Five jobs worth trying first
The first useful experiment is usually a familiar task you can already judge. I would choose work that repeats, takes attention, and ends in a draft a person can inspect. The point is to recover capacity for work that matters.
What you will leave with
A one-week experiment with a clear success measure.
Practice time: about 20 minutesPick work you know well enough to check
Look at the last week. Where did someone turn notes into a document, rewrite the same explanation, or sort a pile of similar requests? Start where the input is available, the output has a recognizable shape, and a mistake can be caught before it affects anyone.
A first experiment should be easy to stop. Keep sending, publishing, pricing, and changes to business records outside the experiment. Use fictional or approved material and one approved tool.
Five places to look
- Customer replies. Supply the question, verified facts, and an approved tone example. Ask for a draft with unresolved details marked. Check every promise and next step before sending.
- Scope outlines. Turn site or discovery notes into included work, exclusions, assumptions, and open questions. Keep prices and commitments with the responsible person. Check that missing details have not become assumptions.
- Meeting or thread summaries. Extract decisions, owners, deadlines, and open questions. Ask for supporting source lines. Distinguish a proposal from a decision and an offered deadline from an agreed one.
- First drafts from a brief. Try a public announcement, instruction sheet, or newsletter outline. Provide the audience, facts, and format. Review accuracy and usefulness before polishing the voice.
- Sorting by explicit rules. Group incoming questions by topic or documents by type. Define categories and include "unclear." Check a varied sample and every ambiguous item before using the grouping.
These are candidates, not guaranteed wins. A short reply may be quicker to write yourself. A sensitive record may require a different process. A task with no usable source material needs that problem solved first.
A summary should preserve what was not decided
Worked example / Fictional practice material
Notes: Alex will draft an event outline by Friday. Sam suggested an evening format. No date was chosen. Room availability needs checking; nobody took that task.
Request: Separate agreed actions from suggestions. For each action, list owner and deadline exactly as stated. Mark missing owners or dates as unassigned. Include the supporting sentence.
Expected result: Alex: draft outline, Friday. Evening format: suggestion. Event date: undecided. Room availability: needs an owner and deadline.
A summary saying "Sam will book an evening event" fails. It has converted a suggestion into an assignment. This is the kind of error your review must be designed to find.
Test for a week, then decide
Time one or two examples done your usual way. Then try several similar cases with AI and record setup, drafting, checking, and correction time. Keep quality standards the same. Five cases can reveal friction; they do not establish reliability for every situation.
At the end, choose: keep using it as a draft assistant, improve the brief and retest, or stop. Count the time spent checking. A fast draft followed by a long repair session is not recovered capacity.
If it helps, name what the reclaimed time is for: faster follow-up, better preparation, or an opportunity you have been putting off. Before making it recurring team work, fill in a Reviewer Card.
Keep this
One-week experiment sheet
Copy this into your own document and fill in the brackets. Use approved or fictional material when trying it with AI.
TASK: [One recurring piece of work.]
WHY IT MATTERS: [Capacity, response time, quality, or another outcome.]
INPUT / OUTPUT: [Approved material / draft to produce.]
BASELINE: [Usual total time and quality standard.]
TOOL / ACCOUNT: [Approved tool and workspace.]
REVIEWER: [Name and the mistakes to check.]
FOR EACH CASE: [Setup + draft + review + correction minutes; errors; usable?]
SUCCESS: [A specific improvement without lowering the standard.]
STOP IF: [A consequential error, data problem, or excessive repair.]
REVIEW DATE: [Date.]
DECISION: [Keep / change and retest / stop.]
USE THE CAPACITY FOR: [The work this makes room for.]Want help applying this to your organization? Bring the job you have in mind, and we can work out a useful next step.