LifeOS

Principle № 08 · Method

Prompt the Ideal State

A prompt describes the outcome, the constraints and the tools, and leaves the method to the model.

A tilted tray with one precise socket and a steel ball rolling freely toward it, its route lit teal, the old maze walls shown only as bone-grey ghosts standing in their empty slots.
Fig. P08A precise destination; the scripted walls removed

Every prompt in LifeOS, whether it’s a skill, a workflow or a brief handed to an agent, describes the ideal state. It says what done looks like, what limits apply and which tools are available, and it leaves how to think to the model.

Why not script the steps

Step-by-step instructions on how to reason (“first analyze the inputs, then list the edge cases, then decide”) fail in two ways. They hold a capable model to a path it could have beaten, since the script can only be as smart as whoever wrote it. They also go stale: each new model makes more of the script unnecessary, so a prompt full of method written for last year’s model can make this year’s worse.

This is the Bitter Lesson applied to prompting. General methods that use the model’s own ability keep winning over hand-built methods, and the gap grows as models improve. Any line a smarter model would make unnecessary gets cut.

Precision moves to the outcome

Prompting this way is the more precise option, because the detail moves from the method to the result. “Produce a threat model that names every trust boundary and a concrete attack against each” is tighter than five steps on how to think about threats, and it still holds after the next model upgrade. The ISA is this principle applied to tasks: its criteria are the prompt.

What stays

Four kinds of how-to survive the cut, because no amount of intelligence makes them unnecessary:

  • Safety gates: confirmations, approvals and guards on destructive actions.

  • Verified gotchas: a documented, non-obvious failure the model would otherwise hit, with a dated incident behind it.

  • Tool contracts: exact command syntax, parameters and paths.

  • Output formats: the shape the deliverable has to take.

Everything else that reads as procedure is a candidate for deletion, and LifeOS has an audit built for finding it.

The Skill System → A growing library of self-activating, composable units of expertise. ISA System → One document that captures what 'done' looks like — the testable spec the Algorithm climbs toward.