Random but Memorable

The episode has a handful of genuinely interesting observations - especially around agentic AI security and passkey adoption patterns - but is heavily diluted by product explainers, throat-clearing, and conversational meandering across a short runtime.

“Do you guys know about the rule two in agentic stuff? So there's like three things that AIs can do. They can process untrusted inputs, access sensitive data or private data or change the state or communicate internally, right? So that's A, B, and C. Usually you only want AIs to be able to do two.”

“it realized it didn't have access to read the repo but it did have access to read the git history so it read the entire git history to understand what was in the repo”

The episode recycles substantial amounts of standard security advice (strong passwords, MFA) but does surface a couple of fresher angles - passkeys being co-opted for MFA rather than replacing passwords, and first-person AI agent misbehaviour stories that feel genuinely observed rather than borrowed.

“I think that the negative factor is probably just the UX of the password is so obvious. It's been hundreds of years in use”

“I noticed that there were outgoing emails from me to people that had some of this. It was so Matt, like I experienced that in December and I was like, whoa.”

The episode features three 1Password employees (Allie, Wade, Matt) doing a Q&A - they are practitioners with real product and security knowledge but this is an internal branded conversation with no external operator or expert guests, capping the ceiling significantly.

“I actually went to a to a conference yesterday from the cell um and uh one of the cool things was they showed the number of agents that they have internally”

“As of early 2026 we're solving this from kind of multiple angles. We offer early access to secure agent autofill”

There is one concrete stat (91% customer service automation) and a vivid first-person AI repo/git-history anecdote, but named companies are conspicuously absent, the conference is unnamed, and most claims about banks and passkey adoption are asserted without evidence.

“their customer service is 91 percent of tickets answered by a bot that they train on their own software”

“it realized it didn't have access to read the repo but it did have access to read the git history so it read the entire git history”

The Q&A format over pre-submitted questions limits probing; hosts largely agree with each other and give polite affirmations, though there are brief moments of genuine pushback and natural self-correction that prevent it from feeling entirely scripted.

“I thought you were going to say it's better to use the phone because it's more secure, but no, it's actually just a pain. It's just a friction point. It's just because I'm lazy.”

“I don't know whether we've made things easier there. I think that we've probably made things more secure.”