02 / AI at Work · Field briefing
Where AI earns its place
Choose the task before the tool.
A business rarely needs an AI strategy for every task at once. It needs a useful first experiment. Start with work that has clear inputs, a recognisable output and someone who can judge whether that output is right. A capable model cannot compensate for an undefined decision.
Which piece of work should we try first?
An explanatory schematic—not a live system or measured performance.
KAF World Digital · Prepared 4 October 2026 · First editorial edition. Worked examples are invented for teaching.
Try the example02 / Follow the work
Follow the work
Separate three things: the task, the information available, and the consequence of an error. Drafting a reply from an approved returns policy is different from sending the reply. Summarising a meeting is different from deciding what was agreed. Extracting invoice fields is different from authorising a payment. The same model can be useful at one stage and unsuitable at another.
Consider an invented small repair business. The owner spends time finding the right service instructions, drafting responses to common enquiries and reconciling disputed bills. A search-assisted draft from current instructions could be a useful pilot. A disputed bill needs a responsible person to assess the facts. The distinction comes from the workflow, not a universal list of tasks AI can or cannot do.
Use a four-question screen. Is the task frequent enough to observe? Can you obtain authorised, representative inputs? Can a person identify a good result without doing all the work again? Can a failed attempt be stopped or corrected cheaply? Unknown answers are research tasks before they are automation opportunities.
Do not confuse visible speed with end-to-end benefit. A thirty-second draft that takes fifteen minutes to repair may not help. Record the existing process, then count preparation, review and rework in the trial. Pay attention to the people receiving the output: if they need to verify every detail from scratch, the work may simply have moved.
There are often three reasonable outcomes: use ordinary software or a template, use AI to assist a person, or explore a more automated workflow with explicit permissions. Keeping a task manual can also be a sound decision. A pilot is useful when it helps choose between these outcomes.
Evidence for the technical mechanisms: Microsoft · Using Generative AI Responsibly, Anthropic · Building Evals. The worked examples and practical framework are KAF editorial recommendations.
03 / Worked example
Try the task map
Four questions for finding a useful first experiment. This interactive lesson uses prepared synthetic examples and local calculations.
Use the companion worksheet and the worked example. All assumptions are illustrative.
No account, model API or external service is connected. Inputs are not submitted or stored by this page.
04 / Go deeper
Useful words. Important details.
- Reviewer
- The person who can judge whether an output meets the task’s requirements.
- Baseline
- The current process measured before a proposed change.
- Recovery
- How a failed or mistaken step is stopped, corrected or reversed.
- Accepted task
- An output that meets the stated criteria after required review.
Separate the stages
Map input collection, transformation, review and execution separately. A task can be suitable for drafting while its execution remains unsuitable for automation. Evaluate the workflow you will actually run.
Avoid a universal opportunity score
Frequency, input rights, reviewability and failure consequence are not interchangeable quantities. This guide does not combine them into a pseudo-scientific score. A failed prerequisite can outweigh an apparently attractive speed gain.
What to record
Save source examples, the approved task definition, observed preparation/review time and failure categories. Use the same acceptance rules when comparing the existing and proposed processes.
The technical notes qualify the explanation; they are not product-specific assurance or a certification.
05 / Evidence & scope
Inspect the source.
Keep the context.
Sources were accessed on 4 October 2026. These technical and security references support the mechanisms described; they do not validate our fictional examples or establish comparative product performance.
- Microsoft · Using Generative AI Responsibly
Responsible-use principles, representative testing and operational ownership. Technical guidance, not legal advice.
- Anthropic · Building Evals
Task-specific evaluations and code, human and model-based grading. Provider examples are not independent comparative evidence.
What this briefing does—and does not—establish
The worked examples, suggested questions and worksheets are original KAF teaching material. Numbers are explicitly illustrative. No customer data, observed adoption results, market-size forecast or live provider pricing is presented.
Check the exact product, release, configuration and policy before using a workflow. Source documents can change; this is a dated educational briefing.
Keep a useful copy.
Read the full briefing offline, or use the companion worksheet to apply it.