Speaker 1Yesterday, Anthropic announced Claude Duet, their answer to Grokbot and Metamuse and OpenAI's Dots. And yet, somehow, Duet is not the focus of our main episode. How could that possibly be? Well, in short, it's because that right now we are living through what may ultimately be seen as a key transitional moment between two very different eras of cybersecurity. One where long-term norms held, and even exploits and challenges were manageable, and the other, newer era, where it's AI versus AI, and a battle for who can coordinate best. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Section, CDATA, and Blitzy. 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 if you want to learn more about sponsoring the show, send us a note at sponsors at ai-daily-brief.ai. And we kick off today with a story that, on almost any day on this show, would absolutely be the main, and in fact, would usually be a main-only type episode. Anthropic has joined the personal agent market with their own twist on Muse slash Dot slash Grokbot, and they're calling it Claude Duet. The two reasons that today, for us, it's just a brief headline introduction are that, one, the massive slate of cyber issues going around the internet right now are fairly unignorable, both in the specifics, but also what they allude to for the future. And secondly, I'm working on a full-length, long-term, and long-term plan for the future of the show. Which is going to go much deeper into Claude Duet, and exactly how I'm using it and its competitors and what I'm finding so far. So come back on Sunday to check that out. For now, it appears that Anthropic is, perhaps unsurprisingly, aiming at the professional and small business side of the personal agent market, with their main differentiator seeming to be a focus on being proactive. Describing the core functionality they wrote, each morning Duet sends you a short brief on the most important items across your connected apps. Throughout the day, it keeps tabs on those apps, and it's a lot of fun. So if you're interested in that, messages you only when something needs your attention. You can also hand it tasks of all shapes, from simple administrative tasks of scheduling meetings, to more complex ones such as building dashboards with dozens of tools, creating financial or sales forecasts, prototyping and building web apps. Besides the proactive approach, Claude Duet has all of the features that people have come to expect from the current crop of agents. Duet has its virtual machine, persistent memory, and a full suite of connectors. And Anthropic also emphasized a focus on safety. Duet will ask for explicit permission before taking major actions, like making a payment or deleting files. Anthropic has also created what they call a new dedicated classifier and safety system prompt that checks each message and guards against leaks of sensitive information. Right now, Duet is being rolled out as a research preview, and you can't just decide to use it, you have to get on a waitlist. Anthropic warned that this is an early version, which might have some issues like misreading threads or miscalibrating what's urgent. And they also warned that it won't be as capable at personal errands like booking tickets as other agents, because of the extent to which their focus is on work tasks. In general, the community treated this with excitement, if not surprise. And for many, as summed up by Dan McAdier, the big and immediate differentiator wasn't anything about the proactivity or even necessarily its focus on work tasks. It was the fact that it was built on Opus 5-5. There were a few folks like Chubby who were surprised to see Anthropic pursuing a personal agent. He basically said that he wasn't sure who the target was for this, but I think that that might be a category error in thinking that personal agents are just going to be for personal uses. It seems to me that there is a bit of a recognition that even though we have work and personal lives, the boundaries of those things are fairly permeable. And when it comes to these so-called personal agents, right now, it seems like the labs are betting that while we might gravitate towards one that has our particular use cases emphasized, they just have to be able to do everything that we might want them to do. Signal summed it up this way. Pretty much every major AI company now has a personal agent product except Google. Meta has Muse, Anthropic, Duet, OpenAI, Dots, SpaceX, Bots. Slowly but surely, the competition is going to shifting from who has the smartest model to who owns the persistent relationship with the user. This will be even more the case going forward. You're seeing this because we likely reached peak intelligence required to power most of the world's workflows probably earlier this year, if not late last year. Mostly, I think the discussion shows just how nascent this product category remains. There is nothing even approaching consensus on whether these things are the right form factor, or what their most important modes of interactions are, or what even the essential features are. People disagree around whether we'll have one or many, and what the most important distinguishers will be. I know that subsequent to my time using these personal agents so far, I'm not 100% convinced that they're going to ultimately be an essential part of the portfolio. But like I said, I'm going to get a lot deeper into that on Sunday. And I guess a last note on Duet for those who are keeping track. If I were a betting man, I think this is part one of a one-two punch heading into the IPO, at least from a product perspective for Anthropic. Release your answer to all the personal agents floating around, and then give it a new model which makes everyone feel just so far behind. In other words, maybe this is just prelude to Fable 5.5. Next up, staying on the theme of agents, contra to some other businesses out there, Walmart has decided to welcome personal agents onto their platform. Recent weeks have seen corporate America divided on the risks of personal agents interacting with services on their owners' behalves. Shopify was of course an emphatic yes, while Amazon, JP Morgan, and Bank of America were a clear no. It should be noted that the reason for the no's between those companies was a little bit different. JP Morgan and Bank of America were a clear no. It should be noted that the reason for are thinking about risk, in terms of agents doing something with people's money that they shouldn't, while Amazon is focused more on competitive risk. In that light, Walmart allowing third-party agents is kind of a surprising move. The retail giant was very early to AI adoption, and one of the core pillars of their strategy was Sparky, their proprietary shopping agent. Earlier this month, Walmart said customers who used Sparky had order values 40% higher than non-agentic online shoppers. Still, it seems now that Walmart wants to encourage agentic commerce and they don't particularly care which agents their customers are using. As of next month, customers will be able to set up a linked account for their personal agent with configurable spending limits, kind of similar to how a child's account might work. Aside from supporting third-party agents, Walmart's new policy included lengthy terms and conditions that could become a template for safer agentic shopping. Walmart included rules that an agent must not take actions beyond what the shopper intended, and that agents can't modify a customer's payment methods. Walmart will require purchases to be approved by default, and said they would ban agents who are not compliant with their guidelines. However, for compliant agents, Walmart said agentic purchases would have access to the returns policy even if an agent makes a mistake. Once again, we find ourselves in incredibly early days for agentic commerce, and I think trying to make definitive statements now about how this is going to play out is very hard if not impossible. Still, early evidence suggests that for those who are using personal agents, agentic commerce does seem to be an early use case that people are exploring. During Wednesday night's earnings call, Meta said that one quarter of agents who are using personal agents are not compliant with their guidelines. For those who are using personal agents, the average value of their purchases is $200, which surprised even the e-commerce veterans who have been paying attention to the space. Mark Zuckerberg commented, Plaid told us that they hadn't seen activity like they're seeing with their connector on any other service before. In that light, Walmart's strategy begins to look a little bit clearer. If the early adopters of agentic shopping are very valuable customers, Walmart just wants to let them shop. Speaking of Meta, the company has announced a new class of non-voting stock to help fund the next leg of their AI. The announcement came alongside those same third-quarter earnings, which featured a very minor upwards revision to their data center budget. Meta now expects to spend between $135 billion and $145 billion this year, adjusting the low end up from $130 billion. The new stock will begin trading in January and allows Meta to raise funds by selling equity without diluting Mark Zuckerberg's voting control of the company. CFO Susan Lee wrote in the earnings report, the scale of investment required to compete, sustain growth, expand infrastructure, and pursue new strategic initiatives amidst the AI transformation taking place in the industry is significant. During the earnings call, Lee added that the new class of stock will expand our funding tools and give us flexibility to invest behind the unprecedented momentum we're seeing across every part of our business driven by AI. There are a few reasons why Meta would pivot to funding their data center build-out with equity rather than debt. Some thought that this was a sign that the bond market is completely tapped out on AI debt. But of course, it also might be that Meta sees that as a possibility moving forward, rather than a current state condition, and wants to have a new class of stock. The new class of stock will expand our funding tools and hedge against it. Others noted that Meta has $900 billion in liabilities, including $350 billion in non-cancellable data center contracts. That's a rough gauge of spending over the coming decade, but only about $83 billion of that total is currently on issued long-term bonds. Perhaps the scale of the build-out suggests they won't be able to fund it all with debt. Now, ultimately, the Occam's razor answer, or I guess the generous read depending on your position on Zuckerberg, is that Meta is tapping all available pools of capital and keeping their options open. Issuing stock is a deep pool of capital, as long as Meta isn't too concerned about diluting existing shareholders. Financial planning advice from Mollus, which was attached to the earnings report, said Meta would need to raise several hundred billion dollars over the coming years in any scenario. They noted that if they did issue voting stock, they could still raise as much as $700 billion without diluting Mark Zuckerberg's vote below 50%. Interestingly, Meta stock was up in overnight markets, suggesting shareholders are willing to deal with some dilution for now. What this means for the stage of the build-out remains to be seen, but it is another important signal. The current plan barring some of the most important stocks in the market, happening on Friday, is a much deeper dive into what I think are the actual most concerning bubble signals. So if you're interested in that, come back tomorrow. For now, though, that is going to do it for today's headlines. Next up, the main episode. A new study from KPMG in the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes. Researchers studied more than 500 early career professionals and found that the best performers consistently amplified the number of people who worked with AI. simplified the value of AI by guiding, evaluating, and refining its outputs. These top performers, called AI amplifiers, weren't defined by what they knew alone, but by how they worked with AI. Learn more about what separates AI amplifiers from everyone else at kpmg.com slash US slash AI amplifiers. 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 at sectionai.com. That's S-E-C-T-I-O-N-A-I dot com. This episode is brought to you by CDATA. Companies rolling out AI keep hitting the same wall. Token spend is climbing, security is holding agents back, and no one knows how AI is truly being used. The root cause? AI is cut off from the systems and context your business runs on. That's why CDATA built Connect AI, which is a tool that can help you One control point between your AI and your systems. Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more, each agent gets exactly the data and tools it needs. It governs down to the record and lets you finally see your org's complete AI usage in one place. Accurate answers, a fraction of the token cost, and a gateway that learns your business with every interaction. See how at cdata.ai. Blitzy's deep code-based understanding unlocks the thing every roadmap owner cares about, shipping new features, and making changes to your code. Here's the truth about building inside a massive enterprise codebase. Writing code was never the bottleneck, context is. Which system does this touch? Which contracts can't break? Which standards apply? Blitzy already knows because it reverse engineered your entire codebase into a dynamic knowledge graph before feature work began. With that complete picture, Blitzy builds features end-to-end. Architecture, APIs, UI, and tests all validated against your existing systems. One Blitzy customer built an AI-native application from scratch with 100% autonomous completion, saving over 2,700 engineers. And that's what Blitzy is all about. It's a world-renowned AI-native application that's able to do everything it needs to do for its customers. And it's a world-renowned AI-native application that's able to meet the needs of thousands of people and make the most out of it. It's a world-renowned AI-native application that's able to meet the needs of thousands of people and make the most out of it. due thanks to AI defenses finding issues before the inevitable wave of incoming cyberattacks by other AIs. Now consumers are going to have the same experience of getting everything updated often and fast. Now, one question some had was around the disclosure method of this particular exploit. Maple AI CTO Anthony Roening wrote, If they've worked on the PHP vulnerability for over two months, then why has there been such a poor disclosure and rollout strategy? It's basically been a strategy of, quote, in 24 hours, you're all screwed. Cybersecurity professional Tay Monahan wrote, This might be a stupid question, but given the discovery last week and that it started leaking 12 plus hours ago, why weren't the PHP patches ready to be dropped as soon as the vulnerability slash POC fully leaked? Why is everyone okay arbitrarily waiting another 12 hours to patch? Christian the Pav writes, I'm no security expert, but telling the hacking world there's a mega huge apocalyptic exploit for PHP and giving them a day to figure it out and do as much damage before they release the fix seems like a bad idea. Couldn't they release the update then after a day explain what it is? Now, my understanding from reading accounts from people who, again, know a lot more than I do about this is that the patch itself makes the exploit. obvious, so the concern was that as soon as they released the patch, they were also essentially making the exploit known in the wild. This is why the NSA warned everyone to take their sights down ahead of the disclosure. Frederick Willems wrote, The real hacking world isn't the issue. They probably won't target most of us. Releasing the exploit slash patch unleashes a storm of script kitties that will scan and breach everything they find. They tried to get everyone ready to update the moment it lands. Now, before we get into broader implications, some are lamenting what might be the legacy of this particular exploit itself. Kenny Lauderdale writes, This PHP stuff is so depressing because there's going to be 15 years worth of old sites that are going to become lost media now. Cyber Security News reporting account The International Cyber Digest wrote, 4chan has gone read-only and something awful has shut its forums as American sites brace for the PHP patch. Bookmarking service Pinboard, game publisher Choice of Games, history site Damn Interesting, and the Explain XKCD wiki are offline too. And then to add to the idea that these types of attacks were not going to be limited in their frequency, Yesterday also saw, Railway hit by a major exploit. Railway is an AI hosting service similar to Vercel. In a disclosure post, Railway wrote, Earlier this morning, Railway detected anomalous activity involving systems that store customer environment variables. After further investigation, our investigation to date indicates that an unauthorized third party may have accessed encrypted environment variables, and for some of them, obtained the decrypted contents. Environment variables may contain secrets such as API keys or database and registry credentials. We believe this is part of a series of attacks recently targeting platform companies with the aim of compromising AI inference providers. At first glance, this looks very similar to the open code exploit from last weekend, which again saw attackers gain access to user API keys, allowing the attackers to steal tokens and run up big bills. Like the open code situation, Railway is suggesting that all users should rotate API keys and roll all environmental variables stored on the platform. Leo @Synthwave wrote, Keys actively being drained, and they've disabled the ability to deploy anything so can't deploy with new variables. Nightmare day for cyber, and it's only going to get worse from here, I fear. Dax from OpenCode responded to the Railway team and summed up the new normal. He wrote, I'm sure you're all better than me at this, but just some of what we've seen if it helps. They're not going to stop. It's going to get increasingly sophisticated, and they're going to go after every surface of your system and all of your employees. Any old projects you've long forgotten about will be found and used against you. The only reason we're surviving is because we have agents going 24/7. We're still barely sleeping. Good luck. Sorry you're dealing with this. But really, it's a collective problem we're all facing together. Now, if this is the new normal. If this is the new normal, there are very different paths that that normal can take. On the one end of the spectrum is what was suggested by former OpenAI staffer Joshua Achayim who wrote, "Feels like March 1st, 2020 for cyber. Not normal. Something is happening. Internet lockdowns in a couple of months?" On the flip side, there are those who are recognizing that we're probably going to have to fight AI with AI. Entrepreneur John Ennis wrote, "Well, we just rotated every single key across a large number of projects very quickly, and I'm struck by the fact that this would have been an absolute nightmare not too long ago. The defensive aspect of AI for cybersecurity is really underrated. Where the nuclear weapon analogy breaks down with AI is that you can't stop a nuclear weapon with another nuclear weapon, but you can stop an AI attack with an AI." Box's Aaron Levy built on those themes, writing, "Cyber is going to be one of the most profound aspects of AI. We now have a tireless workforce that can hunt down, discover, patch, or attack any system and software ever created. We will use agents to identify the long-standing risks in existing systems and infrastructure. Attackers will use agents to accelerate the discovery. We will equally use agents to defend attacks and discover anomalous patterns to better protect our data and much more. We're entering a new cyber regime, and we'll probably use 100x more tokens on this loop than we're imagining right now, because we can scan every log, look at every line of code, look through traffic, and process data at a scale that wasn't comprehensible before. More than ever, enterprises are going to care deeply about how they manage their networks, systems, software, data, identity, devices, and more. Agents will largely become the solution to these challenges, but they will keep everyone up to date on how they're going to manage their networks, systems, software, data, identity, devices, and more. They're going to be the key to solving these problems.
PodcastTranscript AI