Chapter 22: Eight Guided AI Projects
How to use this chapter
Eight complete pieces of work, each producing something you keep. They are ordered so that early projects feed later ones, but each stands alone.
You do not need all eight. Pick the three or four that match your life and do them properly rather than skimming all of them. Every project ends with a verification step and a note on what it taught you. That second part is the lasting output.
Worksheets for each are in the Bonus Resource Library.
Project 1: Your Personal Context File
Time: 30 minutes · Feeds: everything else in this chapter
The single highest-return thirty minutes in this book. You are building a document that describes your situation once, so you stop describing it every time.
Build it. Open a conversation and ask to be interviewed:
I want to write a context document I can paste at the start of any AI conversation, so I don’t re-explain myself. Interview me (one question at a time, up to fifteen questions) then write it. Cover my work, what I’m responsible for, my constraints, my standards, and how I like things communicated.
Answer honestly, including the unflattering parts. “I over-explain and need to be cut down” is more useful than a description of your job title.
Refine it. It should be a page or less. Cut anything that would not change an answer.
Test it. Run the same real request twice, once with the file pasted and once without. The difference is the value.
Verify: Read it as a stranger would. Does it contain anything you would not want stored? Remove client names and identifying details; the shape of your situation is what matters, not the specifics.
The practical result: you have seen how context improves the answer and how an interview can produce better material than a blank page.
Project 2: The Document Interrogation
Time: 45 minutes · You need: a real contract, lease, policy, or manual
Do it.
1. Upload the document. Ask for a plain-English summary in one paragraph.
2. Ask for every obligation it places on you, with an exact quotation for each.
3. Ask for anything unusual, one-sided, or worth questioning, and ask what a specialist would notice that a layperson would not.
4. Ask what the document does not address that you would expect it to.
5. Produce a numbered list of questions for the professional involved.
Verify: search the original for four of the quotations. If any is not there verbatim, treat the whole extraction as suspect and redo it section by section.
What to carry forward: require quotations when the source is available, and use AI to develop better questions instead of merely accepting its answers.
Project 3: The Messy Spreadsheet Rescue
Time: 1 hour · You need: real, disorganized data
Do it.
1. Upload it. Ask what is there, what is inconsistent, and what looks wrong before asking for any changes.
2. Ask for a cleaning plan. Read it and approve it. Do not skip this; it is where you catch a plan to discard rows you need.
3. Have it cleaned, with code shown.
4. Ask what patterns and anomalies it sees.
5. Ask for one chart of the most interesting thing.
Verify: all five steps from Chapter 21. Row count first.
The lesson from the numbers: computing and guessing can look identical on the screen, while approving the plan before the work prevents many data disasters.
Project 4: The Recurring Workflow Prompt
Time: 45 minutes · You need: a task you do at least monthly
Do it.
1. Do the task once with AI, using full CLEAR.
2. Refine until the output is genuinely good, not acceptable, good.
3. Write a reusable version of this prompt with blanks where the specifics change. Include the constraints that mattered.
4. Test it cold on a different instance of the same task.
5. Fix what broke. Save it as entry one in your prompt library.
Verify: hand the prompt to a colleague who does similar work and watch them use it. Everything you left implicit will surface immediately.
A reusable prompt succeeds when it works for someone who was not present for the original conversation. Your own tested prompts will usually outperform a published collection.
Project 5: The Verified Research Brief
Time: 90 minutes · You need: a real question you need answered
Do it.
1. State the question and what the answer will be used for.
2. Ask for the landscape first: what the main positions are, what is disputed, what would need checking.
3. Ask for the answer, with every factual claim marked.
4. Run full TRUST (Chapter 9). Trace where each claim should live. Re-ask cold in a different tool. Check every specific at a real source. Scrutinize numbers, dates, names, citations.
5. Write the brief yourself, in your own words, from the verified material.
Verify: count how many specifics survived checking unchanged, how many were wrong, and how many did not exist. Write the three numbers down.
Your calibration: keep those three numbers. They reveal more about the tools and subjects you use than a general warning ever could.
Project 6: The Small Tool You Actually Use
Time: 1 hour · No coding required
Do it.
1. Identify a calculation or tracker you do repeatedly: a quote estimator, a unit converter for your trade, a dosage or dilution calculator, a project cost sheet, a decision scorer.
2. Describe it: inputs, calculation, output, who uses it.
3. Ask for a working version as an artifact or a single-file web page.
4. Test it against three cases where you know the right answer.
5. Refine three times: appearance, edge cases, clearer labels.
Verify: test with deliberately wrong input. Empty fields, negative numbers, absurd values. A tool that fails silently on bad input is worse than no tool.
The dividing line: you can now build small software, but testing it against known answers is what separates a useful tool from a convincing toy.
Project 7: The Difficult Communication
Time: 30 minutes · You need: a message you have been avoiding
Do it.
1. Start in thinking mode (Chapter 11): I need to send something difficult and I haven’t worked out what I actually want to happen. Ask me questions before we draft anything.
2. Once the goal is clear, ask for three approaches with different trade-offs, not three tones of the same approach.
3. Choose one and draft it.
4. Open a new conversation. Paste the draft with no context. This message is going to someone in a strained working relationship. How will it land? What will they react to?
5. Revise using that critique.
Verify: read it aloud. Then wait an hour and read it again before sending. If it still says what you mean, send it.
What changed: the maker–critic loop improved the message, while the thinking conversation helped you decide what you actually wanted the message to accomplish.
Project 8: Your 90-Day Learning Plan
Time: 1 hour to build · You need: something you want to learn
Do it.
1. State the skill, your honest starting point, your weekly time, and what “able to do it” would look like concretely.
2. Ask for a 90-day plan with weekly goals you can measure, and push back on anything unrealistic for your actual schedule.
3. For week one, run LEARN (Chapter 17) on the first concrete thing.
4. Set up your study loop from Chapter 19: active recall one question at a time, explain-back weekly, a dated log of what you got wrong.
5. Schedule a spaced review at weeks 3, 6, and 12.
Verify: at the end of week one, explain what you learned without notes and have the gaps identified. If there are none, the plan is too easy.
Evidence of progress: the plan made you retrieve knowledge instead of merely rereading it, and the dated log made that progress visible.
After the projects
Look at what you produced: a context file, a verified brief, a tested prompt, a working tool, a learning plan. That is the beginning of a personal system, and Chapter 27 assembles it into one.
If even one project becomes part of your routine, that is real progress. The goal is to discover the two or three uses that fit your life well enough to continue after you close the book.
Learn More About Artificial Intelligence
Learn more about the CLEAR, TRUST and LEARN methods for Artificial Intelligence