The AI Daily Brief: Artificial Intelligence News and Analysis - AI Context Gets a Major Upgrade

Starting point is 00:00:00 Today on the AI Daily Brief, AI context, especially for business users, gets a major upgrade. And before that, in the headlines, the biggest debt deal yet for AI infrastructure. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick notes before we dive in. First of all, thank you to today's sponsors, KPMG, assembly, robots and pencils and super-intelligent. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe at Apple Podcasts. And to learn about sponsoring the show, shoot us a note at sponsors at AIDailybrief.aI.

Starting point is 00:00:41 We are now about halfway full for Q1, so as you are starting to do your 2026 planning, let us know we would love to help. Welcome back to the AI Daily Brief Headlined Edition, all the daily AI news you need in around five minutes. We kick up today with an update in the AI infrastructure space. A new record debt deal is coming together to fund Oracle's project Stargate infrastructure. Bloomberg reports a consortium of banks are putting together a $38 billion debt. deal, which will be the largest financing deal to date for AI Infra. The deal is split into two

Starting point is 00:01:12 tranches, 23.25 billion associated with a project in Texas, and a $14.75 billion package that will fund a data center in Wisconsin. Both data centers are being developed by Vantage data centers and will be operated by Oracle to provide compute for OpenAI. The institutions underwriting the deal include a laundry list of the world's largest banks, including JP Morgan, Wells Fargo, Goldman Sachs, Society, Hanna-Ral, Mitsubishi, UFJ. The banks will sell the debt onto high net worth clients, private credit firms, and pension funds. Now, right now, data center debt is red hot, so there will likely be no shortage of buyers to snatch up the record-breaking deal. Earlier this month, META closed a $27 billion deal with PIMCO as the major buyer, and that

Starting point is 00:01:51 debt surged once it started trading on public markets, making PIMCO $2 billion in paper gains. The deal also gives us some insight into how data center financing is being structured. Both tranches have four-year maturities with two one-year extension options. sources said they're priced at 2.5 percentage points above the benchmark, so likely between 6.5 and 7% interest rates. Now, there are a lot of interpretations of this. Of course, the people who want to see an AI bubble say, OMG, look at the size of that debt deal, debt is coming in, that must mean it's bad. There's a real tyranny of big numbers for those folks. On the other hand, are people who realize that at least right now, there is enormous demand in the markets for this

Starting point is 00:02:28 sort of debt. Private credit just has a voracious appetite for this, and so to them, in short, this is fine. Endgame Macro notes that it is part and parcel of a larger paradigm shift. They write, This is about the new arms race and AI infrastructure and who can lock in the physical and financial foundation of that ecosystem first. Data centers have become the modern versions of oil fields. Whoever controls the power, cooling, and fiber capacities controls the economy that runs on them. Oracle's $38 billion debt sale is an attempt to seize that ground before it gets fully priced out. They continue, Oracle is using leverage to buy its way to the front of the line, converting future AI workloads into guaranteed bond-financipal cash flow.

Starting point is 00:03:03 It's turning the data center business into a quasi-utility model with stable, contracted revenue in exchange for enormous upfront CAPEX. They also point out something else important. There's a deeper macro signal here, too. While the government is issuing hundreds of billions in treasuries, private credit markets are happily absorbing corporate infrastructure debt like this, that tells you investors are betting that AI-driven demand will hold up, even if the broader economy slows, that the cash flows from training and hosting large models are the new safe assets.

Starting point is 00:03:31 Ultimately, that means they write, the deal represents two overlap. forces, the financialization of compute, and the monopolization of digital energy. Oracle is trying to own the pipes and power sources that the next economy will run through, and by funding it with record debt, it's making a massive bet that AI demand will become the backbone of global economic growth itself. Next up, staying in the realm of infrastructure, Anthropic and Google have announced a massive new AI compute deal. Anthropic will expand their use of Google Cloud servers, including up to a million of Google's TPUs. The deal is expected to be worth tens of billions of dollars and add over a gigawatt of capacity for Anthropic next year.

Starting point is 00:04:05 The news is in some ways a coming out party for Google's range of TPU chips, which are one of the more credible competitors for Nvidia's industry-leading GPUs. TPUs or tensor processing units are special-purpose chips that can only be used to run AI models and other machine learning applications. In contrast, GPUs are much more general purpose, so have some future-proofing if new incompatible architecture arises. Google is currently ramping up production for their seventh-generation TPU, However, until recently, the chips were largely for internal use only.

Starting point is 00:04:34 They have been available to rent through Google Cloud, but this is the first major deal to establish a dedicated TPU cluster for an outside client. Theoretically, the benefits of a TPU over a GPU are speed and cost. The special purpose design is more efficient, but until now, Google has been a generation or two behind NVIDIA, meaning the tradeoffs didn't stack up in practice. With the new 7th generation Ironwood chip, that gap appears to be closing. The specs suggest they'll be on par,

Starting point is 00:04:58 or even better than Nvidia's latest generation Blackwell chips on performance, speed, and cost, but of course we'll actually have to see how that holds up in the real world. In their announcement, Anthropic was careful to note their continued commitment to Amazon as their primary training partner and cloud provider. They also reconfirmed their work on Project Rainier, a multi-facility compute cluster that will contain hundreds of thousands of chips. Analysts believe this could be the first tentative step for Google to market their chips for sale. With Bloomberg intelligence writing,

Starting point is 00:05:24 Google's deal with Anthropics suggests more commercialization of the former's TPU beyond Google Cloud to other neoclouds. Shea Belure writes, Google just locked Anthropic into the largest TPU expansion ever, which ties a $7 billion revenue run rate directly into Google Cloud and adds a gigawatt of compute capacity by 2026. The takeaway is Google is turning TPUs into the profit engine of its AI business. Microsoft's CEO, Sotia Nadella,

Starting point is 00:05:49 has published his annual shareholder letter, and in case you are wondering, AI remains at the very center of the company's vision of the future. Nadella wrote, imagine a world where every person can get help from a researcher, a coder, or an analyst on demand, not just information, but deep contextual expertise paired with action, or where every organization, no matter its size or sector, can reinvent employee experiences, reimagine customer engagement, reshape business processes, and bend the curve on innovation for their people, businesses, and industries. This is the new frontier and how we will unlock

Starting point is 00:06:19 the next level of productivity and growth for the world. The only reason that this is notable is that there are a number of analysts who are trying to basically suggest that Microsoft is nudging away from AI, given that they've been less enthusiastic about the infrastructure buildout than some of their competitors. And to the extent that you think that that's the case, this letter should disavowalue of that notion. Microsoft very clearly sees its future as AI, period, full stop. Lastly today, a really cool example of why vibe coding is not just for prototypes. The Kingdom of Jordan is partnering with Replit on an AI learning assistant. A pilot of the assistant known as Saraj is already live and has conducted more than 600,000 interactions with more than 100,000

Starting point is 00:06:58 students and teachers. Once the pilot is completed, the plan is to make the assistant available to all 1.6 million students in 90,000 teachers in Jordan's public schools. Essentially, this is a tool to enable self-directed learning as well as a quick reference guide for teachers. Suraj interacts with an Arabic language interface, requires no technical background to use, and functions like a conversational search engine. What's really cool about this, though, and why it's worth shouting out here, is that the pilot version of this was put together in less than a month by a single person. Amr Abulaila

Starting point is 00:07:28 who is a member of the National Council for Future Technology. Abolaila said, building a project of this scale from scratch would have taken months of development. Using Replit enabled us to prototype and deploy the learning assistant in a fraction of the time, effectively transforming vision into reality.

Starting point is 00:07:42 So no, as it turns out, vibe coding is not just some fly-by-night trend. For now, that's going to do it for today's AID Daily Brief Headlines edition. Next up, the main episode. What if AI wasn't just a buzzword, but a business imperative? On You Can with AI, we take you inside the boardrooms and strategy sessions of the world's most forward-thinking enterprises. Hosted by me, Nathaniel Wittamore, and powered by KPMG,

Starting point is 00:08:10 the seven-part series delivers real-world insights from leaders who are scaling AI with purpose, from aligning culture and leadership to building trust, data readiness, and deploying AI agents. Whether you're a C-suite executive, strategist, or innovator, this podcast is your front-row seat to the future of enterprise AI. So go check it out at www.kpmg.org.us slash AI podcasts or search you can with AI on Spotify, Apple Podcasts, or wherever you get your podcasts. If you'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 Cluelly. Now, as I've said before, voices one of the most important modalities of AI. It's the most natural human interface,

Starting point is 00:08:59 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. Today's episode is brought to you by robots and pencils. When competitive advantage lasts mere moments, speed to value wins the AI race. While big consultancies bury progress under layers of process,

Starting point is 00:09:40 robots and pencils builds impact at AI speed. They partner with clients to enhance human potential through AI, modernizing apps, strengthening data pipelines, and accelerating cloud transformation. With AWS certified teams across U.S., Canada, Europe, and Latin America, clients get local expertise and global scale. And with a laser focus on real outcomes, their solutions help organizers work smarter and serve customers better. They're your nimble, high-service alternative to big integrators. Turn your AI vision into value fast.

Starting point is 00:10:08 Stay ahead with a partner built for progress. Partner with Robots and Pencils at Robots and Pencils.com slash AI Daily Brief. Today's episode is brought to you by my company, Super Intelligent. Look guys, buying or building agents without a plan is how you end up in pilot purgatory. Super Intelligent is the agent planning platform that saves you from stalling out on AI. We interview teams at scale, translate real work into prioritized agent opportunities, and deliver recommendations that you can execute on. What to build, what success looks like, how fast you'll get results,

Starting point is 00:10:39 and even what platforms and tools you should consider. All customized for you. Instead of shopping for hype, you get to deploy with confidence. Visit besupor.ai and book your AI planning demo today. Welcome back to the AI Daily Brief. I am on record saying that I think that the two big themes for AI heading into 2026 are context and ROI. And with a set of announcements this week, you'd almost think that these companies were trying to front run the entire side of that context theme by getting a whole new set of features into place. In fact, there's so much going on that even stalwarts like Professor Ethan Malick are having trouble keeping up.

Starting point is 00:11:20 On Thursday, he tweeted, lots of scattered AI announcements in the past couple of days. days that are pretty valuable for business use cases from Microsoft, OpenAI, Google, and Anthropic around memory, business context, and teamwork with AI, real issues. Now, I just need time to try some of them out myself. First up, we have Anthropics Claude getting a memory upgrade. Memory is the absolute Achilles heel when it comes to productive LLM use. If you have spent time having to reintroduce your LLM to a whole slew of context or background about a particular issue that is relevant for the prompt that you're trying to give it, you'll know just what a pain this is. You also might have experienced that challenge where you thought that it had all of the

Starting point is 00:12:03 background context, but then all of a sudden, out of nowhere, it just behaves as though it is forgotten everything. And yet still, one of the biggest reasons, if not the biggest reason, that I have stuck pretty closely to chat GPT as my main tool, even though I use all of the popular chatbots at various points, is that it has a better set of memory and context around my work. Well, now with this new upgrade, Claude is getting its own version of memory. This first became available to team in Enterprise users in September, but is now rolling out to paid subscribers more broadly. And the simple idea is to give Claude access to previous conversations so that you don't have to do all of that sort of background and reminding every single time. Now, Anthropics says that they're trying to be extremely transparent around how memory works.

Starting point is 00:12:46 The new features allow users to both search and reference chats as well as to generate memories from chat history. When it does that memory generation, it gives people the ability to see what things Claude actually remembers. It provides a memory summary, is transparent about which chats it comes from, and also tries to give you more controls around turning memories off. The Verge writes, you could tell Claude to focus on specific memories or, quote, forget an old job entirely. They're also effectively trying to create distinct memory spaces or project-based memory organization so that the memory itself can be organized into different buckets. Now, this is a real issue right now.

Starting point is 00:13:22 I've called it in the past context confusion, and where I see it most acutely in my interaction with ChatGBTBT is that it has a hard time understanding where AI Daily Brief as a business begins and ends as opposed to super intelligent, which, although related via me, are two separate things with different revenue streams, different goals, different players involved. And so I'm excited to see if Anthropics approach to this can actually help solve for that sort of context confusion issue. They're also allowing people to import memories from other platforms. platforms like ChatGPT and Gemini and export memories from Claude so that there isn't memory

Starting point is 00:13:53 lock-in. Most people are just straight up excited about this. Although Ruin Dang does note that one of the potentially negative things that comes with increased memory is the expansion of what personal data people are comfortable giving their AI. She writes, people's tolerance for AI storing their data keeps growing because for users its usability. Just like in the mobile era, we once feared apps knowing too much and exposing us too easily,

Starting point is 00:14:16 then we started worrying about not being seen enough. The wheel of history turns again. Now, speaking of history, we also recently talked about a new feature from Anthropic called Skills. Skills are basically a way to create little packages of context that the Claude models can call on when they make sense, and it's a way to both improve the context that Clod has access to as well as improve the efficiency of the model because they can use a less sophisticated model to figure out which of those skills they should be drawing upon for a particular query

Starting point is 00:14:43 or prompt, and then only deploy the real aggressive use of tokens when they're in that right skills context. Well, as Alex Albert, who has Anthropics big Claude Haidteman on Twitter, points out with this chart, GitHub stars for Anthropic skills are growing at a much faster rate initially even than MCP. Now, real ones might remember that MCP, while it was initially interesting to people, took a few months to really hit its inflection point around March of this year, and then had that sort of parabolic growth curve. Skills, though, at least so far, has been just straight up and to the right. Next up, we move to OpenAI, who has announced a direct business context feature called Company Knowledge. OpenAI, CEO of applications, Fiji Simo writes,

Starting point is 00:15:21 it brings all the context from your apps, Slack, Google Drive, GitHub, etc., together and chat GPT so you can get answers that are specific to your business. Company knowledge is basically exactly what it sounds like. The idea is that a huge amount of the relevant context for a particular business lives inside the documents and history of the other applications that that business uses. Think conversations in Slack, planning documents in Google Docs, contacts and HubSpot, you name it. Company knowledge is a more simplified user experience that gives enterprise users access to all of that information.

Starting point is 00:15:52 In the announcement post, they write, chatchipT can help with almost any question, but the context you need to get worked on often lives in your internal tools, docs, files, messages, emails, tickets, and project trackers. Those tools don't always connect to each other and the most accurate answer is often spread across them. With company knowledge, the information in your connected apps,

Starting point is 00:16:11 like Slack, SharePoint, Google Drive, and GitHub, becomes more useful and accessible. For example, if you have an upcoming client call, ChatGBT can create a briefing for you based on recent messages from your account channel and Slack, key details from emails with your client, and the last call notes in Google Docs, and any escalations from Intercom support tickets

Starting point is 00:16:28 since your last meeting. Now, this is one of those absolutely duh features that is just totally essential and completely game-changing for enterprise users. This sort of enterprise search is so valuable that companies like Gleen have built a nine-figure revenue business around just this core feature. If you were using a version of chat chapit that has company knowledge enabled, under the Ask Anything bar, there should be a little button that says company knowledge,

Starting point is 00:16:52 and when you click it, it gives you the ability to add all of the connected apps that you use at work. As it's working and drawing upon those sources, it shares its chain of thoughts, you can follow along and see what's happening, and importantly it provides citations of the sources it used to inform its responses, along with these specific snippets that it drew from, giving you the ability to dive deeper into that original source to both double-check the work or to go deeper. on some particular question. Now, it seems like the search model that they're using is pretty sophisticated, at least in terms of how they're describing it. They claim that it's smart enough to understand conflicting details and can run multiple searches to resolve those details. It can also

Starting point is 00:17:27 provide comprehensive responses that don't just rely on one source. In other words, it's not necessarily optimized to just find the fastest answer. It's got a prerogative around comprehensiveness, and it even has the ability to rank sources by recency and quality, making it so that you don't necessarily have to specify time or dates for it to get you the most relevant and recent information. Now, of course, they also give a whole bunch of provisors and guarantees around privacy, and one interesting note is that when the company knowledge feature is turned on, chat GPT does not have access to search the web or to create charts and images. You can manually turn it off midstream and continue working in the same conversation to use

Starting point is 00:18:03 those capabilities, and it doesn't lose that existing context, but right now they're separate features. The reason for that, as Andrew, who's the AI obsolete at Berkshire Gray points out, is that it's actually powered by a new model. Andrew writes, Quietest Agent Release I've seen. What he's referring to is this. It's powered by a version of GPT5 that's trained to look across multiple sources to give more comprehensive and accurate answers.

Starting point is 00:18:26 In other words, this is a version of GPT5 that is optimized for this particular use case. I imagine that this is going to be an extremely unlocking feature for a huge number of enterprise users. Certainly the people at OpenAI who are using it are already reporting positive results, and I would expect this to become completely dear rigor and a huge competitive advantage for ChatGPT's enterprise version. One other interesting OpenAI announcement. On Thursday of this week, the company also announced that it had bought the bombastically named software applications, Inc, which is the two-year-old AI startup behind Sky, which they describe as a powerful natural language interface for the Mac. From the blog post, with Sky, AI works alongside you, whether you're

Starting point is 00:19:06 writing, planning, coding, or managing your day. Sky understands what's on your screen and can take action using your apps. You'll remember when we talked about the AI browser that the two prospective value propositions of that are on the one hand, agent mode, and the ability for the browser to actually do things for you, but that my feeling about that set of benefits is that they are kind of locked in for the future as opposed to something that's going to be super relevant right now, whereas the immediate benefit is the ability to use your chatbot with the context of what's on your screen. Instead of having to drag a tweet that you're drafting over into chat GPT,

Starting point is 00:19:39 if you've got the sidebar pulled up on the window, it can just see what you're drafting, which is a reduction in the cognitive load that comes with context switching. Sky is something similar to that, but instead of it being a browser, it's your entire operating system that they have the context of. A video shows someone grabbing a message from my message and dragging it into the sky window, and that being able to unlock a whole set of next steps, including putting it on a calendar, really making this an operating system parallel to the context benefits of the AI browser.

Starting point is 00:20:05 Most people were just absolutely gobsmacked that Apple had let this team, who's so deeply integrated into the Mac operating system, go to Open AI. Finally, on this theme of context getting an upgrade, we have the fall announcements for Microsoft co-pilot. Now, they set this up as a broad release with a bunch of different pieces. And while Microsoft AI CEO Mustafa Sullyman said that the announcements all boil down to the one core idea of them betting on humanist AI, I think that the subtext is all about context. So there are a couple of different ways that this is manifest. One interesting one is co-pilot groups. Co-pilot groups are kind of exactly what you would imagine.

Starting point is 00:20:45 If you are using co-pilot to plan or brainstorm or iterate on something that involves a group, maybe you're planning a trip with friends or you're thinking through a problem with classmates, groups allows any particular co-pilot conversation to become a group thread. The friend trip planning example was the one they gave in their demo video. after getting the conversation started, a link populated that the prompter could use to invite a set of their friends to all be part of the conversation.

Starting point is 00:21:09 Now, this is obviously quite valuable for that use case, but I kind of imagine this being the type of feature that quickly becomes table stakes across all of the different tools. I just think that there are enough times when you want to be actively engaging with other people, particularly in the work context, where it's going to be more valuable

Starting point is 00:21:26 to bring them into the conversation early as opposed to just sharing a link to the conversation later. One of the things that happens at superintuitive, all the time, is one of us will have some extended thread with one of the tools and then have to catch people up by using the link to that conversation that they then have to go back through and read and try to grok. This way just seems much more efficient and again reduces context switching. They also explicitly added what they call deeper memory and shared context. In their announcement blog, they write,

Starting point is 00:21:52 copilot now has long-term memory, helping you keep track of your thoughts and to do-do list almost like a second brain. With memory and personalization, you can ask co-pilot to remember important information like training for a marathon or an anniversary, then recall it during future interactions. We're also beginning to roll out the ability to reference explicitly past conversations, making it easier to pick up where you left off after some time has passed. Then hearkening to the company knowledge features that we were just discussing, they're also adding connectors. With connectors, you can link co-pilot to OneDrive, Outlook, Gmail, Google Drive, calendar, etc., and bring all of that context into your co-pilot conversations.

Starting point is 00:22:27 Microsoft, of course, also got its own version of an AI browser. with them explicitly saying that copilot mode in Edge is evolving into a full AI browser. We've already extensively covered the context implications of that, and they're also using their integration with the Windows operating system to, as they put it, turn every Windows 11 PC into an AI PC. Of course, for many, there's no bigger news from this Microsoft announcement of the return of Clippy, who is this time named Miko, but for me, as you can tell, this announcement is all just about making AI work better

Starting point is 00:22:56 by giving it more information about the person who's piloting it. I think we are going to see more and more efforts around memory, long-term context, understanding, but there is no doubt that after this week, context has gotten a big upgrade. For now, that's going to do it for the AI Daily Brief. Appreciate you listening or watching, as always. And until next time, peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - AI Context Gets a Major Upgrade

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.