You will finish a substantial piece of work by chaining smaller requests and checking each step.
Open a new document now and title it with the name of a piece of work you need to produce, followed by "step log". You'll write one entry in it after each step of the exercise below. When you finish, the log shows how long the work took and where you had to step in.
Lessons 3.1 to 3.4 each taught one move. Lesson 3.1, "Why one giant prompt gives you a mediocre result", showed why you split a big job up instead of writing one giant prompt. Lesson 3.2 was about asking for the plan before the work. Lesson 3.3 was about letting the assistant ask you questions first, and lesson 3.4 was about knowing when a reasoning mode is worth the time it takes. The chain puts those four moves into a fixed five-step sequence: questions first, outline, section drafts, critique and final edit. You run it on a real piece of work, from a blank page to a finished document.
Set aside about forty minutes. You can spread the work over more than one sitting, as long as you do the steps in order.
Choose something you actually need to produce, with enough substance for several sections. A proposal, a project plan, a training guide, a policy explainer for your team or a detailed handover note would all work. Skip anything shorter than a page, because the chain will feel like overkill for it. Also skip anything containing confidential client data you aren't allowed to put into your assistant.
The step log is the separate document you opened before starting. Each entry records the prompt you used, what you corrected and roughly how long the step took. With five steps you'll have five entries. At the end you add the total time and one sentence comparing it with doing the job without AI.
Keep to the order. If a step's output looks good enough to let you jump ahead, do the next step anyway.
Questions first: give the assistant your context and the task, and ask for up to five questions before it starts. Answer them, then ask it to restate the brief. Outline: ask for an outline with a line on what each section will contain, and tell it not to write the full piece yet. Fix the outline before going on. Section drafts: ask for one section at a time, pasting in any facts, figures or source text that section needs. Correct each section as it arrives. Critique: open a fresh chat, paste in the full draft, and ask for its three weakest points and any claims likely to be wrong. Final edit: go back to the main chat, ask for the fixes you agree with, then ask for a final pass on length and wording.
Step 4 uses a fresh chat because a chat that has just written a draft tends to defend it. Module 7 explains why this happens and what this kind of critique can and cannot catch.
What you do between steps matters more than the prompts themselves. After each output, read it against your brief and fix anything wrong before you ask for the next thing. Then write the log line for that step. Most of your corrections will be about facts only you have, and that's typical. The model still won't know your business when you finish. The chain gives you five places to put that knowledge in, where one big prompt would give you a single place.
Aisha (an invented example, also used in lesson 3.3) chose a training guide for the six graduates from lesson 3.3. The guide covered how to record their time in the firm's timesheet system, which new staff always got wrong in their first month. She had written a similar guide by hand the year before, so she had something to compare against.
In step 1 the assistant asked whether the graduates would use the system on laptops or phones. She hadn't mentioned this, and she answered laptops only. In step 2 the outline included a section on approving timesheets, which graduates don't do. She removed it and added a section on the three most common mistakes. In step 3 the billing codes section invented a code format, so she pasted in the real list of codes and asked for a redraft. In step 4 the critique pointed out that the guide never said what to do if you forgot to submit on Friday. It was a fair point she had missed. In step 5 she asked for that fix and for the guide to fit on two pages.
Her log showed 15, 10, 40, 10 and 15 minutes, a total of 90 minutes. Last year's guide had taken her most of an afternoon. She saved real time, though less than she expected, because the section drafts needed real checking. The bigger gain was in quality. Last year's guide had the same Friday gap, and nobody had noticed it for twelve months until the critique raised it.
When the piece is done, add up the times in your log and compare the total with a similar task you did without an assistant. Be honest about both methods. Count the time you spent reading and correcting as well as the time spent typing prompts. You may find the chain was faster with about the same quality, or slower with a better result. For a task you know very well, it may occasionally turn out not to be worth it. Each of these is a useful finding, and your findings feed into the module 9 decisions about which tasks get a saved template.
You finish with two things. The first is the finished piece, checked and ready to use or send. The second is the step log, with five entries (prompt, corrections, time taken), the total and your one-sentence comparison with doing the job without AI. Save your prompts too. The ones from steps 1, 2 and 4 are especially easy to reuse almost word for word on your next big task.
Now pick your task, open the log, and write the context paragraph that will start step 1.
Hand in the finished piece with a step log showing each prompt, what you corrected and the total time taken.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).