Align purpose while preserving leverage
Set purpose, sources, authority and constraints without scripting away useful breadth, synthesis or adaptation.
AI Shaping practice
AI Shaping principles, lifecycle and operational patterns
AI Shaping practice is governed by seven established practice-derived principles, organised through a complete capability lifecycle and applied through operational patterns. Public disclosure narrows as operational specificity increases: principles and their governing relationships are stated directly, the lifecycle is shown through broad capability bands, and operational patterns are explained by role and benefit while their catalogue and operating machinery remain protected.
Principles govern what must remain true. The lifecycle governs when work, decisions and transitions occur. Operational patterns form the mixed lower layer of reusable structures, techniques, recurring practices and governance mechanisms through which the practice is applied.
Practice architecture
One hierarchy, a narrowing public view
The explanatory hierarchy is Principles → Lifecycle → Operational patterns. Principles constrain the whole practice. The lifecycle organises capability work over time. Operational patterns are the mixed lower layer through which the principles and lifecycle are applied.
The public disclosure funnel is different from the hierarchy. It controls depth, not importance: operational detail becomes progressively more protected as it becomes easier to reproduce, transfer or operate independently.
- PrinciplesSeven compact statements and five governing relationships are presented directly because they establish the strategic discipline without transferring the operating method.
- LifecycleThe complete capability-over-time coverage is shown through five broad public bands rather than the canonical phase structure.
- Operational patternsTheir mixed role and benefits are explained, while the classified catalogue, selection logic and operating machinery remain protected.
The narrowing widths represent disclosure depth, not strategic rank. Principles remain the highest layer; operational patterns remain essential to disciplined practice.
AI Shaping principles
The seven principles are established within the current AI Shaping body of work. They are practice-derived, evidence-sensitive and iteratively improvable under human direction. They are not an external standard, universal doctrine, complete body of knowledge, mandatory seven-step process or independent implementation authority.
Work with AI’s nature
- Align before shapingBegin with the model’s strongest available source-grounded account of established practice; introduce deviations only where purpose, evidence, context or human judgement justifies them.
- Design for AI variabilityTreat generative-AI behaviour as probabilistic, changeable and platform-dependent; use AI shaping only where expected benefit justifies continuing review, testing, renewal and platform-specific validation.
- Preserve AI leverageDirect and constrain work without unnecessarily suppressing the breadth, depth, synthesis and adaptation that make generative AI valuable.
Structure human–AI work
- Bound AI workBound complex AI work through purpose-, role-, source- and stage-specific units; preserve continuity through explicit state, review and handoff rather than expecting one execution to carry the whole problem.
- Maintain human directionMaintain active human direction over purpose, priorities, boundaries, trade-offs, stopping conditions and continuation while AI carries work within that direction.
- Shift suitable work to AIOnce purpose, source basis, boundaries, authority and review conditions are established, shift suitable work to AI while retaining human judgement, review, approval, communication, decisions and real-world action.
Assure and renew the work
- Validate before continuingTreat AI-carried work as provisional; validate it against purpose, sources, constraints, acceptance conditions and prior state, then remediate, revalidate, redirect or stop before consequential continuation.
The groups are memory aids. The principles recur, interact and constrain one another throughout the lifecycle.
Cross-principle relationships
These relationships show how the established principles reinforce, constrain and complete one another in recurring human–AI work. They are not additional principles.
Bound work while shifting burden
Shift coherent, reviewable units to AI while human direction, judgement, accountability and consequence boundaries remain controlling.
Design for variability and validate continuation
Treat AI work as variable and provisional; validate before consequential continuation, then remediate, redirect or stop where required.
Set direction without pre-performing the work
Define outcomes, sources, boundaries and acceptance without manually performing the synthesis through elaborate instructions.
Enable initiative while preserving repairability
Allow useful initiative while keeping differing assumptions, interpretations and emphasis visible, so divergence can be repaired against the controlling basis.
Lifecycle
The lifecycle covers the whole capability over time, not only establishment. For public orientation, the complete lifecycle is grouped into five broad bands. These bands show completeness and decision flow without publishing the canonical phase catalogue or practitioner operating sequence.
Qualify and bound
Determine whether shaping is justified, define the capability and establish its purpose, authority, sources, boundaries, suitable approach and intended value.
Establish and accept
Create the bounded capability, test its fitness, remediate material findings and obtain human acceptance before consequential use.
Operate, reuse and improve
Apply the accepted capability across recurring work while preserving continuity, reviewability and controlled improvement under human direction.
Adapt, renew, extend or transfer
Respond to material change and separately evaluate broader related use or movement into another work domain.
Retire, supersede or archive
End reliance responsibly while preserving required state, history and residual obligations.
Public-band boundary: these five bands are disclosure categories, not replacement phase names. The complete lifecycle, phase gates and practitioner detail remain private or protected. Improvability by design is also distinct from actual improvement through accepted revision and measured improvement supported by defined measures.
Operational patterns
Operational patterns are the mixed lower layer through which principles and lifecycle intent become more consistent, reviewable and resumable practice. They include reusable structures, genuine techniques, recurring practices and governance mechanisms. The public site explains their role and benefits rather than publishing the classified catalogue, selection logic or procedures.
Consistency and reuseReduce repeated reconstruction by carrying a stable work pattern and reusable work basis into recurring work.
Continuity and reviewabilityKeep sources, authority, current state, decisions, unresolved matters and review status visible across work cycles.
Assurance and controlled changeSupport proportionate validation, correction, acceptance and deliberate change without uncontrolled propagation.
An operational pattern may support one lifecycle phase, span several phases or recur across the lifecycle. The complete catalogue, classifications, applicability rules, inputs, activities, outputs, gates, exceptions and recovery machinery remain private or protected.
What this page makes visible
The public site shows the seven compact principles, five governing relationships, complete lifecycle coverage through broad bands, the role and benefits of operational patterns, observable work-burden shift, source basis and the human-authority boundary.
This is enough to understand the practice architecture, recognise potential relevance and evaluate whether a defined deeper question is justified.
What remains protected
The canonical lifecycle phase structure, classified operational-pattern catalogue, selection logic, complete procedures, detailed gates, decision rules, readiness tests, prompts, templates, schemas, validation and recovery machinery, sensitive records and sufficient material to reproduce or independently operate the method remain private or protected.
Protected-stage discussion is for a defined question, minimum necessary detail and agreed protective conditions—not unrestricted method transfer.
Connect practice to the category and work system
Use the AI shaping category page for the whole public concept architecture, Operating-model design for the work-system lens, or Productive AI work for the outcome.