@DanielMiessler joins @labenz on @CogRev_Podcast for a deep dive into how to actually build a personal AI infrastructure that works.
They cover:
Nathan's 1GB second brain: 5 years of emails, calls, DMs, podcasts & social media — all in a local SQLite database, indexed and searchable, so Claude can retrieve things Nathan himself has forgotten
The two-computer architecture: a laptop agent with high access but low autonomy (draft, don't send) vs. an always-on Mac Mini running more autonomous agents with their own Gmail, GitHub, and Mercury virtual cards
Why a clear hierarchy among AI agents beats emergent teamwork — and how Nathan structured one-way control so the Mac Mini agents can never reach back up into the laptop
The social norms question around AI agents interacting as you: why Daniel never lets AI write as him ("if it's doing the writing, it's doing the thinking"), and what Nathan settled on with his agents Aid and Clay
Daniel's core security heuristic: minimize the number of companies touching your sensitive data — treat anything stored at a small company as eventually compromised, and design around the Titans
The "current state → ideal state" frame: why Daniel treats this as the only container that matters for your entire AI ecosystem, and how to start articulating your own ideal state
Why you should build a continuous attack surface assessment into your AI stack — and the incident response skill that rotates all keys with a single command
