Editor-selected existing cases and prompts, separate from AI news. Selection dates use Hong Kong time.
Oct 26: five existing cases rechecked, not new or tested. Use shareable or fictional text, not source agents/integrations. Gemini web lists HK/MO/TW and Traditional Chinese; ChatGPT/Claude list TW, not HK/MO. Check free/basic limits; local paid prices and Chinese quality unknown. App/API differ. Human review; no staff/student scoring, account links, sending or shopping.
Text exercise: Provide consented team updates, a time period and an audience. Draft a source-linked summary, marking missing updates without judging performance.
Prompt
Summarize [consented team updates] by [period and project], identifying blockers, decisions and source links. Do not infer progress, ownership or performance; list missing updates separately. Produce a draft for [approved readers] without sending reminders or publishing.
Evidence limits and cautions
This is an interviewee’s team account without summary-quality tests; the page shows an update date, not a confirmed publication date.
These steps and prompts are original, untested exercises by this site, not the author’s verbatim workflow. Reported results do not guarantee exercise outcomes.
- Do not use summaries to score employee performance.
- Do not automatically chase people or share across channels without agreement.
Original source: Using AI to accelerate hiring and productivity at Zapier
Provide authorized materials, learner level, local curriculum goals and lesson time. Draft activities with purposes, materials, estimated duration and goal sources.
Prompt
Draft a lesson map from [authorized material, learner level, local objectives and class time]. Include activity purpose, materials, time estimates and objective sources. Adapt to [teaching language and class needs], flag teacher-review items and include no student data or validated-effectiveness claim.
Evidence limits and cautions
He saw no obvious errors on initial review but allowed that some might remain; no learning-outcome evaluation was supplied.
These steps and prompts are original, untested exercises by this site, not the author’s verbatim workflow. Reported results do not guarantee exercise outcomes.
- Foreign standards are not automatically local requirements.
- Do not use this to grade students or determine educational opportunities.
Original source: When you give a Claude a mouse
Give a text AI authorized annotations, page references and bibliography to draft a research-note draft traceable to annotations for human review. This is a drafting exercise; the source used Claude Code.
Prompt
From [authorized annotations/page references/bibliographic metadata/research question], draft a research note. Trace every point to an annotation location and distinguish author claims, my comments and open questions. State coverage limits when only annotations are available. Do not invent papers, write to Zotero or reproduce long source passages.
Evidence limits and cautions
A bounded library does not guarantee accurate summaries or interpretation of citations.
These steps and prompts are original, untested exercises by this site, not the author’s verbatim workflow. Reported results do not guarantee exercise outcomes.
- Do not reproduce long passages from papers.
- Do not upload a private library or overwrite notes automatically.
Original source: Academic Use Cases with Claude
Give a text AI synthetic error logs and redacted investigation notes to draft an internal escalation and customer-facing draft for human review. This is a drafting exercise; the source used Cursor.
Prompt
Using [synthetic error records/public documentation/deidentified investigation notes/intended recipients], draft an internal escalation and customer explanation. Separate facts, possible causes and open checks. Keep conclusions consistent while retaining necessary diagnostic detail only internally. Check the external draft for confidentiality; do not create tickets or send replies.
Evidence limits and cautions
Productivity multipliers are not adopted as general effects; internal data must not flow directly into customer replies.
These steps and prompts are original, untested exercises by this site, not the author’s verbatim workflow. Reported results do not guarantee exercise outcomes.
- Do not connect production logs or customer accounts.
- Similar errors do not prove identical root causes.
Original source: How technical support at Cursor uses Cursor
Provide synthetic clothing specifications, sizing systems, product options and a cart sample. List matches, differences and missing details.
Prompt
Using [synthetic clothing specifications/sizing system and units/product options/cart samples], check variant consistency across model, color and every sizing dimension. Without a supplied conversion, do not guess equivalent sizes. Return discrepancies only; do not shop, alter carts or guarantee fit.
Evidence limits and cautions
Finding a correct product option does not establish the same variant in the cart.
These steps and prompts are original, untested exercises by this site, not the author’s verbatim workflow. Reported results do not guarantee exercise outcomes.
- Do not infer body measurements or impose the source’s fixed style and color.
- Do not operate real carts or promise fit.
- Hong Kong and Macao are absent from ChatGPT’s current support list. Taiwan’s listing does not guarantee the reported agent, old plugin or store is available. Analyze supplied data only; do not bypass regional restrictions.
Original source: Hands on with OpenAI’s Operator