cases than the stuff that's in your Gmail.

21:05Apple also has something else, ownership of devices that operate in the physical world. And I think when you start to view everything through the lens of this battle for personal context, OpenAI's hardware decisions start to make a little more sense. Hardware allows them to go after a very specific type of personal context, which is the personal context of how you interact in the physical world. One thing that Apple did last year that did capture people's attention was the new live translation feature they announced for Apple AirPods. Unlike many other form factors for AI devices, AirPods are something we already interact with.