October 17, 2026
Amulya Vishwanath spent seven years running product marketing for Nvidia's robotics platforms, which makes her skepticism worth taking seriously. She told TechCrunch that physical AI is still in what she calls its 'Lego era' -- the basic pieces, actuators, robotic hands, software stacks, safety standards, are still being built and rebuilt, not assembled into something that just works. That's a direct rebuttal to the popular narrative that one giant foundation model is about to make general-purpose robots an overnight reality.
I think she's right, and I think business readers should actually find this reassuring rather than disappointing. Hype cycles make it tempting to delay automation decisions while waiting for the 'real' breakthrough. But if the fundamentals, hardware, safety, interoperability, are still being figured out at the infrastructure level, then the near-term returns on AI investment are still in software: workflows, internal tools, and the systems that run a business day to day. That's not a consolation prize; it's where the ROI actually sits right now. If you're evaluating where to put automation dollars in 2026, workflow automation inside your existing operations will pay off faster than betting on humanoid hardware that's still assembling its own Lego set.
Veritas is expanding its anti-counterfeit chip beyond its original use case into jewelry, trophies, and fragrance bottles, according to TechCrunch. The pitch is straightforward: as counterfeiters get good enough to produce 'superfakes' that fool even trained eyes, brands need a way to verify authenticity that doesn't rely on logos or packaging alone, something embedded and hard to clone.
What strikes me here is less the chip itself and more the pattern it represents. We've spent two years talking about AI's ability to generate convincing fakes, images, voices, now entire product experiences. Veritas is a reminder that the market response to synthetic deception is going to be verification infrastructure, not just better detection after the fact. That's a useful lesson for any business building trust-dependent products, whether it's physical goods or software records. Companies that bake in verifiable provenance, clear audit trails, documented data handling, are going to have an easier time with increasingly skeptical customers. It's part of why we think about security and our trust center as product features, not afterthoughts. The tradeoff worth noting: chip-based verification adds cost and complexity to manufacturing, so it'll likely start with high-margin luxury goods before trickling down anywhere else.
Cal AI, the calorie-tracking app, generated more than $30 million in revenue in two years and was acquired by MyFitnessPal in March, TechCrunch reported. The detail that matters most is how it got there: not a novel model or a defensible technical moat, but a small army of paid influencers driving viral growth.
This is worth sitting with, because it cuts against the assumption that AI products win on AI quality. Cal AI's core function, estimating calories from a photo, is a feature, not a company, on its own. What turned it into a $30 million acquisition was distribution strategy executed well. For founders and small teams building AI-flavored products right now, the takeaway is blunt: your model or your prompt engineering is rarely the moat. Marketing execution and growth channels are. If you're weighing how to launch something similar, it's worth comparing how far a scrappy go-to-market can take a thin technical layer versus investing in marketing and campaigns infrastructure that compounds beyond one viral push, since influencer spend is a rented channel, not owned growth.
Which of these three stories actually changes how you'd spend your next automation or marketing budget: the slow grind of physical AI infrastructure, the move toward verified-trust hardware, or the proof that distribution still beats a clever model?
Sources