Talking AI

The episode surfaces a few genuinely useful concepts - provably correct vs. merely 'guaranteed,' hardware-in-the-loop testing, and fuzzing/concolic testing as verification layers - but the majority of runtime is consumed by general AI-enthusiasm talk, the host's own AI anecdotes, and vague platitudes about adoption. The core thesis is stated early and never meaningfully deepened.

“99% correct is still failure when it comes to mission critical systems”

“it's not really just a coding problem. It's more of a, like, behavioral. Behavioral problem at scale”

The 'last mile verification' framing for mission-critical code is a genuinely differentiated angle, and the distinction between 'guarantee' and 'provably correct' is a sharp conceptual move. Everything else - use AI or fall behind, AI makes mistakes like humans do, SaaS isn't dead - is standard 2024-era podcast filler.

“it's sort of like I want to rewire a city, but I don't want the power to go out”

“Prove is even a stronger word than guarantee. Right. Because it's provable”

Ryan Atay has real executive credentials - former CEO of Tableau, Salesforce background, now President/COO of a $125M-funded company in a novel niche - which is legitimate. However, he repeatedly disclaims engineering expertise and cannot explain Code Metal's technical differentiation in depth, limiting the substantive value a practitioner can extract.

“I'm going to get quickly out of my realm of expertise because I'm not an engineer. We should put a little caveat on this”

“I had all this great experience and I'm very grateful for my time at Salesforce and Tableau and the things before that”

A handful of concrete numbers appear - million lines of C, a few weeks vs. 50-100 developers over two years - but they are unverified assertions with no methodology attached. Every named customer example is deliberately anonymized, and the technical claims are stated without any supporting data or third-party validation.

“I can't mention the name yet, uh, soon”

“This would have taken you, I don't know, 50, uh, to 100 developers for two years? Well, there's a cost to that”

The host raises a few genuinely interesting angles - accountability/insurance markets, human-in-the-loop as a Jenga tower, token-cost pressure - but never challenges the boldest technical claims, accepts deliberately vague customer examples without pushback, and repeatedly inserts his own multi-paragraph AI usage stories that stall the conversation.

“I was a very early OpenAI chatgpt, uh, adopter and I went through this period of using both and now I've kind of shifted over. I still use both, but I went through this exercise with our quarterly planning. I had it hooked up to our CRM and all the other data”

“Wait, you lost me at fu. Was it fuzzing? Was that. I've never heard of that term”