Why non-deterministic AI needs deterministic infrastructure

Hinzugefügt: 2026-05-28

[00:00:00]Do you feel like there's anything that the AI ecosystem is getting wrong or missing when it comes to distributed systems?

[00:00:05]>> Yeah, I think well, so the models are really good. So, like Anthropic's got like 4/7s Opus and they got 5/5 for OpenAI.

[00:00:13]They're really good at things like, you know, having random Janet Jackson songs from 1992, but also writing code. Yeah, yeah, escapades.

[00:00:21]But what I'm getting at is like it knows too much about too many things. So, when you think about the thing it gets wrong is if you write Go or Rust, it tries to reinvent everything from bottom up. It never looks for libraries. Is there something you think is the right direction people have been going in lately with what they're building and how they're getting AI into production?

[00:00:38]Yeah, I I think actually folks who are doing the proper context engineering was like super hot last summer. Now we have skills. We've got MCP. So, like it's really about your tooling cuz like you have non-deterministic AI, you want to get more determinism within your infrastructure.