6QQMN331 Chap.2 Agents, Skills and Augmented Work
Agents, Skills and Augmented Work
Choose the lightest capable system
This chapter distinguishes one-time prompts, reusable skills and agents that plan across steps, tools and changing conditions. It starts from the work pattern rather than the product label, then decomposes inputs, decisions, actions and permissions.
The central boundary is simple: more autonomy creates more coordination and control work.
Design meaningful oversight
The worked workflow shows how a repeated issue digest can justify a supervised agent while leaving category changes and consequential alerts with a manager.
Practice focuses on permission limits, failure recovery and the division between AI speed and human judgement.
Read autonomy as an organisational choice
Tool access, memory and the ability to choose a next step change more than technical capability. They redistribute who notices an error, who can interrupt execution and who explains the result.
The chapter therefore treats every proposed action as a permission decision. Readers learn to distinguish drafting from sending, detecting from changing and recommending from approving.
Those distinctions make oversight concrete because they show exactly where a person can inspect evidence or halt a workflow.
Compare normal work with exceptions
A design that succeeds on the ordinary route may still fail when information is missing, categories conflict or an external tool is unavailable. The method maps a normal path and an exception path before adoption.
Recovery requires a retained prior state, useful logs and a named decision owner. This comparison also reveals when a reusable skill is enough: if a person can supply the changing context and choose the next action, added autonomy may create control costs without solving a real problem.
What this chapter covers
- 01
Prompts
- 02
Reusable Skills
- 03
AI Agents
- 04
Agentic AI
- 05
Harnesses
- 06
Augmented Workforce
Worked example · free
Route a weekly issue digest
- 2Decompose the repeated workflow and branching decisions.
- 2Select the lightest capable system.
- 3Set permissions, approval and recovery routes.
Key terms
- Prompt
- A Prompt is a one-time instruction requesting a defined output.
- Reusable Skill
- A Reusable Skill packages instructions and context for repeated work.
- AI Agent
- An AI Agent plans and uses capabilities or tools toward an objective.
- Harness
- A Harness coordinates models, tools, safety checks and execution.
Agents, Skills and Augmented Work FAQ
When is a prompt enough?
A prompt is enough when the user can supply the information, the desired transformation is clear and the task has few steps or branching choices. Adding an agent to such work may increase permissions, failure points and monitoring effort without adding useful capability.
What makes a system agentic?
An agentic system works toward a goal by selecting and coordinating actions rather than only returning one response. The analysis should still identify permitted tools, stopping conditions, adaptation limits and the human role that owns purpose, exceptions and consequential decisions.
What does meaningful oversight require?
The person needs relevant context, time, authority and a clear action when the system is wrong or uncertain. A formal approval step is weak if staff routinely accept outputs, cannot see the basis or are penalised for justified disagreement.
How should permissions be specified?
Write each permission beside the task that needs it and separate reading, drafting, editing and sending. The system should receive the narrowest access that completes the approved workflow. Also identify who grants access, how it expires and what happens after unexpected use. This turns a broad safety promise into an auditable boundary.
Why does exception handling matter?
Agentic work can branch when inputs are incomplete, tools fail or evidence conflicts. A design should state when execution pauses, which information reaches the reviewer and how the prior state is recovered. Without an exception route, routine efficiency can hide the moment when a system exceeds its competence or authority.
How can augmentation be evaluated?
Compare the old and proposed division of labour. Identify whether people gain time for judgement or merely inherit monitoring, correction and escalation work. Useful evaluation measures decision quality, workload distribution, missed exceptions and recovery as well as speed. The conclusion should say which human capability is strengthened and which new responsibility appears.
Assessment move
Take three familiar tasks and route each through the chapter decision tree. Explain why a prompt, skill or agent is the least complex adequate choice. For any agent case, list inputs, choices, actions, permissions, human approvals and recovery. Practise narrowing access rather than adding a generic oversight sentence. Redraw one workflow after removing a tool permission and decide whether the task still works.
Then introduce missing information or a failed action, mark the exact stopping point and write the evidence a reviewer would need to recover safely.
Working through Agents, Skills and Augmented Work in 6QQMN331? Sia is AskSia’s AI Management tutor — ask any 6QQMN331 Agents, Skills and Augmented Work question and get a clear, step-by-step explanation grounded in how 6QQMN331 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.