AI Agents as Engineers, Optical Interconnects, & Secure AI

In this episode, Vikash Sharma breaks down the latest in AI for builders: - **P-1 AI's "Archie":** A $50M Series A fuels an "agentic AI engineer" designed to automate end-to-end design tasks across mechanical, electrical, and thermal workflows. This signals AI moving from a mere tool to a true engineering teammate, potentially compressing product development cycles. - **GlobalFoundries' $300M CHIPS Award:** A major U.S. government investment backed by NVIDIA, AMD, Meta, and Microsoft to accelerate silicon photonics and optical interconnect research. This is critical for overcoming the copper bottleneck in AI data centers, enabling denser, lower-power, and higher-bandwidth compute clusters. - **Rakuten & HP's "Rakuten AI for Desktop":** A hybrid AI application combining a local 7B LLM with cloud-based agents for real-time services. This offers a blueprint for balancing performance, security, cost, and addressing data sovereignty in both B2B and B2C products. - **NVIDIA's Open Secure AI Alliance (OSAA):** Formed in response to an agent-related security incident, this 37-member coalition aims to develop open-source defensive tools and standardized protocols for AI agent security, speeding up safer deployments. - **Monorale AI's Funding for Multi-Model Orchestration:** A new platform focused on governance, orchestration, and compliance across diverse AI tools, reflecting the growing need to manage complex multi-model AI stacks efficiently. **Deep Dive:** We explore P-1 AI's "Archie" in detail, discussing its implications for founders, product managers, and engineering leaders looking to scale capacity and accelerate innovation. Vikash shares a no-BS take on the hype versus the practical realities of agentic AI. **Practical Takeaway:** Learn how to quickly evaluate agentic AI for a specific engineering task in your team, helping you understand its true potential and limitations without falling for marketing fluff. ---

Transcript

Copy Transcript Copied

SPEAKER_00 0:00

Your engineering team might just get a new teammate, an AI agent capable of handling full design cycles and iterative tasks. We're also diving into a massive $300 million investment that could reshape the very backbone of AI data centers and how to deploy your AI agents safely in a world of new security risks. NoBS 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 some high signal items that hit the wire recently. First up, we've got P1AI who just raised a hefty $50 million Series A round with NEA leading the charge and former GE CEO Jeff Imelt joining their board. This funding is all about propelling their agentic AI engineer named Archie. Now in plain English, Archie isn't just a code generator or a simple design assistant. It's an AI built to integrate deeply into existing mechanical, electrical, and thermal engineering workflows, automating entire end-to-end design tasks. For us builders, this is a clear signal that AI is moving beyond being just a tool in our toolbox and becoming more like a virtual teammate, potentially expanding our engineering bandwidth without the proportional hiring headaches. It also validates a strong market demand for these multidisciplinary, agentic engineering workflows that can seriously compress product development cycles, pushing ideas to market faster than ever before. Next, something a bit closer to the hardware layer. This isn't just a random grant. It's specifically aimed at accelerating research in silicon photonics, advanced packaging, and optical interconnects. And here's why it matters: this isn't just global foundries working in a vacuum. This effort is backed by giants like Nvidia, AMD, Meta, and Microsoft all collaborating to develop these near-packaged optics and co-packaged optics architectures. For builders thinking about the long game in AI infrastructure, this is crucial. It directly targets the looming copper interconnect bottleneck in current data centers. Optical links can dramatically raise bandwidth while significantly cutting power consumption in AI clusters, which is huge for scaling. It also provides a much clearer US-based path to integrate these high-efficiency optical interconnects, which will ultimately enable us to build denser, lower cost, and more powerful AI compute clusters down the line. Definitely one to watch. Also, Rakuten and HP have teamed up to unveil Rakuten AI for desktop, a hybrid AI solution specifically for eligible HP PCs in Japan. What's happening here is the launch of a hybrid application that cleverly combines a local 7 billion parameter large language model with cloud-based agents from the broader Rakuten ecosystem. So the local on-device models are handling those privacy-sensitive tasks and ensuring low latency for immediate responses while the cloud agents are powering the more complex real-time services like personalized shopping experiences or travel bookings within the Rakuten ecosystem. For builders, this provides a really concrete and practical blueprint for designing hybrid AI architectures. It shows us how to balance performance, data security, and cost effectively. Plus, it's a smart way to address data sovereignty concerns and even enable offline functionality, which can be incredibly useful for both B2B and B2C products, especially in markets with strict data regulations. Moving to a really important area, Nvidia has launched the Open Secure AI Alliance, or OSA, with a laser focus on agent security. This isn't just a proactive measure, it's a direct response to a real incident in July 2026 where autonomous agents apparently breached a platform. The alliance, which boasts 37 members, is working to develop open source defensive tools, secure architectures, and standardized protocols specifically for AI agent security. For us builders, this is huge. It means we're going to get shared open defenses, think zero trust identities, better sandboxing mechanisms, and advanced vulnerability scanning tools, all designed to make our agent deployments much safer. This alliance could significantly reduce the need for individual teams to build bespoke agent security from scratch, helping us speed up our time to production for agent-powered products without compromising safety. And finally, UK-based Monora AI recently raised $250,000 in SES funding and is now opening a larger 3.75 million EIS round. Their mission to build a multimodal AI orchestration platform designed for unified tool access and management. Their focus areas are governance, orchestration, security, and compliance across what they anticipate will be a diverse array of AI tools. This smaller funding round, but with a clear vision, reflects a critical shift we're seeing, the move toward sophisticated multimodal routing and management. Why does this matter? Because as builders, we'll increasingly need to optimize for cost, performance, and vendor flexibility across different models. For larger enterprises, especially, this kind of platform could be invaluable for ensuring centralized governance and compliance across complex AI stacks. Alright, for our deep dive today, I want to hone in on P1AI's Archie and this idea of AI as a teammate, not just a tool. It's a subtle but profound shift. What happened is that P1AI announced a significant $50 million Series A led by NEA, which is a big vote of confidence. Their product, Archie, is positioned as an agentic AI engineer built to handle end-to-end design tasks across mechanical, electrical, and thermal engineering. This isn't just about assisting an engineer, it's about the AI taking ownership of iterative design cycles within specified constraints, aiming for production-ready outputs. That's a pretty bold claim, and the funding suggests investors believe it. Why this matters right now is because we've seen AI tools emerge in engineering for things like code generation, maybe some finite element analysis or basic simulation. But they've largely been point solutions. Archie represents a move towards a more holistic, autonomous approach. It implies that for certain well-defined engineering problems, the AI can iterate, optimize, and even resolve conflicts across disciplines. Something that previously required significant human coordination and time. This could drastically compress product development cycles, allowing companies to innovate and iterate at speeds we haven't seen before. Think about the impact on prototyping and RD budgets. So who should really care about this? Well, if you're a founder ASO, especially in hardware or physical product spaces, this could be a game changer for scaling your engineering capacity without having to proportionally scale your hiring. Imagine your small team having the output of a much larger one. For product managers, this means potentially faster iteration loops, getting more designs in front of customers quicker, and focusing your human engineers on the truly novel, unsolved problems. Engineering leaders could use this to redeploy their most talented engineers towards higher value, more creative work, while Archie handles the routine, albeit complex, iterations and even that to indie hackers. In specialized hardware niches, might find access to advanced automated design capabilities that were once exclusive to large corporations. How I think about it as a builder is to view Archie not as a replacement for human ingenuity, but as an incredibly powerful engineering co-pilot for specific, repeatable tasks. Think of it like this: if you were building a custom car, you wouldn't ask an AI to invent a completely new type of engine from scratch. But if you gave it precise parameters for optimizing the existing engine's airflow, thermal management, and weight distribution for maximum efficiency, that's where Archie could shine. It thrives on well-specified constraints and optimization challenges. My mental model here is to identify the assembly line parts of your engineering process, those stages where you're iterating on known principles within a bounded problem space and then ask, can an agent like Archie automate 80% of this, freeing up my human talent for the true breakthroughs? The opportunities are huge for accelerating development and reducing human error in repetitive complex calculations. My no BS take here is that while the term agentic AI engineer definitely carries a bit of hype, it's important to remember this is likely domain-bound, not a general-purpose engineer who can invent new physics. We shouldn't expect it to magically solve entirely novel design problems or operate outside of its specified engineering domains. The limitations are real. It's going to be best suited for well-defined constraints and optimization tasks, less so for radical creative invention. And there's a significant risk involved. Design failures in production driven by an autonomous agent could raise serious liability and trust concerns demanding incredibly robust validation processes. That said, a solid $50 million series A suggests real traction and belief in its execution. The category of autonomous engineering agents is definitely one to watch, but keep your hype filters on for the generalist claims. If you want one practical takeaway from today's episode, here it is. Experiment. Evaluate an agentic AI solution for a specific repeatable engineering task within your team. This isn't about replacing your engineers overnight, but about understanding the practical boundaries and benefits of this emerging category. Here's how to try it in under 60 minutes. 1. Identify a bottleneck workflow. Sit down with your engineering lead or a senior engineer. Pinpoint one specific design or optimization task that's highly repeatable, takes significant time, and follows a clear set of rules or parameters. Maybe it's component placement optimization, thermal profile generation for a standard enclosure, or even generating a specific type of bill of materials based on a design brief. 2. Research existing tools or APIs. Spend 20-30 minutes searching for existing commercial agentic tools or even specific APIs that claim to automate aspects of this identified workflow. Look for solutions in fields like caddy automation, simulation scripting, or even AI-driven parameter optimization? Don't worry about perfect fit yet, just get a sense of what's out there. You might find a prototype or a free tier to play with. 3. Define a micro prototype goal. For your identified workflow, articulate a small measurable goal that an agent could theoretically achieve. For example, can an agent generate five valid thermal designs for X component within Y constraints in under five minutes? Or can it suggest optimal material pairings for Z load conditions? Why this specific experiment is worth your time right now? By focusing on a well-defined, repeatable task, you bypass the hype and get straight to understanding the current practical capabilities and limitations of Agentic AI. It's a low-risk way to gauge the feasibility and potential ROI for your specific product or engineering process, helping you decide when and how to integrate these powerful new capabilities without getting caught up in the abstract promises. This small experiment gives you a tangible benchmark for future evaluations. 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.