Small Tools for Thriving Together
GOOD COMPANY PROJECT
Three things to try, with room for doubt, corrections, and the occasional raised eyebrow. Pick whichever helps with something already on your plate. No special enthusiasm for AI required.
Confidently Wrong, Together
AI can be confidently wrong. People can too. Agreement between us is pleasant, but it is not a receipt from reality.
Before using an AI-assisted answer, try this shared checklist. The person owns the final check; AI can help organize the questions.
Name the claims. Underline dates, quantities, names, instructions, and anything presented as a fact. Ask AI to list them without adding new ones.
Find the evidence. For each important claim, open the original source or check the actual item. A plausible link, repeated answer, or confident explanation is not enough.
Check the fit. Does the source concern this place, version, date, and situation? A perfectly accurate answer to yesterday’s question can still spoil today’s plan.
Mark the gaps. Write “unknown” beside anything you cannot confirm. Ask what information would settle it. Do not let tidy prose fill the hole.
Decide what travels. Correct the draft, remove unsupported details, and keep any uncertainty the reader needs. If the missing fact matters, pause before using it.
Fictional, untested example: Rowan asks AI to turn notes about a board-game afternoon into a packing list. The notes say “four games”; the draft says “four players.” Rowan checks the original note and changes the list to “four games; player count unknown.” Nobody has tested whether this method helps a real gathering. The example simply shows the difference between checking a source and accepting a smooth sentence.
Try this prompt: “List the factual claims in this draft. Match each to the information I supplied. Label unsupported claims and questions. Do not invent evidence.”
Then check the list yourself. The fact-checker may also need fact-checking. Bring curiosity; leave the ceremonial rubber stamp in its drawer.
Give the Boring Part to AI, Keep the Judgment
Choose one small task with a result you can inspect: sorting a non-sensitive reading list, turning your own notes into headings, or drafting labels for a shared craft cupboard. Start with the work, not a hunt for somewhere to insert AI.
Before prompting, fill in these five lines:
Goal: I want ___ so that ___ can ___.
Facts: These details are confirmed: ___. These are unknown: ___.
AI’s part: Organize, compare, or draft ___. Ask about gaps; do not fill them.
My part: Supply context, decide what matters, verify the result, and approve any use.
Stop line: Do not send, buy, delete, make commitments, or act on anyone’s behalf.
For example, you might ask for three ways to group a list of craft materials you already own. You choose labels that make sense to the people using the cupboard. AI does not know whether “miscellaneous treasures” is charming or a warning sign in your household.
Set a privacy boundary first. Use invented details or information you are allowed to share with the tool. Leave out passwords, private correspondence, personal records, and details about other people. If removing names still leaves someone recognizable, simplify further or keep the task offline. Check the tool’s data settings before using work material.
Set a review boundary too. Compare the output with your inputs. Check that nothing disappeared, changed meaning, or became a made-up fact. Read it as the eventual user: is it understandable, respectful, and usable? Keep medical, legal, financial, and other consequential decisions outside this exercise.
Finish with one sentence: “AI helped with ___; I checked ___; ___ is still uncertain.” That makes the division of work visible. If the task was easier with a pencil, the pencil wins. It does not require a consolation speech.
A Small Experiment in Thriving Together
Pick a modest task you already need to do. One option: make a clear list of materials you have for a creative project, so you can start with what is on hand. Use non-sensitive details. This is an invitation to test a method, not a claim that it works.
First, define useful. Write a single sentence: “This will help if I can ___.” For a materials list, that could be finding each item without reopening every box. Choose a ten-minute limit, or another limit that fits your day.
Make a simple first version yourself. Spend a couple of minutes arranging the list. Keep it. This gives you something concrete to compare with the AI version; it is not a controlled study.
Offer AI one bounded job. Try: “Group these items into a short list. Use only the items provided. Keep quantities unchanged. Put unclear items under ‘Questions.’” Check every item and quantity against your original. Stop when your time limit arrives, even if the machine has more suggestions. Machines are generous with suggestions.
Keep a tiny record:
Usefulness: better, the same, or worse than my first version? Why?
Time: minutes preparing, prompting, checking, and correcting; total ___.
Corrections: what changed or went missing?
Judgment: what could only I decide or verify?
Necessity: did I need AI, or would my original list do?
Next time: repeat, change one thing, or skip AI?
Include disappointments. A slower result is useful information, not a failed audition for the future. One attempt cannot establish a general benefit or prove time savings.
Keep the record for yourself; there is nothing to submit. Let care for people and the living world shape the task you choose. You can value using what you already have without claiming this experiment saved resources. Small, honest observations are enough to begin.