~/cybernetics
Cybernetics_
“The agent isn’t the problem, you are.”
We have to revisit the concept of cybernetics, in the context of recent advancements. We now have the ability to augment our own intelligence. Can our wisdom keep up?
01Two levers for increasing capability.
You can improve yourself, and you can extend yourself. Improving is slow, compounding, and invisible on a quarterly basis. Extending is fast, easy, and looks like progress from the outside.
Both are real levers. The interesting question is the balance between them: how much of your capacity should come from your own judgment, and how much from systems that carry your judgment forward when you aren't in the room.
02An agent is an extension of the person who built it.
Every agent you build is made out of you: your model of the problem, your vocabulary, your taste, the failures you happened to have seen. You write your judgment into prompts, contracts, gates, and permissions, culminating in an engineered system. You are performing systems engineering.
That also makes you its ceiling. An agent can't be more rigorous than the person who defined "done", can't catch a class of failure its builder never imagined, and will happily automate the bias nobody thought to check. When agents aren't producing good work, the diagnosis is usually upstream: the builder hasn't decided what good looks like, or hasn't got the reps to recognize it.
And because the extension copies you, it repeats your mistakes at scale, in public, until you notice them. The agent doesn't only do your work; it tells you what your work has been all along.
03Your second brain is a career asset.
A second brain is the external store of what you have read, decided, shipped, and gotten wrong: the notes, decisions, and postmortems that would otherwise dissolve with time. Held outside your head, it survives context switches, busy quarters, and the version of you that has already forgotten why.
When the store is real, prompts and gates get written from evidence instead of from memory, and the mirror above stops depending on how well you slept that week. Capture what you learn; distill it until it is load-bearing; let it feed the next thing you build.
04The balance, in practice.
Extend only, and you automate your current level of understanding and freeze it. Capability plateaus at whatever you were when you wrote the first prompt, then quietly decays as the world moves on.
Improve only, and you're correct and slow while someone who extended themselves does the same work in a fraction of the time.
What I'm aiming for: enough of the work stays in my hands that I'm still learning, and enough of it moves into agents that what I learn shows up in the system the same week. Improvement feeds the extension; the extension makes improvement matter.
05Where the line sits.
The diagram places every task on two axes: how often it comes up, and what a mistake costs when it does.
A moving highlight walks the four outcomes, from keeping the work yourself to handing it off entirely.