Output
Is the result fit for the task and its context?
Technology / Artificial intelligence
Artificial intelligence · one of five concerns
AI is a change to work before it is a change to software.
Artificial intelligence can infer, draft, classify or recommend. The difficult question is what people do with that output, under which conditions, and who remains responsible when it is wrong.
The same decision, four questionsPerspective → Practice
A tool drafting a customer reply
The model can produce words quickly. The organisation still has to define review, escalation, records and the cost of error.
01
Perspective
Perspective
02
Performance
Performance
03
Precision
Precision
04
Practice
Practice
01 · Perspective
Perspective starts with the task and its consequence.
Ask what the person is trying to accomplish, what information they need, and what could happen when the system is wrong. Automation is one option among process change, training, better information or no change.
What must remain a human responsibility because the consequence is too important to outsource?
02 · PerformanceSignal Work Consequence Question
Performance is more than model output.
A useful AI-enabled workflow considers quality, review effort, repeat work, incidents, user understanding and the experience of people affected by the decision.
Review
Can the responsible person check it with enough time, information and authority?
Impact
What happens to customers, staff, data and decisions when it fails?
Ask
Where is the error detected and who can stop the route?
03 · PrecisionDefine before selecting Accept before scaling
Precision defines the safe use case.
Name the task, inputs, prohibited uses, review conditions, record, incident route and change owner. Confidence language alone does not create oversight.
Define the use
What may the system do, for whom, and with which information?
Define the stop
Which signal, error or new condition requires a person to pause or change the system?
04 · Practice
AI becomes practice when responsibility stays visible after deployment.
Competence, monitoring, incident response and change control keep the tool inside the organisation’s policy and the task owner’s real authority.
Continue to Practice →Sources and notes
Read the basis, then test it in context.
The sources below are primary management and regulatory references. Applicability depends on system, role, sector and jurisdiction; they do not make one universal AI workflow.
- ISO/IEC 42001:2023 — AI is managed through policies, objectives and processes.
- NIST AI RMF · Govern — roles, responsibilities and communication should be clear.
- EU AI Act — AI literacy and deployer duties depend on system and role.