Fable 5 launches while Siri partners with Gemini | Now Shipping

In this week's AI briefing for product people, Mike Belsito unpacks Anthropic's Fable 5 launch, Codex's expansion to non developers, and Siri's partnership with Gemini. 

Transcript

SPEAKER_00 0:00

I'm Mike Belcito, and this is Now Shipping, the weekly AI news show for product people. Every week I'll bring you three AI news stories that matter to you, as somebody who's actually building products for a living. Not the hype, not the noise, just the stuff you need to keep up with. Brought to you by the team at Mind the Product, this is Now Shipping. We have a lot to get into today, three big stories. Um, so I'll jump right in with a bit of a preview. Story one is all about OpenAI's Codex platform and how it just stopped being a developer tool. Now, story two is about anthropic. They just made their most powerful model available to the public. Claude Fable 5 launched earlier this week, and it came with something that I have not seen before. And then there's story number three. Apple just reset what Siri actually is. Apple's Worldwide Developers Conference headline this week was a ground-up rebuild of Apple's long struggling assistant Siri. Those are the stories. Let's dive right in. Okay, story number one is about OpenAI's Codex platform. I'll say going mainstream. Um, I'm sure you're familiar with Codex. Uh, maybe you've used it before, but if you haven't, it's OpenAI's coding agent for Chat GPT. It's a developer tool, right? Maybe not, though, because I'm gonna give you a number. 20%. That's the share of Codex's 5 million weekly users who are now non-developers. They're not engineers, they're not people who write code for a living. This may be product managers and designers, um, the people that are in our world, but think about people outside of our world too. Think about bankers, marketers, investors, people who, until very recently, they never had a reason to open a coding agent at all. And that non-developer group, it's growing faster for open AI than their developer core. In fact, it's growing three times faster. Now, think about that. Codex launched as a coding agent. I mean, that's really what it is. It's built around writing and running code. And somehow, non-developers are now its fastest growing user segment. Now, is this surprising? Probably not really, right? I mean, many of you, non-developers watching, you've probably dug into clawed code for various reasons. I've used clawed code to create an AI developmental editor for the book I'm writing right now. Um, so these days, for non-developers using codecs, it's not like they're using it to develop code. In many cases, um, I suppose maybe they could, but I'd say in many cases, they're using it to probably automate much of their work. It could be things like moving data from one tool to another, or auto-generating specs from research, or turning a Figma file into a structured handoff doc without going through an engineer. Um, basically building lightweight automations across the apps that they're already using. Now, along with releasing stats on their fastest growing non-developer user base, OpenAI also announced six new role-specific plugins with different apps, skills, instructions, and workflows across 62 applications and 110 skills. Now, according to OpenAI, um, I'll give you an example of some of these plugins, like the Data Analytics plugin. It helps explore product and business data and creates reports using tools like Snowflake, Databricks, Genie, Hex, Tableau, uh, the Creative Production plugin that helps marketing teams turn briefs into campaign assets using Figma, Canva, Shutterstock, Pixart. Um, now the product design plugin turns early ideas into prototypes, reviewable right in Figma and Canva. And there's additional plugins for corporate finance, private equity investing, marketing strategy, strategy consulting, legal, that's all coming down the pipeline. Now, when we hear 110 automated skills, it sounds like a lot, but like, what does it actually mean? It means a designer or a product manager can say, Hey, pull the user research from our Notion doc, map it against our open Jira tickets, and surface the three biggest gaps. And then Codex can run that across the three separate tools without any engineering help. You don't have to actually go to an engineer to make use of that. You don't have to write a single line of code to get there either. The agent does the work. Human reviews the output and makes the call, but the agent does that early work. Now, we're not talking about the future. For a lot of people in product and design roles, you can actually do this right now. So, why should this matter to you? Well, first, if you're building anything that product teams or designers use, maybe it's a collaboration tool, a design system, a product management platform, a road mapping product, you can no longer assume that your power users are technical. Let that sink in for a minute. I mean, the non-technical user who used to need a technical champion inside of engineering to get any value out of your tool, they may actually be the one to use your product now. So, what does your onboarding look like for them? You know, what does your AI feature set look like for this non-technical group? You might want to think twice about building it out exactly the same way you would for, say, your developer core. May need to be totally different. It's also a great reminder that in this AI everything world that we're in right now, we have to constantly be watching who's really using our product and getting value from it. It used to be that we'd build a product for a very specific group that had a very specific problem and we focus on that group, right? We we get to product market fit. It might take months, heck, might take years, and then we expand. But things are moving so fast. I mean, my take is that we don't have time to ignore other user segments who are clamoring for our products early on. Just like these AI tools are moving fast, the needs of our customers are moving even faster. We have to make sure that we're ready to serve them. So that's one thing for you to take away from story one. Um, let's move on to story two. Okay, story number two, it's about Claude's release of its brand new AI model and the trapdoor that came with it. Earlier this week, Anthropic released Claude Fable 5. This is the first publicly available version of their Mythos class model. This is the most powerful AI model that they've ever shipped. It had been locked up since April, available only to a small group of vetted cybersecurity researchers under very tight controls. In fact, when Anthropic announced Mythos in April, they said it was so powerful that if it got into the wrong hands, like some black hat hackers, things could be bad. Very, very bad. And of course, that's when government officials freaked out, that's when executives started to panic, all thinking about the tech that was going to wreak havoc on all of us. And hey, look, I admit I had some sleepless nights just wondering what this all meant, what our future was going to look like. Anyway, now it's just available. It's available to anybody because Anthropic released Claude Fable V of the Mythos class. But what does this really mean? Now, I should say, they didn't just release it as is, they didn't make these, you know, claims before about it being too powerful and it can't get in the wrong hands, then today they're just like, eh, actually, have at it. No, it's that's not the case. Look, Fable V isn't just a capable model with a new name. Anthropic built something structurally different inside of it. A safety trapdoor. And here's how it works: when a query comes in that touches a high-risk piece of territory, think like cybersecurity, biology, chemistry, classifiers intercept it before it reaches the full model and it routes it to Claude Opus 4.8 instead. Now, you're notified when this happens. You're not going to be charged for the Fable 5 prices for the rerouted request. System doesn't just like swap models on you without you knowing. But what does it mean for you as a product builder? Well, first, I do want to outline there is some opportunity here, right? Like you now have access to a genuinely frontier class model, the same capability tier that was locked behind Anthropic's most restricted research program just two months ago. I mean, we're just talking about April when they were saying, nope, the public can't have access to this yet. Now, for most product teams, this could be amazing. You know, longer reasoning, better vision, sustain multi-step tasks, things that used to require significant prompt engineering or chaining multiple models together, Fable 5 just handles it. But with most models, there are two states your product has to handle. The model helps or the model refuses. You design your UX around both of those states, but Fable 5 introduces a third state. What happens when the model hands the query off to a different model entirely? One with a different capability, different latency, potentially different output quality. Now, your user might not notice, but you will when the experience feels inconsistent. So the question for product builders is not just like what can Fable V do, but what happens in my product when Fable V decides that it can't serve my product? And if you're building anything that touches security, biotech, chemistry, uh research tools, you need to actually map out that fallback experience. Don't let anthropics architecture make the UX decision for you by default. You have to own it. You're gonna have to design for it explicitly. Now, the teams that treat this as a feature and not a constraint, they're gonna be able to build more resilient products because your users don't actually care what model answered their need, right? They care about whether your product works. Now, there is an aside here. Um, I think the timing is kind of wild when this happened because the same week this happens, Anthropics executives or you know, two top officials, they published this blog post saying, and I'll paraphrase it, that it might be good for the world to slow down or even temporarily pause frontier AI development to let society and safety research catch up. Specifically, this came from Anthropic's head of research institute, Marina Favaro, and one of their co-founders, Jack Clark. They suggested that AI labs should build international agreements and countries can work together to monitor how AI is developed. And they stressed that the world doesn't have decades to figure this out. So let's recap this, right? Anthropic releases their most powerful public model ever. And also they said, hey, maybe we should pump the brakes, or at least everybody else should pump the brakes. This is a wild world we live in, but I think it reflects something very real about where we're at within AI right now. The people building this technology, maybe they are uncertain about it. Maybe they're not pretending everything is fine. Um, yes, they are releasing the model, but they feel the need to sound the alarm at the same time. Or maybe Anthropic is just preparing for a well-timed IPO. Maybe all of the above. You decide on all this. I just think it's really interesting timing. All right, one final story of the day. All right, final story of the week. A big, giant Siri reset. Because recently at Apple's Worldwide Developer Conference, there were a ton of announcements that were made by Apple, but the big one focused on Siri AI. A ground up rebuild of the assistant that, let's be honest, it's been struggling for years. I mean, hey Siri, set a timer. That was pretty much been the ceiling of what we could trust Siri to do. Um, if you asked it anything complex, it would either just search the web or it wouldn't even understand what you were asking for in the first place. Now, Apple was sort of one of the first to introduce what a voice agent could actually feel like, but it was pretty much just the voice, not the agent, right? But those days might very well be over. The new Siri can hold a real back and forth conversation. It keeps and maintains context. So if you ask a follow-up question, it knows what you're actually talking about. It pulls from your personal data, emails, messages, photos, and it can take action across apps, not just open them, but actually do things inside of them. That might be rescheduling a meeting, uh, might be adding something to a shopping list, or pulling a flight confirmation from your email and actually checking you in for your flight, which is something I need to do right after I record this episode because I'll be headed to London for Mind the Products MTP Con. Um, anyway, this is all a fundamentally different product than what Siri was two weeks ago. Now, there was something about this announcement that was kind of a buried lead of sorts. Google Gemini is actually powering everything. That's right. Apple, the company that really spent the last several decades building the world's most tightly controlled hardware-software ecosystem, the company that treats its supply chain basically like a state secret, that famously doesn't really play that well with others. It's partnered with Google to power its flagship AI feature. Now, Apple has its own models. Apple Intelligence has been running on device processing for a year now, over a year actually. But they chose for the most capable parts of Siri AI to go external, to go to Google specifically. And what does this tell us? Well, it tells us that even Apple doesn't think it can win the model race alone. That, yes, they built in-house models, but compared to what Frontier Labs are actually producing, those are two totally different tiers. And they'd rather partner than compete. And Google's Gemini is capable enough and trusted enough that Apple is willing to actually build its flagship-facing products right on top of it. But what does it mean for you as a product builder? Well, let's start with the obvious. If your product lives on iOS, if you're building for consumers or knowledge workers, it probably does. The assumption that Siri was too limited to compete with your in-app functionality, that's gone. You can't really think like that anymore. Now, for years, that was kind of emote, right? Siri couldn't do what your app does. So users had to open your app. They had to navigate your UI, they had to use your features directly. That's how the engagement happened. But now that's changing. Maybe it's not gone overnight. I mean, Apple has a well-documented history of announcing things at their worldwide developer conference. And then it takes 12 to 18 months to actually ship it at scale. But we now know what's coming. And the teams that actually wait for it to be shipping before they start thinking about it, you're going to be 18 months behind. Now, I think another interesting design question is you know, not just can Siri replace my app, but it's more, what does my app become when Siri is the interface layer sitting on top of it? Because Siri AI is built with app integration in mind. Developers can expose specific capabilities directly to Siri, letting users trigger features in your app through conversation without ever actually opening it themselves. Siri can do more inside your product. That's an opportunity for you, right? That's maybe it's not a threat at all. It could be the biggest opportunity for your team, depending on what your product is. Now, most teams might read the WWDC news as like, hey, there's a new Siri, and you just sort of file it under, you know, Apple stuff to deal with later. But if you go the other direction, if you actually start asking, okay, how do we make our product work beautifully inside the new Siri experience? I think you're going to be in a completely different position when this actually rolls out, when this is all being used by everybody. And another thing on the Google partnership, I mean, every Siri interaction on an Apple device is at some level running through Google's infrastructure. I mean, two of the biggest AI competitors in the world, now they're sort of stitched together inside of a device that's inside over a billion people's pockets. The privacy, competitive, and ecosystem implications of that, we're gonna have to see how it all plays out, but this is what's happening, and it's now a part of the environment that you're building in. That's going to do it for today's episode of Now Shipping. If you enjoyed today's episode, if you got value from it, please subscribe to the channel, spread the word, share it with a friend. Seriously, every little bit helps. So that would mean a lot to us. And if you want to help improve the show, leave a comment, uh, share your feedback. I'm definitely open to it. We want to make this even better. I promise you, I will read any comment that you leave here. So, with all of that, once again, my name is Mike Belsito, and from the team at Mind the Product, this is now shipping.