Anti-Vision as Debugging for Personal Operating Systems
anti-vision plus daily levers maps to goal loops as an inspectable operator state machine instead of vague discipline theater
Dan Koe uses an X Article to make a simple claim: New Year’s resolutions fail because they try to change actions without changing the person those actions naturally come from.
The clever bit is the protocol shape. This is not “be more disciplined” with a better font. It’s closer to a one-day debug session for your life: surface unconscious goals, write the anti-vision, interrupt autopilot, synthesize what you actually want, then turn it into a game with a yearly mission, monthly project, daily levers, and constraints.
That framing is useful for joelclaw because vague goals are poison for humans and agents alike. If a goal can’t be decomposed into state, constraints, feedback, and daily levers, it’s just vibes in a trench coat.
The best part is the anti-vision. Before deciding what to build, describe what happens if nothing changes. That gives the system a negative attractor, not just a shiny desired state. For agent loops, that’s a solid pattern: define the failure mode vividly enough that the loop can steer away from it.
Key Ideas
- Dan Koe argues that changing actions is second-order; changing identity is first-order because behavior follows the person you believe you are.
- The article leans on Alfred Adler’s “trust only movement” idea: stated goals matter less than observable behavior.
- The “anti-vision” works as a negative spec: document the life you get if current defaults continue, then use that tension to choose a better target.
- The “personal vision game” turns aspiration into structure: yearly mission, monthly project, daily levers, and constraints.
- The protocol connects cleanly to state machines: current state, desired state, forbidden state, transitions, guards, and feedback.