The Next AI Platform Isn't a Model -- It's Your Context

0:00This podcast is supported by Google. Hey folks, Stephen Johnson here, co-founder of Notebook LM. As an author, I've always been obsessed with how software could help organize ideas and make connections. So we built Notebook LM as an AI-first tool for anyone trying to make sense of complex information. Upload your documents and Notebook LM instantly becomes your personal expert, uncovering insights and helping you brainstorm. Try it at notebooklm.google.com. Today on the AI Daily Brief, why the next AI platform war is being fought over your context. Before that in the headlines, OpenAI's next bet is building their own chips that design themselves.

0:42The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

0:54All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, Gemini, KPMG, Robots and Pencils, Notion and Blitzy. To get an ad-free version of the show, go to patreon.com slash AIDailyBrief, or you can subscribe on Apple Podcasts. And if you are interested in sponsoring the show, shoot us a note at sponsors at AIDailyBrief.ai. And while you are there, you might notice that we have a new tab on the website for jobs. The short of it is I am exploring a single growth role, helping me make this thing as absolutely gigantic as possible. The show has grown more than 300 % over the last year.

1:27It is consistently very high in the rankings. It gets millions of downloads and views every month. But the inescapable reality is that in this world, the total addressable audience who should care about what's happening in AI is somewhere around everyone, meaning that we have a lot of opportunity. The job then is to help me grow the podcast as big as possible. The responsibilities are whatever it takes to do so. And how to apply is pretty simple. Impress me, grab my attention, show me how you think, send me a note at jobs at aidailybrief.ai, and let's see what we can do. With that, let's get into today's episode.

2:02Welcome back to the AI Daily Brief Headlines Edition, all the daily AI news you need in around five minutes. We kick off today with another big infrastructure deal from OpenAI. The company has signed a multi-year partnership with Broadcom. The two companies will collaborate on custom silicon and networking equipment and plan to deploy 10 gigawatts of data center capacity powered by the Broadcom hardware. The goal is to start deploying server racks in the second half of next year. Now, the deal is non-binding, allowing Open AI to walk away if things go pear-shaped. But Open AI said that their pursuit of custom silicon would allow them to embed what they've learned, quote, directly into the hardware, unlocking new levels of capability and intelligence.

2:41In a podcast released alongside the announcement, Open AI president Greg Brockman explained that this literally means that GPT-5 is making improvements to the chip design that will improve the performance of the next model. He said, We're at the point now where I don't think any of the optimizations we have are ones that human designers could have come up with. Usually our experts take a look at it later and say, yeah, this was on my list, but it was a list of 20 things that would have taken them another month to get to. Broadcom benefited from the OpenAI bump, with the stock rocketing by 12 % in overnight trading.

3:11Now for those keeping track, that means that OpenAI has made around 26 gigawatts worth of chip deals over the past month between NVIDIA, AMD, and now Broadcom. And while it's difficult to get an accurate read on exactly how large the supply of AI data centers is currently, estimates range from 30 to 60 gigawatts of total data center capacity in the US, with between 10 and 20 % of that representing AI workflows. Basically, regardless of which estimate you go with, OpenAI is looking to more than double the supply of AI data centers in the U.S. all by themselves over the next five years. Unsurprisingly, what you think about this deal is pretty much informed by your pre-existing opinions of OpenAI's deal-making frenzy.

3:50Just like we have the Elon Rorschach test, where anything that his enterprises announce is really just a referendum on what people think about him, that's increasingly what we're getting with OpenAI as well. Some see a generational company scaling up their already huge bet. In that same podcast released alongside the announcement, Sam Altman explained the reason for wanting to make their own chips, saying, by being able to optimize across the entire stack, we can get huge efficiency gains, and that will lead to much better performance, faster models, and cheaper models. Broadcom CEO Hawk Tan put it simply, if you do your own chips, you control your destiny.

4:22Trader SpacePixel noted that this isn't just a big deal for OpenAI, but ratchets up the competition in general. They wrote, game theory would suggest Meta, XAI, Google, and Anthropic are going to have to double their compute due to the last week of OpenAI deals. AMD, Nvidia, and Broadcom are going to have some serious revenue growth over the next five years. Still, a far larger category of people were adding a few additional spokes to their circular investment map. In addition to the Broadcom announcement, OpenAI also announced a multi-billion dollar deal with chipmaker ARM to produce CPUs for servers powered by the custom silicon.

4:54ARM is 90 % powered by SoftBank, which is also involved in many of these big OpenAI deals. Last week, Bloomberg reported that SoftBank was seeking a$5 billion loan collateralized by their Armstock, and the OpenAI bump pushed that stock up by 11%, giving SoftBank a lot more room to borrow. Interestingly, the market wasn't all that enthusiastic about SoftBank, given these arrangements. The stock was actually down by more than 6 % during the Tuesday session in Tokyo. The one interesting thing that I think is kind of missing from the chatter is the possibility that these companies are all right and we actually are going to need this much compute.

5:27As OpenAI's Rune put it in a viral tweet, not enough people are emotionally prepared for if it's not a bubble. Now, staying on the chip theme for just a minute, more than half of AWS's AI services are now running on their own custom chips. Amazon has been steadily building out an ecosystem of their own chips to power their AI ambitions. Late last year, they released Tranium 2, which is the latest version of their AI accelerator. Tranium 3 was intended to be released towards the end of this year, but has hit delays with the design of liquid cooling systems. Tranium 2 was praised as a significant course correction for Amazon after the lackluster first edition.

6:00The chips aren't up to par with NVIDIA's leading chips, but they come with a significant reduction in operating costs. Tranium 2 handles both training and inference and was designed as a natural fit for Anthropik's heavy reliance on reinforcement learning. Amazon's compute deal with Anthropik has allowed them to pursue this aggressive build-out, safe in the knowledge that they have a large anchor customer. Speaking with the information, AWS Chief Marketing Officer Julia White said, Our long-term investment in our own chips, with Tranium being the most current example, is really a wonderful advantage for us from a price performance perspective.

6:30Now for the second part of this headlines episode, we have a bunch of product or feature announcements. The first comes from N8N, who has introduced a new workflow builder that allows users to build agents using natural language prompts. Oh my god, thank goodness. A long-awaited feature that is finally here. Since the workflow design tool started catching on earlier this year, the interface has been a huge blocker. For non-technical folks, node-based agent building can be a steep learning curve. And even for those who are somewhat technical, it still can be a pain. The new Agent Builder lays out a proposed workflow based on user prompts presented as a node system.

7:05Users can then tweak the workflow with the benefit of AI doing the heavy lifting. Right, so Zaire Akhtar, and they then delivered what everyone was expecting from OpenAI's Agent Builder. Next up, we have Google, who are integrating NanoBanana absolutely everywhere. The image generation model will be added to Google Search, Notebook LM, and soon to the Photos app. In Search, Nano Banana will be paired with the Lens tool to allow quick photo editing. It wasn't clear from the demo how it integrates with an actual search, but it's another pathway for users to get quick access to image editing. In Notebook LM, Nano Banana will now drive the Video Overviews feature, enabling six new visual styles for the generated slideshows.

7:41Google hasn't yet provided details about the forthcoming integration with the Photos app, but it's a pretty safe bet that we're getting embedded AI photo editing. Generally, the strategy is pretty clear, and it is to get NanoBanana's powerful editing features easily available everywhere they possibly could be useful within the Google ecosystem. Lastly today, Microsoft has announced their first in-house image generator. The text-to-image model called MAI Image One is part of an initiative kicked off in August to build that company's internal capacity for model training. The first batch included a small chatbot model and an accompanying voice model.

8:15Microsoft's AI CEO Mustafa Suleiman said at the time, it's critical that a company of our size is able to be self-sufficient in AI if we choose to. Now, so far, nothing Microsoft has produced is state-of-the-art, but MAI Image 1 seems relatively good for a first attempt. Microsoft said they worked with human artists to avoid repetitive or generically stylized outputs. They claim the model excels at photorealistic images, including features like lighting and landscapes. Speed is also a big factor, with Microsoft claiming faster generation than, quote, larger, slower models. While benchmarking image models is inherently subjective, Microsoft has managed to rank ninth on Ella Marina in preliminary testing.

8:52GPT-1, which is the model that drives OpenAI's ImageGen, as well as Nano Banana, are each ranked ahead of the Microsoft model. Whatever the specifics of this model, it is part of a larger trajectory of Microsoft's independence, which matters greatly, especially in the context of what we're talking about in our main episode, which is the coming AI platform where we're all about context. So with that, we'll close the headlines and move to the main episode.

9:42Whether 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.us slash AI podcasts, or search You Can With AI on Spotify, Apple Podcasts, or wherever you get your podcasts. AI changes fast. You need a partner built for the long game. Robots and pencils work side by side with organizations to turn AI ambition into real human impact. As an AWS-certified partner, they modernize infrastructure, design cloud-native systems, and apply AI to create business value. And their partnerships don't end at launch.

10:18As AI changes, Robots & Pencils stays by your side, so you keep pace. The difference is close partnership that builds value and compounds over time. Plus, with delivery centers across the U.S., Canada, Europe, and Latin America, clients get local expertise and global scale. For AI that delivers progress, not promises, visit robotsandpencils.com slash AI Daily Brief. Chatbots are great, but they can only take you so far. I've recently been testing Notion's new AI agents, and they are a very different type of experience. These are agents that actually complete entire workflows for you in your style.

10:53And best of all, they work in a channel that you already know and love because they are purpose-built Notion super users. Notion's new AI agents completely expands the range of what Notion can do. It can now build documents from your entire company's knowledge base, organize scattered information into organized reports, basically do tasks that used to take days, and get them complete in minutes. These agents don't just help with work, they finish it. Getting started with building on Notion is easier than ever. Notion agents are now your very own super user to help you onboard in minutes. Your AI teammates are ready to work.

11:23Try Notion AI for free at the link in our show notes. This episode is brought to you by Blitzy, the enterprise autonomous software development platform with infinite code context. Blitzy uses thousands of specialized AI agents that think for hours to understand enterprise-scale code bases with millions of lines of code. Enterprise engineering leaders start every development sprint with the Blitzy platform, bringing in their development requirements. The Blitzy platform provides a plan, then generates and precompiles code for each task. Blitzy delivers 80 % plus of the development work autonomously while providing a guide for the final 20 % of human development work required to complete the sprint.

11:58Public companies are achieving a 5x engineering velocity increase when incorporating Blitzy as their pre-IDE development tool, pairing it with their coding co-pilot of choice to bring an AI-native SDLC into their org. Blitzy is providing a limited-time, 30-day free proof-of-concept for qualifying enterprises. The team will provide a 5x velocity increase on a real development project in your org. Visit blitzy.com and press book demo to learn how Blitzy transforms your STLC from AI-assisted to AI-native. That's blitzy.com. Welcome back to the AI Daily Brief. I am very publicly on record at this point of saying that I think the two big themes for enterprise AI in 2026 are context and ROI.

12:39And this idea of context engineering is both a term and a framing and a discipline that have been on the rise throughout this year. Back in June, for those of you who want a primer, we did an episode called Context Engineering, What It Is and Why It Matters. That was June 25th, or you can just search for it on Google. You will definitely find it. And today what we're talking about is a news announcement that seems pretty simple at first, but I think is revealing of a broader context war that's going to be a key feature of product developments over the year to come. So first, let's do a little tiny bit of background on what this idea is.

13:11Around the middle of the year, you started to see tweets like this one from Toby Lucky from Shopify who wrote, I really like the term context engineering over prompt engineering. It describes the core skill better, the art of providing all the context for the task to be plausibly solvable by the LLM. So you're getting a couple of things from this. First of all, what context engineering is, is about organizing and orchestrating all of the data and information that an AI or an agent would need to successfully complete whatever task gets assigned. The connection to prompt engineering is the idea that while prompt engineering was about how to specify and name your task in ways that the AI could best understand and give you the type of response you wanted, context engineering refers to giving it access to everything it needs to not only do that better, but to take on bigger and bigger tasks.

13:57Anthropic, in a recently published guide called Effective Context Engineering for AI Agents, which we're going to talk about a little bit later in the show, describes the difference this way. They said, building with language models is becoming less about finding the right words and phrases for your prompts and more about answering the broader question of what configuration of context is most likely to generate our model's desired behavior. Back around that same time, OpenAI co-founder Andre Karpathy weighed in, saying plus one for context engineering over prompt engineering. People associate prompts with short task descriptions you'd give an LLM in your day-to-day use, when in every industrial strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step.

14:37Science, because doing this right involves task descriptions and explanations, few-shot examples, RAG, related possibly multimodal data, tools, state, and history. Too little or of the wrong form, and the LLM doesn't have the right context for optimal performance. Too much or too irrelevant, and the LLM costs might go up and performance might come down. Doing this well is highly non-trivial. And art because of the guiding intuition around LLM psychology of people's spirits. Now, I actually think that in some ways, when we use the term context engineering right now, we're referring to two entirely different things.

15:08There is, on the one hand, this highly technical process for how we're designing AI and agentic systems to be able to access and capture the right context at the right time. This is the sort of entrepreneurial or technical side of context engineering, and where a lot of the focus in the discussion is because people are still figuring out the best ways to give AI and agents the right context. There is another side, though, which is the one that's going to be a little bit more relevant for enterprises, and where I'm taking the concept and dragging it a little bit to the left, which is context engineering and context orchestration as the art of putting together in accessible ways all of the data and information people within your business context need to get the most out of the LLMs and agents that they're using.

15:50Basically for enterprises, context engineering is closely related to the sort of data readiness, data accessibility, and data fragmentation issues that we've been talking about recently, particularly in that episode I did about the lessons that we've learned from superintelligence thousands of interviews. And this is, of course, what I mean when I say that 2026 is all about context engineering. What I mean is that I think it's going to be about how enterprises organize and connect their data in ways that make it accessible to the AI systems that their people are using and that they are deploying to get ever increasingly complex sets of tasks done.

16:23So that is the background context through which I want to recontextualize. And man, we really need a word other than context here. This announcement from yesterday that ChatGPT is now in Slack. Slack posted, big news, the ChatGPT app for Slack is here. With Slack's new real-time search API, the ChatGPT app for Slack brings the power of ChatGPT into a dedicated Slack sidebar, a space for you to ask questions, brainstorm ideas, draft content, and solve problems. This is just the beginning of a smarter way to work together. Basically, the new app allows users to use ChatGPT from right within the app.

16:58Whatever the types of things that you would do for work tasks that you would do on the native ChatGPT app, you can now do directly from within Slack. This follows, by the way, a similar announcement from Claude and Slack from just last month. Now, part of the value here is just not having to switch between different applications. There is an inherent lag and a time drain when you have to move around between different environments. And so one part of it is just bringing that convenience and keeping people in Slack more where they're already doing that work. Importantly, though, for both integrations, they're connected to Slack's new real-time search API.

17:28What this means is that the models, whether it's Claude or ChatGPT, are able to search through your Slack instance to access the full context of your work chats. In other words, instead of having to explain and give a bunch of preamble around the context for a particular request, by having this embedded in Slack and connected to the real-time search API, if that context exists in your Slack conversations, the idea is that these LLMs can just draw upon that without you having to go explain it. One of the things we've been talking about a lot recently is how powerful memory is as a moat when it comes to LLM usage.

18:06At this point, I am very regularly using ChatGPT, Claude, Gemini, and Grok. For me right now, there is no one model to win them all. In fact, in a tweet this week, I compared them to a team of different interns, all with different personalities. And yet, despite knowing what I like for different use cases, my default has been and will continue to be, at least for now, GBT5 because of how much context ChatGPT has about me, just through the basic memory that it has. I don't have to re-explain everything about the AI Daily Brief or Super Intelligent every time I ask it to engage with a strategic question, and that's an example of where its own memory is the context.

18:44Now, bringing it back to Slack, this is actually part of a broader announcement. Alongside ChatGPT, the company reframed this as the next evolution of the Slack platform. And basically what they're trying to do is be the foundational infrastructure for agentic work. They write, now you can build and use powerful context-aware AI apps and agents that securely connect to your conversational data right in your flow of work. And they point to applications from OpenAI, Anthropic, Google, Perplexity, Writer, Dropbox, and Notion as all taking advantage of that context. Now, Slack and its parent company, Salesforce, know exactly how valuable this context data is.

19:24We know this because back in June, right around the time, by the way, that we were first having these context engineering conversations, Salesforce started blocking companies like Glean from accessing Slack data. Now, this was a big hit on Glean, which has started to carve some of its own moat as an enterprise search tool. Glean tried to turn their customers against Salesforce, basically making the argument that the data doesn't belong to Slack and Salesforce, even though it happened in Slack, that it belongs to the customers and that they should be able to bring it wherever they want into Glean, despite Glean being a competitor to certain types of AI features that Salesforce and Slack wanted to build.

19:58According to an internal email that was intended for Glean customers, they claimed that Salesforce and Slack were, quote, hampering your ability to use your data with your chosen enterprise AI platform. Then a couple of weeks ago, however, Salesforce reversed positions and open Slack backup for external AI. And now alongside Dreamforce, which is going on in San Francisco right now, we can see how their approach has changed. Instead of them trying to just keep everything in their own AI ecosystem, it seems like they are now instead making a bet that the incredible context represented by Slack make it the perfect place to be a context platform for all other AI apps.

20:36Indeed, the company is now positioning Slack as your agentic OS. As part of that, they're launching a new personal AI companion called Slackbot, as well as their own version of Enterprise Search, but they're also making this a platform play, quote, powering an open ecosystem of agents that connect to your entire enterprise. And this is why the title of this show is about why the next platform war is in a model, but all about your context. Dong Ming writes, Slack seems like an obvious place for people and AI agents to interact. As AI agents become more common in the future, I guess that's where they will hang out to gain more context and do better work.

21:11MJ Kang writes, Post-MCP, everyone's scrambling to become everyone else's aggregator. The winner will be the product with the richest personalized context. And to do that, the company should, one, have the longest session time and strongest engagement, and two, collect as much information as possible. And indeed, if those two things are the big criteria, you can see why a work communication app like Slack might be such a contender. However, it is not the only contender. There have been a set of announcements over the past couple of months that all suggest how this context platform war is going to be fought.

21:45Back in June, Grammarly announced that it was acquiring Superhuman. Now, when it was announced, Grammarly said that the acquisition was about accelerating its evolution into a, quote, AI productivity platform for apps and agents. And that this acquisition, quote, positioned email as a critical communication service in the company's vision of an agentic future. A lot of that announcement was about email as a workspace into which you could embed agentic and AI tools. And certainly given the amount of time we all spend on email, that made sense. However, Perplexity's launch of a personal email assistant kind of puts that acquisition into a different light.

22:20In that announcement, Perplexity gets more directly at just how much context about you email has. They write, email is more than a message center. Your inbox contains your professional memory, your relationships, calendaring, and coordination. And in Perplexity's case, while the goal is a personal assistant that takes advantage of that context, it's clear that what they are trying to access uniquely is the context that email provides. Context is also one of the big reasons that many people think Google has such an advantage when it comes to the long-term, particularly for enterprise AI. A huge amount of work already happens in the Google workspace.

22:57Think about your Gmail, your Google Drive, your calendar, slides, forms, you name it. That entire suite is just buckets and buckets of work context specific to you. Now let's reframe last week's announcement of Gemini Enterprise in light of this broader platform war for context. Gemini Enterprise effectively pulls together context from Google Drive, Gmail, and Google Calendar and layers a powerful agentic interface on top of it. Now, there is, of course, one other company that has a ton of work context that isn't currently in this conversation, but which lurks just around the edges. And that is Microsoft.