L18 · Use the method on your own project
Skill packs and prompts are provided in English so they work best with the AI tools. Use them as-is.
Use the method on your own project
Learning time: about 24.0–35.4 active minutes. Choose ONE presentation: the text (about 5.4 minutes, estimated) OR the lesson video (3:59, measured). Then 20–30 minutes of practice and checking (estimated). Do not count both presentations. Setup delays and optional extra study are separate; learner timing has not yet been piloted.
Where you are: Module 9, lesson 2 of 2. You have prepared the fictional kit for handover. Choose a real application or a clearly labelled simulation before the independent capstone.
What you will do
Translate the method to a small project, identify required permissions and choose a useful measure without claiming a business result you have not observed. Outcomes O1 and O8.
Teaching
A suitable first project has a clear user, a small output and an affordable way to check mistakes. Examples include an internal workshop budget aid, a checklist generator using synthetic information, or a local document organizer working on disposable copies.
Start with one specific moment in someone's work. A coordinator needs to compare budget scenarios before requesting approval. An administrator needs a consistent checklist before a handover. A project owner needs a clean change summary before review. Ask what the person will do with the result and which mistake would make it unusable. Name the action rather than promising to automate operations.
Use the first course's distinction between actual application and simulation. A real task needs authorization for the task, data and tools. Permission to discuss a workplace problem does not automatically allow uploading its files or connecting an account. If access is unclear, use invented data or a permitted public source and label the result a simulation. Do not create a false record of supervisor approval to make the exercise look complete.
Choose tools after defining the work. Claude Code may build one part; Codex may inspect another. A connector is useful only if a specific required input or action needs it. If the project can be completed with local synthetic files, connecting the company's calendar adds unnecessary scope. Record the reason for each tool, the access required and a fallback. A fallback may leave live integration unverified.
Measure what you can observe. You can record task time, whether an output passed named checks, the number of corrections or feedback a recipient actually gave. Faster drafting alone does not prove equal quality or savings. Compare equivalent tasks and conditions where practical, retain the quality checks and state the limits of a small trial. A demonstration with synthetic data does not establish market demand.
Worked example
A fictional training administrator wants to generate a workshop checklist. Their first idea connects email, a customer database and a public website. The immediate need is narrower: prepare a checklist for one approved internal session. A first simulation can use a local file with invented date, venue and equipment fields, generate a local checklist and test a missing-venue case.
That result can reveal whether the instructions are understandable and missing inputs are caught. It cannot prove that email delivery works or customers received anything. If the administrator later wants automatic messages, that is a new scope requiring recipient, content, permission and duplicate-send controls. Keep the first project's conclusion within what was actually tested.
Do it yourself
Complete a transfer brief with these fields: intended user; recurring situation; useful artifact; smallest version; authorized inputs; prohibited actions; chosen agents/tools and reasons; three observable acceptance checks; recovery route; evidence you will retain; decision needed before execution.
Use this complete review prompt:
Review a proposed small project for transfer from the Workshop Launch Kit course. I want an artifact that an ordinary colleague can use and inspect. Do not build, connect accounts, access workplace files, send messages, spend, publish or modify anything in this review.
Check whether the proposed user and task are specific, the smallest version is bounded, data/tool permissions are explicit, each tool has a necessary job, and the success measures support only claims we can observe. Identify a safe simulation if permissions are missing. Give a short recommendation and the smallest unresolved decision. Do not invent authorization, demand, user feedback or business savings.
TRANSFER BRIEF:
[Paste the complete brief and mark unknown fields unknown.]
Save the actual review and your response. Explain any rejected suggestion. Keep a list of decisions rather than accepting every proposed feature. Do not begin a real workplace build solely because a model says the idea is safe.
Feedback and retry
Pause here. A good transfer brief can be understood by someone outside your role. It names a person using an output, a failure that can be tested and an honest authorization state. “Everyone needs this” is not an audience. “The AI will handle it” is not a test or recovery plan.
For the retry, remove access to the most convenient connector. Decide whether a local public/synthetic input permits the same learning task, a narrower task or only preparation. State exactly what the fallback cannot prove. Then revise the claim about completion.
Before continuing, write one sentence explaining what your proposed first trial could show and one it could not show. If you have no approved real project, carry the simulation forward. The capstone uses a separate unseen case to test the course's skills consistently.
Check your understanding
Choose an answer and write one sentence explaining why before you read the answers.
- Must every installed connector be used in a project?
- A. No: select by a necessary task and permission.
- B. Yes, to get full value from the tools.
- C. Yes, unless it costs money.
- Does saving ten minutes on one draft prove ongoing financial savings?
- A. Yes.
- B. Yes, if it happens twice.
- C. No: quality, comparison and costs remain relevant.
- Can a labelled simulation be useful without demonstrating workplace adoption?
- A. No: only real workplace use counts.
- B. Yes: its claims must match the observed simulated work.
- C. Yes, and it can be described as adopted.
Answers
- A. No. Select by a necessary task and permission.
- C. No. Quality, comparison and costs remain relevant.
- B. Yes. Its claims must match the observed simulated work.
Summary
Choose a small project whose result you can inspect and whose permissions you understand. Preserve useful work while leaving unsupported claims and unapproved actions out of the next assignment.
Sources
Anthropic, MCP connections; OpenAI, MCP, reviewed 8 November 2026, support checking actual configuration and access. Transfer planning, trial measures and fictional examples are original course design, not evidence of demand or business impact.
