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
OpenAI Chips, Google Agents, and AI Supply Chain Risk for Builders
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
OpenAI just unveiled its own custom inference chip, signaling a big play for cheaper, faster AI. Meanwhile, Google's making agentic AI dramatically easier to build by integrating computer use directly into Gemini. And uh in a surprising turn, even the NSA found itself cut off from a frontier model overnight. We're talking about real shifts in the AI landscape today, from hardware to policy to practical agent building. No BS AI briefing brought to you by Proactive AI. Welcome back. I'm your host, Vikash Sharma, and this is where builders get straightforward AI news without the fluff. Alright, let's dive into what's been shaking up the AI world for us builders. We've got some high signal items that really highlight the evolving dynamics. First up, OpenAI just unveiled its first custom inference chip, codenamed Jalapeno, developed with Broadcom. This isn't just a fancy press release, it's a strategic move. OpenAI says these chips are designed specifically for real-time coding workloads, and early tests suggest they're offering better performance per watt than current alternatives, although they're still very much in the testing phase. For us builders, this could mean a significant shift. Potentially, we're looking at cheaper, faster, and more reliable OpenAI inference APIs down the line. It's a clear signal of vertical integration with OpenAI looking to reduce its heavy dependency on Nvidia's GPUs. If they can bring down the cost and boost the speed of their underlying infrastructure, that directly translates to better margins for them and hopefully more competitive pricing or more powerful capabilities for us consuming their APIs. Next, Google has integrated computer use directly into Gemini 3.5 flash. This is a huge step for Agentic AI. What this means in plain English is that now a Gemini agent can actually see, reason, and act across your browser, mobile, and desktop environments. It's not a standalone model anymore. It's a core feature available via the Gemini API and their Gemini Enterprise Agent platform, complete with enterprise level safeguards. Think of it like a truly intelligent virtual assistant that can navigate your UI, click buttons, input data, and understand context from what it sees on your screen, not just what you type into a prompt. For builders, this significantly lowers the complexity of multi-step automation. Suddenly, a single API call can drive incredibly robust agent behavior across different applications, making it far more practical to build production grade workflows that interact with legacy systems or web UIs without custom integrations. Also a fascinating and somewhat concerning story. The NSA reportedly lost access to Anthropics Mythos V after US export controls were put in place. According to the New York Times, Mythos V was a model the NSA used for rapidly finding software weaknesses. The critical point here is that the US government, even for its national security functions, relies heavily on private AI companies. When export controls were imposed on Anthropic, it created an immediate tangible supply chain disruption for a critical government agency. Now, why does this matter for EUs? It's a stark reminder that access to frontier models or even specialized models can be cut off abruptly, not just for geopolitical reasons, but potentially due to regulatory shifts, supplier changes, or even a provider's internal policy changes. Builders need to prioritize contingency plans and diversification in their AI stack. Can you switch providers if one becomes unavailable? Do you have an open source fallback? It underscores a very real non-hypothetical risk. Moving on. This isn't just another SaaS offering. It's a full-stack solution designed for high security, air-gapped environments. It includes hardware agnostic compute, multi-agent orchestration, a crucial human-in-the-loop governance layer, and mission-specific applications. The sovereign part is key here. It's built for entities that need complete control over their data and AI operations, often due to national security, regulatory compliance, or proprietary data concerns. For builders, this highlights a rapidly growing demand for tech sovereignty and compliance-ready AI stacks. If you're building for sectors like finance, government, defense, or even just large enterprises with strict data residency rules, you absolutely need to be designing for auditability, robust security, and the ability to keep data and models within specific geographical or organizational boundaries. Finally, by Jan Lakun is once again arguing that open source is the only viable path for AI. The meta chief AI scientist advocates for a federated approach, exemplified by what he calls project tapestry. This concept allows for contributions via parameter vector exchange, meaning models can be improved and shared without ever having to share raw, sensitive training data. His argument is that for most of the world, proprietary, closed-off AI models just aren't sustainable or accessible long term. From a builder's perspective, this bolsters the case for investing in open source infrastructure. If you are concerned about resilience, data sovereignty, or ensuring global accessibility for your products, then leaning into open source models and contributing to that ecosystem could be a crucial strategic choice. It's about building a future where AI isn't locked behind a few corporate gates. Alright, so if I had to pick one story that really signals an inflection point for builders this week, it's got to be Google's integration of computer use into Gemini 3.5 flash. This isn't just an incremental update, it fundamentally changes how we can think about building intelligent agents. What happened is that Google took what was previously a more standalone or complex agent infrastructure and integrated the ability for their Gemini 3.5 flash model to see, reason, and act across any digital interface, browsers, mobile apps, desktop applications. It's no longer just a language model talking to APIs. It can effectively operate a computer like a human would, but at machine speed and scale. This capability is now directly accessible via the Gemini API and their enterprise platform, complete with those necessary safeguards and guardrails. Why does this matter right now? Well, for years the promise of agents has been intriguing but often fell short in practice, especially for complex multi-step tasks that cross different applications. You'd need a web scraper here, a custom API integration there, a bunch of conditional logic. It was a nightmare to build and maintain. Now, with computer use embedded directly, it significantly streamlines the development of sophisticated production grade agents. This isn't just about chatbot automation. It's about automating entire workflows that involve human-like interaction with software. Think about customer support agents that can actually log into a CRM, pull up order details, and initiate a refund, all driven by a single AI model. Or internal tools that can onboard new employees by navigating multiple HR systems. The market for agentic automation just got a massive catalyst. So who should really care about this? Founders absolutely need to pay attention because this opens up entirely new product categories for automated workflows or allows them to build more comprehensive done-for-you solutions rather than just API wrappers. By me, the product manager's job should be evaluating how this new capability can be woven into existing products to add new powerful features that were previously too complex or expensive to build. Imagine adding a just get this done for me button to your SaaS product that actually automates tasks in other applications. Engineering leaders will find this interesting because it centralizes agent capabilities, potentially reducing the need for disparate automation tools and simplifying the architecture for integrating AI into operational workflows. And for tense indie hackers, this is a playground. You can prototype full stack automation ideas much faster, potentially building incredibly useful tools with far less code than before. How would I think about this as a builder? I think of it as moving from an API-centric world to an agent OS world for many business processes. Instead of mapping out every single API call and UI element, you're now giving the AI a high-level goal and letting it figure out the steps across different interfaces. The opportunity is massive for internal tool automation, things like automating data entry, report generation, or cross-platform data synchronization. You can build much richer customer-facing experiences too, where the AI isn't just answering questions but executing tasks. On the risk side, you've got to watch out for vendor lock-in. Google is making a strong play here, and while powerful, relying solely on one provider always has its downsides. Also, latency and cost for these kinds of multi-step visual interactions can be higher than simple API calls. I'd caution against getting swept up in the hype of full AGI agents just yet. This is about practical, task-oriented automation. My no-biest take on this is that it's a genuine step forward. It isn't AGI, but it's a significant enabler for building highly capable real-world agents. Builders who can grasp how to effectively prompt and constrain these computer use agents will be able to unlock serious value both for internal efficiencies and for entirely new product offerings. Don't underestimate the power of an AI that can actually click buttons and read your screen in a reliable way. If you want one practical takeaway from today's episode that you can act on this week, here it is. Experiment with building a 10-minute automation prototype using Gemini's computer use. This directly leverages Google's new capability and can give you immediate, tangible insights. Here's how to try it in under 60 minutes. Step one. There's step one. Identify one mundane repetitive workflow in your team. Think about something someone does daily or weekly that involves clicking through a browser, copying data, or interacting with a simple desktop application. Maybe it's pulling specific data from a web dashboard into a spreadsheet or automatically populating a new project in your project management tool based on an email. Choose something with two to three distinct steps across different apps if possible. Step two, access the Gemini API with the computer use capability. Google has made this available, so your first task is to get API access and understand the basic prompt structure for this feature. You'll need to grant the agent the necessary permissions to interact with your chosen environment, which is part of the enterprise safeguards they've built in. Dumma Step 3. Prompt the agent to perform one, two steps of that workflow. So don't try to automate the whole thing at once. Start simple. Give it a clear goal and show it the first couple of actions or describe them in detail. For instance, navigate to this URL, login with these credentials, and click on the export report button. Observe how it interacts. Step 4, track the initial results. Even a short prototype can reveal a lot. How much time did it save you for those 1-2 steps? What was the error rate compared to manual execution? Crucially, what was the cost of running that operation via the API? This initial data will inform whether this approach is viable for larger, more complex automations in your product or internal operations. Why is this specific experiment worth your time right now? Because understanding this capability firsthand is the fastest way to grasp its potential. It's not about theoretical AI, it's about what it can do for your team or your users. It will help you see if this agent OS approach can truly reduce manual effort, improve accuracy, or enable entirely new product features that rely on cross application interaction. Don't just read about it, try it and see the impact. That's it for today's NoBS 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.
