Andy Walsh2x Exited Founder, Startup Strategist & Podcast Host

Your AI twin works, until it doesn't

19 Jun 2026 · 1 min

Watch on YouTube

0:38

An AI digital twin can mimic someone at about 80% accuracy, but only briefly. Without constant new data, it freezes in time, becoming outdated as the real person learns and changes. One person could theoretically quit and let their twin handle work emails, but the trick only lasts until the twin falls behind.

Chapters

can you fire me? And I think there is a set of truth to that where it's like, hey, if I can create someone that is 80% of you, it is going to work, but for only for a limited time. Because as new inputs come in, it's not going to be able to say like, hey, Andy would've changed it this way. He actually stopped using this approach. He started to use this terminology instead. Those things it doesn't have access to. So it's gonna be Andy from 2 years ago. At a certain point, Andy from 2 years ago is not as smart as Andy today. Yeah. And so you want that training data. So, so there is an ephemeral period where it is valuable, where people are going to see that benefit, where if you decide to quit, you're like, hey, I don't have to give 2 weeks notice, just use my digital twin. Yeah.