You will be able to sort your tasks into delegate, assist and keep, based on stakes and skill.
Ravi, the graduate from lesson 7.2, asks his manager a reasonable question: which of my tasks am I allowed to use AI for? His manager starts answering, "Well, it depends", and realises she's never actually thought it through. Some things clearly seem fine. Some clearly don't. Most sit somewhere in between, and nobody has drawn the line.
This lesson gives you a way to draw it. Every task falls into one of three groups, based on two questions: how much does it matter if the output is wrong, and does doing it yourself keep a skill you need? AI fundamentals: what it is, how it works, where it fails has a related lesson, 8.3, Hand over, work together, or keep, which looks at it from the angle of what AI is capable of. Here the angle is risk and skill.
Delegate tasks are low-stakes, and errors in them are easy to spot. If the AI gets something wrong, you'll see it at a glance, and the consequences are small.
Examples include reformatting a table, turning meeting notes into a tidy list, converting a document from one layout to another, drafting a first-pass summary of a long article you'll read anyway, suggesting subject lines, or tidying the grammar of an internal message. Ravi's delegate list includes renaming and sorting files, and turning his rough notes into a clean agenda.
For delegate tasks, a quick glance is enough checking. You don't need to read every word, because a mistake would either be obvious or wouldn't matter much.
Assist tasks are ones where AI can do much of the work, but the output matters enough that a mistake could cause real trouble. So AI drafts and you check, carefully, line by line, before anything is used.
Examples include client emails, reports for management, proposals, code that will run in a live system, and content that will be published. Siew Ling's variance commentary from lesson 7.1, Why we trust confident machines too much, belongs here. The assistant drafts it, and she checks every direction word and figure against the table.
The key to assist tasks is that the check is real and planned in advance. "I'll look over it" isn't a check. "I'll compare every number to the source table" is.
Assist tasks also need a named human reviewer for anything that leaves the team: a specific person who reads the output against the source before it goes to a client, a regulator or the public. Often that's you. For a junior colleague like Ravi, it might be his manager for the first few months. Naming the person makes it clear who's responsible, and stops each person assuming someone else checked.
Keep tasks are ones where the stakes are high, or where the task builds or uses a core skill you need to hold on to. AI may still help at the edges, for example by finding a reference or checking grammar, but the substance is yours.
High-stakes examples include final figures in financial statements, legal or contractual wording, medical or safety information, and any decision about a person, such as hiring, promotion, discipline or a performance rating. Module 3's bias cases, especially Amazon's recruiting tool in lesson 3.2, show why decisions about people belong here.
Core-skill examples depend on your job. For Siew Ling, it's reading a set of accounts and forming a view. For a lawyer, it might be drafting the key clause. For a designer, the first concept sketch. Lesson 7.2, Skills you lose when you stop practising them, explained why keeping these by hand protects your ability to judge AI output elsewhere.
You'll keep meeting new tasks, so it helps to have a one-sentence rule you can apply without thinking each one through from scratch.
Ravi's manager wrote this for the team: "If a mistake would be obvious and harmless, delegate it. If a mistake could reach a client or cost money, AI can draft but a named person checks every line. If it decides something about a person or a final number, or it's a skill you're still learning, do it yourself."
Your rule will be different, because your work is. Someone in marketing might put published content in assist and brand strategy in keep. Someone in customer service might delegate templated replies and keep anything involving a complaint or a refund. The point is to decide once, in calm conditions, so you're not deciding under deadline pressure.
Notice that tasks can move between groups. As Ravi gains experience, some of his keep tasks may move to assist, once he can reliably spot the errors. And if a tool starts making more mistakes on a delegate task, it moves up to assist.
Before writing your own rule, picture a task you were given recently that you weren't sure how to handle, and ask which group it would land in under each of the two questions.
Write the rule you will use to decide which group a new task belongs in, in one sentence.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).