YOU

Thinking

Thinking is the deliberative reasoning a harness or model may produce before or alongside tool use and final answers, including thinking controls that trade quality against latency and cost.

What It Is

Thinking is the deliberative reasoning a harness or model may produce before or alongside tool use and final answers. A harness is the agent runtime the factory drives, such as a coding agent under a shared factory process. Thinking is not the factory workflow system itself, and it is not the factory thoughts work type used for planner or loopback work.

Why It Matters

Thinking and thinking controls trade answer quality against latency and cost. Raising a thinking or effort control can improve hard tasks by giving the model more deliberative reasoning; lowering it can make runs cheaper and faster. Not every harness exposes the same thinking controls, so operators need to know which knobs their chosen harness actually offers.

Simple Example

You start a factory run on a hard coding bug. Before the harness edits files, you raise its thinking or effort control so the model spends more deliberative reasoning on the diagnosis. On a routine rename in the same harness, you lower that control so the run finishes faster and costs less.

Common Confusions

Thinking is not factory thoughts work. Thoughts work is a planner or loopback work type the factory schedules; thinking is deliberative reasoning inside a harness or model turn. Thinking is also not tokens—neither LLM cost or context units nor factory work tokens that move submitted work through places. Tools are named callable actions; compaction shrinks context. Thinking is the deliberative reasoning that may happen before or alongside those actions, not the actions or context shrink itself.

Tags

On this page

What It IsWhy It MattersSimple ExampleCommon ConfusionsTags