When the govt shut down Fable this past weekend, some of us wondered if we'd grown too dependent on a single model. A single model is a single point of failure.
To be clear: I was loving Fable, and I can't wait to use it again. But I've been using Codex to build a DTC company's agentic operating system, and when Fable went down it hit me that I could just point Codex at the same repo and (maybe?) get similar results. I'd planned to play with Hermes this week but this seemed more relevant.
So that's this week's CoS test: can I run my chief of staff (Kubrick 2600) on two different models at once and compare apples to apples? So far? Codex is playing catch-up. Cowork's still my main squeeze, even for this post.
But! Kubrick now runs on Anthropic's Claude Cowork and OpenAI's Codex at the same time. Same brain, two repos, two sets of rails. It's interesting if nothing else, because it proves there's something durable I'm creating outside of the model walled garden.
For a quick catch-up, I've been building Kubrick, my open source personal chief of staff, in public, in Cowork. It's not actually an app. It's a stack of plain markdown files, with a frontier model and a desktop app as the UI. The files explain who I am, what I'm working on, how I want to be pushed, how to write in my voice (ish, I've still got a lot of editing to do).
The stages are a HAL reference. v1000 · Desktop. Ran on my Mac. Worked, as long as the laptop was awake. v2000 · Always-On. Moved to Claude Routines. Now it runs while I sleep. v2600 · Codex. This week. The same brain, now also running on OpenAI's Codex.
It's a bit split-brain: two firewalled versions running at once. But the Claude version and the Codex version did instantiate from the same SOUL.md, the same USER.md, the same second brain in a private GitHub repo. I didn't change a line of the brain to make it work.
One brain, two models, side by side. (Held together with more duct tape than I'd like.)
Here's why you should care - personal intelligence is coming. We're in the earliest innings: the Altair moment, the Homebrew Computer Club moment, the early hackers figuring out the interaction patterns and the value props moment. None of it is there yet. But some day, in the next year (or five), you'll have a chief of staff (or five) running your home and your professional life. It feels inevitable at this point and I get glimmers of magic every day even in my duct tape version.
My particular version is interesting because I've chosen to build in a GUI, while the best work is being done in custom harnesses on the command line, on a Mac Mini, or in the cloud. It hits different when it's just a project in Cowork or Codex. My thinking is it might be a little more accessible to everyone and I think everyone should play with these sooner than later.
And the cool part? My second brain is just a bunch of .md files in a private repo that's mine.
Own the brain. Rent the rails.