No‑BS AI Briefing
No‑BS AI Briefing is for builders who don’t have time for hype. Each episode focuses on a handful of high‑signal stories in AI and AGI, unpacked in simple language with a builder’s perspective. You’ll hear what changed, why it matters, and how you can experiment with the tools, ideas, or strategies yourself—whether you’re leading a team, shipping a startup, or exploring AI side projects.
No‑BS AI Briefing
No-Code Voice AI: xAI's Builder, Meta Cloud, & Cheap Media Generation
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Imagine voice agents becoming as easy to build as a chatbot. And what if generating high quality images and video could cost you just pennies for thousands of assets? That's where we are heading today as we break down some big moves in AI that are shaking up what's possible for your products and your teams. No BS AI briefing brought to you by Proactive AI. Welcome back. I'm your host, Vekas Sharma, and this is where builders get straightforward AI news without the fluff. Alright, let's dive into the headlines for this week because we've got some really interesting shifts happening that directly impact how you build. First up, a quick but significant development on the policy front. The US Department of Commerce has lifted export controls on Anthropic's Fable 5 and Mythos 5 models. Now, if you remember just a couple of weeks ago on June 12th, these models were placed under restrictions due to a detected jailbreak vector. What happened is Anthropic collaborated super quickly with the US government, retrained its safety classifier to effectively block that vulnerability. And now, as of July 1st, these frontier models are globally available again across all of Anthropic's platforms, Cloud Platform, Claude AI, Cloud Code, and even Cloud Cowork. For builders, this is a pretty strong signal. It shows that access to these cutting-edge models can and does hinge on policy compliance and safety measures. It's a bit of a tightrope walk for companies like Anthropic balancing innovation with global regulatory demands. But it also sets a precedent. If you can address a vulnerability quickly and robustly, those controls can be lifted, restoring global availability. It reinforces that regulatory risk is a very real factor in your product planning, but also that fast, effective safety fixes can resolve these issues without necessarily degrading the model's core capabilities. Next, a very exciting launch from Xi. They've rolled out their voice agent builder in beta and its no code. This is a big one because it really unifies the entire stack needed for Voice AI. We're talking speech to text, the large language model itself, and text-to-speech, all integrated with subsecond latency. Think about what that means for real-time interactions. It's crucial. The platform also includes built-in telephony, retrieval, augmented generation, tool use, guardrails, and observability features. And here's the kicker on pricing. It's just 0-05s per audio minute. For builders, this is a game changer because it completely removes the headache of integrating multiple APIs from different vendors, each with their own quirks and latency profiles. That sub-second latency and the low transparent pricing mean you can now rapidly prototype and even ship production grade voice agents for things like customer support, accessibility features, or internal productivity tools without breaking the bank or needing a team of specialist AI engineers. The no-code aspect is also massive. It significantly broadens the pool of people who can actually build and deploy sophisticated voice agents. Moving this capability out of the deep engineering labs and into the hands of product managers and even indie hackers, also in the realm of specialized AI tools, Databricks, and Nvidia have teamed up to launch Genesis Workbench specifically for drug discovery. This is an open source blueprint that integrates proprietary data sets, Nvidia's BioNemo models, and GPUs all within a unified Databricks workspace. They're bringing in tools like MLflow for experiment tracking and Unity Catalog for data governance. The Workbench supports complex workflows like genomics, protein engineering, and small molecule discovery, all while keeping sensitive pipelines in secure, isolated environments. Now, why does this matter for you even if you're not in biotech? It signals a clear trend towards highly integrated domain-specific AI workflows. We're moving beyond generic LLMs into platforms designed to handle the unique data models and compliance needs of specific industries. For builders, it means less tooling fragmentation. You're not stitching together a dozen different services. It also hints at a future where similar integrated platforms could emerge for other complex verticals, giving us a template to follow when thinking about how to build industry-specific AI solutions. Shifting gears to infrastructure, Bloomberg is reporting that Meta is planning a cloud infrastructure business to sell AI compute. Now, this is fascinating. Meta has invested massively in data centers and GPUs for its own internal AI needs, and it seems they're looking to monetize that excess capacity by offering it to external customers. They'd be going head-to-head with the big players like AWS, Azure, and Google Cloud, which is a bold move. But think about the implications for you. It means a potential new source of AI compute that could be significantly more cost competitive, especially given Meta's sheer scale and efficiency in running its own infrastructure. It also opens up the opportunity for managed access to Meta's own models, Lama perhaps, and whatever else they're building directly on their infrastructure. This move really underscores a broader trend we're seeing. Hyperscalers aren't just selling raw compute anymore, they're moving towards offering a full stack of AI services from infrastructure to models and everything in between. It could shake up the market dynamics and provide more choices for where you host and run your AI workloads. And finally, on the model front, Google has launched Nano Banana 2 Lite and Gemini OmniFlash. Let's unpack these. Nano Banana 2 Lite is all about image generation, capable of producing images in under 4 seconds and at an incredibly low price, $0.034 per 1000 images. That's not a typo per thousand images. Then you have Gemini OmniFlash, which is designed for video editing priced at 0.110 per second of video. Both of these are available through the Gemini API and Google AI Studio. So why should builders pay attention? Because ultra low cost and low latency media generation fundamentally changes what's possible in terms of creative features for your products. Suddenly, real-time image and video generation at scale isn't just a fantasy, it's an economic reality. That aggressive pricing makes high-throughput creative workflows not just viable but potentially central to your product's value proposition. It dramatically broadens Google's generative ecosystem, offering new building blocks for integrating creative AI directly into your applications, whether that's for personalized content, dynamic marketing materials, or user-generated creative tools. Now, out of all those stories, the one that really stands out for me and that I think represents a true inflection point for builders is XAI's new voice agent builder. This isn't just another voice API, it's a no-code unified platform that could fundamentally change how we think about building voice-first experiences. What happened is XAI, leveraging its Grok voice technology, has released a beta platform that brings together all the complex pieces of a voice agent, speech to text, the underlying LLM, and text-to-speech into one cohesive sub-second latency system. On top of that, they've included critical features like telephony integration, retrieval capabilities for connecting to your data, guardrails to keep interaction safe, and observability tools to understand how your agents are performing. And they've priced it very aggressively at just 5 cents per audio minute. Why does this matter right now? Well, for years, building a sophisticated voice agent has been a multi-API integration nightmare. You'd be stitching together different services, dealing with varying latencies, and wrestling with inconsistent pricing models. That fragmentation meant higher development costs, longer time to market, and often a sub-par user experience due to noticeable delays. Excise builder streamlines all of that. It takes a previously complex multi-vendor, multidisciplinary engineering challenge and productizes it into a single no-code interface. This isn't just an incremental improvement, it's a paradigm shift towards making high-quality real-time voice AI accessible to almost anyone. It's commoditizing the underlying components and emphasizing the end-to-end product experience. So who should really care about this? Founders and product managers absolutely need to pay attention. This dramatically lowers the barrier to entry for voice-first features. Can your customer support offer instant natural voice assistance? Can your app offer voice navigation for accessibility? These become viable, budget-friendly options. Your teams can focus on core product logic and the unique value you provide. Indie hackers and small teams, especially. This could be your moment to build truly differentiated products with Voice UX without needing a huge dev budget or a deep bench of AI specialists. The no-code aspect is huge for rapid prototyping and iteration. Even yet, incumbents in the CCIS contact center as a service and existing voice AI spaces need to watch this closely. They're now facing pressure on both price and simplicity from a challenger that's effectively democratizing advanced voice capabilities. How I'd think about it as a builder is this. For too long, voice AI has felt like a luxury, something only the biggest companies with dedicated research teams could really perfect. But this is the moment where it becomes a utility. Think about how easy it is to spin up a chatbot today compared to five years ago. This is that same transition happening for voice. The strategic implication for startups is massive. You can now differentiate your product with a smooth responsive voice experience at a fraction of the historical cost and effort. For bigger companies, it's about compressing time to market for new features and improving operational efficiency through voice automation. The shift towards unified opinionated platforms like this reflects a broader trend in AI where API commoditization is driving value up the stack to the integrated experience itself. Don't get stuck in the weeds of building infrastructure that's rapidly becoming a solved problem. My no BS take here, this isn't hype. The promise of sub-second latency, unified tooling, and no code access for voice agents is real and it changes the game. While it's still in beta, the direction is clear. Voice is becoming a default input slash output modality, and the tools to build it are getting dramatically simpler and cheaper. Your users expect natural interaction and now you have a much easier way to deliver it. If you want one practical takeaway from today's episode, here it is. Experiment. Use Xi's no code builder to prototype a voice agent in 60 minutes. Here's how to try it in under an hour. One, sign up and get access. Head over to XIS website, find the voice agent builder beta and get yourself an account. They're making it easy to onboard. Two, define a simple use case. So don't try to build the next Jarvis right away. Start small. For example, build an agent that can answer your company's top five frequently asked questions, or one that can guide a user through a basic troubleshooting step for your product, connect it to a small curated knowledge base. Three, um ta build and test. Use the no-code interface to design the conversation flow, integrate your knowledge base, and set up the telephony or web interface. Then immediately start testing it yourself. Call the agent, ask your questions and observe its responses. Crucially, pay attention to the latency. Does it feel natural or are there awkward pauses? Four, measure and evaluate. Keep an eye on the cost per interaction, which should be pretty transparent at 0.05 per audio minute. Also evaluate the accuracy of the responses and the smoothness of the conversation flow. Try to jailbreak it with unusual questions or out-of-scope requests just to see how the guardrails perform. Why this specific experiment is worth your time right now. This isn't just about playing with a new tool, it's about understanding a fundamental shift in user interaction. Voice is incredibly powerful for accessibility, for quick information retrieval, and for situations where a screen isn't convenient. By getting hands-on with a no-code, sub-second voice agent, you'll quickly grasp its potential and its current limitations. You'll see how easy or not it is to manage the context, handle interruptions, and maintain a natural conversation flow. More importantly, it helps you identify immediate high impact areas within your own product or internal operations where voice could provide a genuinely better user experience or significant efficiency gains, all without a massive upfront engineering investment. It's about bringing voice out of the future tech category and into the what can we ship this quarter conversation. That's it for today's No BS AI briefing. If this helped, follow the show in your podcast app and share it with one builder you know. And if you've got questions or topics you want covered, connect with me on LinkedIn and send them over. See you in the next briefing.
