As breakthrough technology becomes harder for consumers to understand, they will try aggressively to break it.

When I first handed my colleague ChatGPT Advanced Voice, he glanced at me nervously. As a pulsing black dot turned into a whispy blue orb filled with cloud-like smoke on the screen, his mouth went ajar. “Talk to it,” I implored. “Ask it anything.” A smile shot across his face and his eyes went bright. I felt a smile of my own crawl up from my stomach to my throat, ready to burst out. I’m witnessing a magical moment, I tell myself. At last, he pulls the phone speaker towards his mouth and exclaims, “Yo quiero una cerveza,” which he then repeats five times.

This colleague of mine is neither dumb nor, as you might suspect, old. He’s twenty-seven years old, found and sits on the board of a few of our most valuable portfolio companies, and understands deep technology like silicon photonics and advanced semiconductor packaging better than almost anyone. After requesting a six-pack with six sentences, he handed my phone back to me and said, “Yeah, that’s really cool.”

This is just one example of a phenomenon I’ve noticed from investors, engineers, and consumers alike: When a technology is so cool that people don’t know what to do with it, they default to trying to break it. They are then impressed if it doesn’t break, often without testing the technology’s full range of capabilities.

“Try to break” is a better heuristic than it probably sounds. In reality, my colleague was trying to gauge how deterministic the model was, see if it could handle impolite folks who interrupt the model in a customer service setting, and test whether it could switch between multiple languages fluently. As this example illustrates, though, building for resilience rather than feature testing is a completely different mindset.

Rephrased: You can’t move fast and break things in deep tech. There are upsides to this style of company building. Primarily, when a product passes multiple rounds of testing, its revenue ramp can be violent. Old hardware and software infrastructure that was built with more public trial and error—like CDNs, supply chains, and social media—can be immediately leveraged as distribution channels for your unbreakable tech. I firmly believe you’ll see several hardware companies jump from under $10 MM to over $100 MM in revenue in less than a year’s time1. (Hopefully, they are companies in IAG’s portfolio.) You are already seeing this happen with SaaS companies:

Anu Atlaru’s essay “Taste is Eating Silicon Valley” asserted that “[e]ven in the most cutting-edge technical fields, taste is shaping the future as much as the technology itself.” Consider this a corollary: Taste wins the day, but your tech can’t effing break. Whether a product breaks under pressure or not is a binary. Your product won’t be globally deployed if it is still multiple iterations away from not breaking. OpenAi wouldn’t be using Mercor if it wasn’t confident it wouldn’t break.

Barriers to entry are increasing, but barriers to massive distribution are decreasing. Make sure your product is ready for the big leagues.

Addendum for investors

Investors: This is a call for you to update your priors and re-evaluate how you underwrite hardware startups. Unlike software, a "new kid on the block" won't displace a hardware company that is already designed-in and validated for next year's customer design. It may be possible to wait for signs of exponential revenue growth before investing in software, but that will not work in hardware, where costs for fabrication or manufacturing are front-loaded and revenues jump 10x as inventory is sold and then pre-sold. Unless you are a growth stage investor, the time for a "good deal" is before the second derivative of revenues hits 1000%.

  1. However, particularly for hardware companies, the R&D required is front-loaded. These massive deployments will be 2-4 years in the making. I strongly believe there is a massive inefficiency in venture capital where most VCs undervalue enginering-intensive partnerships that hardware startups have with large potential customers. Hardware, particularly for companies worth hundreds of billions or trillions of dollars, is a small universe. You can count on one hand the candidates for a product SKU coming in 3 years. If a product is a year or two out, feature lock likely happened a long time ago. As such, your revenue growth won't matter, until it explodes. Hardware is not software: You can’t just swap out one component for another in a few weeks. I'd rather see a hyperscaler give a startup 20 engineers than $1 MM. ↩