My Hammertime Final Exam
A viral rationality exam teaches AI agents to plan with rollback strategies and small, reversible changes.
A post titled 'My Hammertime Final Exam' has gained traction on the rationality community forum LessWrong. Written by user evjeny, it presents a personal synthesis of productivity and instrumental rationality principles framed as a passed 'exam.' The core contribution is a trio of concepts designed to improve decision-making and system design: a 'Reversible' planning technique that mandates designing explicit rollback steps for any action, the 'One change at a time' principle to prevent small tasks from ballooning, and the identification of the 'eternity' cognitive defect, which blinds us to gradual change.
The framework is particularly resonant for AI and software engineering. The 'Reversible' technique, illustrated with backend deployment and database migration plans, is a formalization of robust system design. It argues that planning should always include a 'rollback' phase, making actions safer and failures more contained. This directly translates to building more reliable AI agents that can backtrack from unsuccessful actions and to creating safer continuous deployment pipelines. The post's blend of personal anecdote and technical rigor provides a concrete mental model for professionals battling complexity.
- Introduces the 'Reversible' technique: every rollout plan must have an explicit, executable rollback plan, modeled on software deployment.
- Defines the 'One change at a time' principle to fight scope creep, using the analogy of a simple task like replacing a trash bag snowballing into a 2-hour cleaning marathon.
- Identifies the 'eternity' cognitive defect: the failure to perceive that 'giant problems' are often solved through imperceptibly small, cumulative steps over time.
Why It Matters
Provides a structured thinking framework for designing safer, more debuggable AI agents and robust software systems, reducing operational risk.