INTRODUCING THE AR PRACTICE   ·   A practical field guide to artificial resources, work and people
THE AR FIELD GUIDE / Resources

Manage the whole AR lifecycle.

Artificial resources need ownership, defined authority, periodic evaluation and planned retirement — not just procurement.

KNOW WHAT YOU MANAGE

Different resources.
Different controls.

The term AR includes the capabilities that perform work and the infrastructure that makes that work possible.

M
Models

Reason, classify, predict or generate

Record version, provider, task fitness, data dependencies and observed limitations. Evaluate changes before they affect critical workflows.

A
Agents

Plan, act and interact with tools

Document permissions, operating boundaries, escalation triggers and audit trails. Autonomous action introduces distinct operational risks.

C
Compute

Provide capacity, resilience and cost control

Plan capacity, latency, unit economics, hosting location, energy use, fallback and supplier dependence.

W
Workflows

Turn capabilities into accountable work

Map decisions, hand-offs, data access and human approvals. A capable model does not guarantee a reliable process.

THE AR LIFECYCLE

Seven stages of management

01

Plan

Define capability demand, alternatives, intended value and acceptable risk.

02

Select

Assess provider, architecture, cost, suitability and contractual dependencies.

03

Onboard

Assign an owner, permissions, test criteria, logging and human escalation.

04

Operate

Monitor actual performance, quality, incidents, cost and human workload.

05

Improve

Address failure patterns and reassess after changes to data, tools or models.

06

Reassure

Periodically verify that evidence, controls and the original purpose remain sound.

07

Retire

Revoke access, manage records, replace dependencies and protect service continuity.

A MINIMUM RESOURCE REGISTER

Make ownership visible.

Every production artificial resource should have an entry that a manager can understand without becoming an AI engineer.

Use the AR Register template to record purpose, provider, version, accountable owner, authority, data sensitivity, human oversight, assurance evidence, cost and next review.

Download the AR Register ↗
EXAMPLE RECORD / ILLUSTRATIVE

Customer response assistant

Purpose: Draft responses to routine enquiries

Authority: Draft only; no autonomous customer contact

Owner: Customer operations manager

Evidence: Test set + monthly quality sampling

Review trigger: Model update, incident or scheduled reassessment

↑ The opportunity

Where value can be created

  • Treat compute and agent permissions as managed assets rather than invisible technical costs.
  • Reuse successful controls across teams and providers.
  • Retire underperforming resources before costs and vulnerabilities accumulate.
↓ The risk

What needs attention

  • Unregistered agents gain unintended access or change behaviour after an update.
  • Capacity, data or supplier failures interrupt core processes.
  • Ownership becomes unclear as resources move between departments.
Decision checkpoint: No production agent without a named accountable owner, defined permissions, monitoring and retirement route.
MAKE THE NEXT DECISION BETTER

A resource is only as reliable as its controls.

Translate the inventory into a testable assurance plan for the outcomes that matter.