AI Shaping: Turning General-Purpose AI into Productive AI Work

This paper defines AI shaping as the operating-model discipline for turning broad general-purpose AI capability into productive AI work by establishing shaped intelligence that carries a reusable and iteratively improvable domain work pattern and reusable work basis across recurring work. The opportunity exists where work retains stable-enough underlying structure despite changing subjects, inputs and work state. The operational effect and principal capability-evidence signal is two-sided work-burden shift: moving more domain-practice and subject-context burden, and more work-state and resumption burden, from the responsible person to shaped intelligence while preserving source basis and the human-authority boundary. Shaped intelligence is the reusable domain-facing capability established through AI shaping. It carries the domain work pattern and work basis into recurring work. Under the direct-shaping approach, the discipline establishes bounded shaped intelligence without first materialising separately reusable AI-shaping intelligence. Under the mediated-shaping approach, separately established AI-shaping intelligence carries the reusable shaping pattern used to develop, audit, improve, adapt, extend, transfer-evaluate or materially renew shaped intelligence under human direction. The paper explains the value hierarchy, work burden and work-burden shift, the direct- and mediated-shaping approaches, the concept architecture, category fit, a public-safe capability lifecycle, domain extension versus domain transfer, adjacent-framework positioning, the work-system and operating-model layer, and protected-method limits. It supports public category understanding only. It does not provide capability evidence, detailed lifecycle gates, implementation guidance, product specification, method transfer or deployment assurance. Where evidence review is warranted, the companion Capability Evidence paper should be used for the structured public evidence package under the public-stage and protected-stage limits described here.