Deciding & Quitting — Action Under Uncertainty

Develop the discipline of acting under irreducible uncertainty without collapsing into paralysis or recklessness. Classify decisions as reversible or irreversible and budget deliberation accordingly. Learn to quit — with pre-committed kill criteria — paths that no longer earn their keep. Apply all of it to the realest case you own: the portfolio of unfinished side projects, each of which gets an explicit, dated decision: continue, sunset, or pause-with-revival-criteria. Type 1 / Type 2 is irreversible migrations vs. feature flags. You already run structurally different rigor for them: the destructive migration gets review, staging, a low-traffic window, a rollback plan; the flag gets toggled and reverted in seconds. Most engineers know not to give a flag question migration-level deliberation — then deliberate over Type 2 life decisions as if they were Type 1. Bezos's framing is your own deployment instinct, made portable. Resulting is blaming the engineer for the bug instead of the system that allowed it. Your 3am-postmortem culture already separates process from outcome; Duke's filter is the same discipline pointed at your own decision history. And quitting is deleting dead code. Senior engineers take pride in PRs that delete more lines than they add; a project kept in ambiguous "still around" status is dead code in your professional codebase — maintenance cost, no value, and it makes everything else harder to reason about. Quit is the case for git rm applied to commitments. The portfolio decision is the cleanup PR, and like all cleanup PRs, the hard part is not the deletion — it's admitting the code was dead.