Who’s Defining the New Default: Speakers’ Highlights (Part V)

From Wrappers to World-Models: How to Build a Physical and Secure Infrastructure for AI?

The architecture of the AI era is being built on three pillars: security-first engineering, embodied intelligence, and the geopolitical economics of open source. In this article we explore how practitioners we talked to, are moving beyond "wrapper" apps toward resilient, sovereign, and physically integrated AI systems.

What happens when AI moves from a chatbot window into the browser, into our glasses, and into the foundations of national policy? As the novelty of LLMs fades, the focus has shifted to the "boring" but critical work of making these systems safe, sustainable, and physically useful.

The New Default speaker series examines the infrastructure required to support a world where AI agents act on our behalf, and hardware finally catches up to software's intelligence.

Beyond the Hype: How Security, Hardware, and Geopolitics are Shaping the Next Era of AI

The New Default is about the structural integrity of the systems that host the chatbots. As we move away from simple API wrappers and toward deep integration, the conversation has shifted toward the heavy lifting of engineering. We sat down with three experts who are redefining these boundaries: exploring how to secure autonomous agents that navigate the web, how to move AI out of the screen and into wearable hardware, and how the "national bet" on open-source models is reshaping global digital sovereignty.

Donato Capitella: The Security Architect of Agentic AI

Donato Capitella, a leading voice in AI security, focuses on the "agentic" shift, where AI doesn't just talk, but acts. As we give LLMs the power to browse the web and execute code, we open a Pandora's box of browser-based vulnerabilities. Here are the areas Donato expanded in our talks: 

• LLM Guardrails as Infrastructure: Donato argues that guardrails shouldn't be an afterthought. They are the essential filtering layer that prevents prompt injection and ensures model outputs remain within safe operational bounds.

• The Browser-Based Agentic Attack: He warns of a new class of threats where AI agents can be manipulated by the very websites they browse. An agent tasked with "summarizing a page" might be tricked by hidden text into "deleting the user's account," requiring a fundamental rethink of web security.

• Spikee: Testing for Non-Determinism: To solve this, Donato introduces Spikee, a testing framework designed specifically for the unpredictable nature of LLMs, allowing developers to stress-test agentic workflows before they hit production.

"The output of the LLM is untrusted input into your system. You need to treat the output of any LLM as if it came from an untrusted source." 

Bobak Tavangar: Hardware, Software, and the Vision of Embodied AI

Bobak Tavangar, Co-founder and CEO of Brilliant Labs, brings a contrarian perspective to the "AI in a box" trend. His vision centers on how AI can finally bridge the gap between digital intelligence and physical experience. Here’s what you can lear from Bobak:

• AI Glasses vs. VR: While the tech world chases high-fidelity VR graphics, Bobak makes the case for AI glasses. His argument? We don't need to be transported to another world; we need our current world to be smarter. AI glasses provide a "heads-up" interface where intelligence is felt through context and audio rather than just pixels.

• Embodiment and Technical Architecture: Bobak highlights that true AI embodiment requires a tight loop between hardware and software. The "New Default" for hardware is a system that isn't just a peripheral but a sensory extension of the AI model itself.

• Community-Driven Hardware: He emphasizes that the most resilient technical architectures are those built in collaboration with developer communities, advocating for open-source hardware that allows for rapid iteration and "embodied" experimentation.

The glasses don't render new worlds. They try to understand the one you are actually in... it's sort of the best form factor for giving an AI agent a front row seat to that life.

Elizabeth Seger: The Geopolitics and Economics of Open AI

Elizabeth Seger, a specialist in AI ethics and digital policy, shifts the lens toward the macro-scale. She explores why "open source" has become the primary battleground for AI dominance. Here are Elisabeth’s insights on the safety and economics of Open AI:

• The Economics of "Free": Elizabeth demystifies the open-source movement, noting that companies often open-source models to commoditize complementary products (like compute or cloud services). It is a strategic play to set the industry standard and prevent vendor lock-in by competitors.

• AI Sovereignty and the National Bet: She introduces the concept of AI sovereignty—the idea that nations must own or control their AI infrastructure to ensure economic and security independence. For many countries, investing in open-source AI is a "national bet" against the hegemony of a few private tech giants.

• Regulation as a Safety Floor: Elizabeth argues that while innovation moves fast, regulation provides the necessary safety floor. The next phase of AI development will be defined by how well we balance the "democratization" of open source with the "guardrails" required to prevent systemic risks.

If there's something you develop that you really want to drive a market demand for, one way to drive that market demand is to make the compliment to that product completely free... You commoditize the compliment to the products you really want to sell.

Why the AI Advantage Belongs to the Architects

The insights from Donato, Bobak, and Elizabeth point to a singular conclusion: the "New Default" isn't just about using AI; it's about owning the architecture. We are moving past the era of the "magic trick" and into the era of the "utility." The "New Default" dictates that the true competitive advantage no longer lies in the ability to write a clever prompt, but in the capability to architect the bedrock beneath it. 

Whether it is securing the digital pathways of autonomous agents, weaving intelligence into the physical fabric of our daily wear, or navigating the high-stakes chess match of global AI economics, the focus has shifted from what AI can say to what it can do safely and reliably. In this new epoch, the winners won't be those who merely consume AI, but the architects who build the resilient, embodied, and sovereign systems that allow it to finally inhabit our world.

Explore the full interviews and deep dives at The New Default.