Make knowledge traceable, actions bounded and collaboration reviewable.
Methodology: The ZHICE Trusted AI Value MethodThe challenge for enterprise knowledge assistants is often not fluency, but whether the cited document is current, whether it applies to the equipment in question and whether the user is entitled to see it. When an application calls systems, drafts work orders or executes actions, errors can also propagate through the workflow. ZHICE designs knowledge, tasks and authorization together. Before a document enters a knowledge base, its owner and scope must be clear. Before a tool is called, its permitted actions must be defined. When a task ends, results and key actions must be recorded. The degree of agent autonomy depends on both task uncertainty and the ability to control its behavior.
Enterprise knowledge must retain its business context when used in AI. Equipment manuals, operating procedures and historical defect records apply to different objects, versions and conditions; similar content is not necessarily interchangeable. ZHICE first clarifies knowledge ownership and task scope, then designs retrieval, citation and conflict handling so the system can explain its sources and applicability. Insufficient or contradictory evidence should be referred to the right person, not concealed behind fluent language. Knowledge organization itself must become continuing operational work.
Agent design must address both task flexibility and action constraints. Where steps are stable and outcomes easy to verify, a fixed workflow is often easier to explain and manage. Dynamic autonomy should be considered only where the task genuinely requires choices about its path. ZHICE authorizes reading, drafting, submitting and execution separately, with explicit responses to tool failures, insufficient permissions and environmental changes. Evaluation examines outputs as well as whether tool actions stayed within their permitted scope, preventing a seemingly correct result from obscuring an inappropriate process.
Swipe to explore the complete table
| Object of diagnosis | Risk to identify | Records to check | Design requirement |
|---|---|---|---|
| Knowledge provenance and applicability | Outdated versions, similar equipment or different regions are mixed together, and responses cannot explain their basis. | Document sources, versions, validity conditions, owners and citation samples from actual tasks. | Preserve provenance and applicability, with explicit handling and handover for conflicting, missing or outdated knowledge. |
| Task path and autonomy | Stable tasks acquire unnecessary dynamic decisions, making behavior hard to reproduce and failures hard to locate. | Task decomposition, fixed-versus-dynamic path comparisons, tool dependencies and failure records. | Begin with the capabilities actually needed, adjusting autonomy to task requirements and validation findings. |
| Tool authorization and human confirmation | Query permissions expand into execution, and generated recommendations directly trigger significant business actions. | Identity and permissions, tool descriptions, approval points, key action logs and actual result states. | Distinguish reading, drafting, submitting and execution, retaining explicit authorization and review for high-impact actions. |
| Exceptions and task recovery | Execution continues despite conflicting sources or tool failures, without pausing, handover or safeguards against duplicate actions. | Exception tests, takeover workflows, tool-result confirmation and retry or recovery records. | Define executable pause and takeover conditions, verify action results and retain a reviewable process record. |
01Can an important recommendation be traced to a current original source applicable to the equipment or business conditions, and how is conflicting information handled?
02Which steps truly require autonomous selection, and could a fixed workflow with human handovers already meet the need?
03May a tool read, draft, submit or execute? Do its actual permissions exceed the task, and who approves important actions?
04When a tool returns an uncertain result or a task is interrupted, how does the system establish whether an action already occurred and avoid duplicate execution?
Three connected levels form the working approach. Select a stage to explore the detail.
Classify policies, procedures, drawings, equipment manuals, fault records and expert knowledge. Check versions, validity periods, applicable equipment, classifications and access restrictions, and identify conflicting content.
Distinguish workflows with fixed steps from tasks that require dynamic choices. Define the boundaries between information retrieval, analytical judgment, drafting and business execution.
Observe when users need to check the original text, consult an expert or seek approval. Identify points where overconfident answers are most likely to be misused.
Preserve provenance, versions, conditions of applicability and permissions. Significant conclusions should link back to evidence. When knowledge is insufficient or contradictory, explain this to the user and refer the issue for human handling.
Validate retrieval, explicit workflows and tool interfaces before deciding whether dynamic planning is warranted. Specify tool inputs, outputs, exceptions and authorization conditions.
Separate permissions for querying, drafting, submitting for approval and execution. Authorized staff confirm high-impact actions. Production control capabilities require separate assessment and cannot inherit the launch conditions of an ordinary office assistant.
Pilot a clearly bounded knowledge domain. First test citation accuracy and task completion offline, then observe use within a controlled user group. Demonstration questions alone do not justify launch.
Cover outdated versions, missing information, insufficient permissions, tool failures and malicious instructions. Check whether the application correctly refuses, pauses or requests human intervention, recording key actions.
Classify feedback as knowledge, retrieval, model or process issues and assign accountable owners. Rerun the relevant evaluations after knowledge or tool changes to detect regressions in other tasks.
Agree depth, data conditions and acceptance criteria at initiation. The engagement scope follows the tasks confirmed by both parties.
Suppose maintenance staff need to retrieve equipment manuals, historical defects and applicable procedures. This is a design scenario for a controlled assistant, not instructions for operating actual equipment.
Limit equipment types and knowledge scope, confirming document versions, local applicability and access permissions.
Have the assistant provide source-linked information indexes and preparation checklists, with important judgments reviewed by qualified staff.
If the application extends to drafting work orders, retain existing approvals and prohibit knowledge responses from directly triggering field operations.
Collect citation errors, omissions and reasons for human edits, then update knowledge and evaluation cases.
01Can a recommendation be traced to the valid version of its original source?
02Which actions may be automated, and which require human confirmation?
03What happens when sources conflict, a tool fails or permissions are insufficient?
We translate management research and technological advances into working methods for energy businesses. The frameworks below are ZHICE's synthesis of the research and industry questions.
Application in the ZHICE approachApply trustworthiness and risk management across AI design, development, use and evaluation, specifying intended use, accountability and continuing monitoring.
Scope and limitationsA voluntary risk management framework, not proof of legal compliance for projects in China. Material described as a concept note on the 2026 critical infrastructure Profile page is not treated as a formally released standard.
Read the original researchApplication in the ZHICE approachIdentify risks specific to generative AI and incorporate trustworthy knowledge, information security, human review and operational feedback into application design.
Scope and limitationsA cross-industry risk guide. Controls must be selected for the task, data, people and system environment rather than applied mechanically.
Read the original researchApplication in the ZHICE approachDistinguish predefined workflows from agents in which models dynamically choose steps, keeping architecture complexity proportionate to task needs.
Scope and limitationsAn engineering practice article, not an industry standard. The official site notes that some tool information has changed; only its architecture selection principles are used here.
Read the original researchApplication in the ZHICE approachEvaluate task outcomes, action records and performance across repeated runs, combining automated checks, model-based assessment and human judgment.
Scope and limitationsGeneral agent engineering experience cannot replace on-site validation, safety assessment or professional judgment in the energy industry.
Read the original researchApplication in the ZHICE approachDraw on perspectives on differences in employee experience and learning effects to test AI outcomes by role, proficiency and task difficulty.
Scope and limitationsThis reference uses the final paper published in QJE in 2025. The study concerns customer support workers at one company; its average productivity change is not extrapolated into benefits for energy businesses.
Read the original researchEach practice examines diagnosis, solution design, implementation and delivery requirements in detail.
Turn a broad range of technical possibilities into prioritized business choices supported by evidence.
Diagnosis · Design · ImplementationExplore this practiceMake knowledge traceable, actions bounded and collaboration reviewable.
Diagnosis · Design · ImplementationExplore this practiceMake launch decisions evidence-based and embed risk accountability in everyday operations.
Diagnosis · Design · ImplementationExplore this practiceDiscuss the choices, constraints and evidence needed, then agree the engagement scope and work programme together.