Working With AI Day to Day
Prompts, workflows, and personal productivity that stick
Listen to executive summary
Audio length ~2 min
Full lesson about ~40 min
Executive summary
~8 min to read the written sections · ~2 min to listen to this summary
This week turns literacy into Monday-morning habits: prompt libraries for recurring work, a short verification checklist before send, and embedding AI where work already happens.
You will prioritize quality under time pressure — facts, numbers, names, tone, confidentiality — so speed does not silently increase error rates.
Outcome: durable personal and team practices, not a one-week tool experiment.
Core concepts
The terms and comparisons you need for this week’s decisions — with examples by function.
Prompt library
Saved, refined prompts for recurring tasks (meeting prep, email, analysis, RFP compare). The team reuses what works instead of reinventing weak chats each time.
Why this works: Under time pressure, free-form chat regresses to vague instructions. Libraries encode quality standards.
Why it matters: Appoint an owner for the library; review monthly; ban confidential data in shared examples.
Example: Shared “board pack summary” prompt used in enterprise ChatGPT or Claude for work.
Start a library this week
Weak / before: Everyone prompts however they like.
Strong / after: Three pinned prompts: meeting prep, customer email, document compare — each with Goal→Context→Constraints→Format.
By function
Legal
Five-line checklist before any AI-assisted external advice summary leaves the firm.
HR
Pinned prompts for FAQs; ban policy invention in constraints.
Purchasing
AI-assisted RFP comparisons still require human commercial judgment.
Marketing
Claims and names checklist before publish.
Finance
Numbers only from ERP/system of record before commentary drafts ship.
Operations
Near-miss sharing when AI invents a process step — no blame theater.
Sales
CRM is truth for price and terms; copilots draft only.
Risk
Reward verification quality in performance conversations, not only speed.
Verification checklist
A short gate before sending AI-assisted work: facts, numbers, names, tone, confidentiality. Most day-to-day risk is skipped checks, not exotic model failures.
Why this works: Speed cultures skip optional steps. Checklists make verification normal, not heroic.
Why it matters: Reward careful quality, not only first-send speed. Leaders should model the checklist in public.
Example: Before send: numbers match source? Commitments real? Any confidential paste?
Customer apology
Weak / before: AI drafted apology with invented compensation.
Strong / after: Checklist catches invented offers; prompt bans inventing credits; policy amount comes from system of record only.
Workflow embedding
Putting AI inside tools and processes people already use (email, CRM, docs) instead of orphan chat websites nobody opens after week two.
Why this works: Adoption follows friction. If the tool is outside the workflow, literacy training will not stick.
Why it matters: Prefer copilots in the system of work; measure use on the actual task, not logins.
Example: Copilot inside CRM vs a separate website staff forget to visit.
Deep dive lesson
~12 min readDeep dive: AI in daily work — habits that survive busy weeks
Learning objective. Leave with a personal verification checklist and three reusable prompts tied to real workflows — not a tool graveyard.
Context. Tools help only when they sit where work already happens and when verification is habit. This week is about Tuesday behavior, not strategy slides.
1. Embed, verify, reuse
Most day-to-day risk is skipped checks, not exotic model failures. Build a five-line checklist before send: facts, numbers, names, tone, confidentiality. Pin three prompts for recurring tasks. Protect confidential inputs even when rushing.
Reward careful quality, not only speed. If the culture praises the person who “sent first,” AI will amplify errors. Leaders should model verification in public: “I used the copilot for a draft; here is what I checked.”
Method: Five-line send checklist
Facts
Every claim has a source I trust.
Numbers
From system of record, not model invention.
Names
People, companies, products spelled and correct.
Tone
Fits audience and brand.
Confidentiality
Right tool for data class; nothing Red in consumer AI.
Worked example: AI-drafted customer apology
Situation
Service lead uses a copilot to draft an apology email after an outage. It invents a compensation amount.
How an executive thinks it through
- Numbers and commitments must come from policy/system of record.
- Checklist would have caught invented compensation.
- Prompt should ban inventing offers.
Decision / what to say
Update prompt constraints; require second check for any credit/commitment; share near-miss with team without blame theater.
Apply in your function
Sales
Ban inventing discounts or SLAs in prompts; verify in CRM.
Customer ops
Checklist before any AI-assisted external reply.
All
Pin three prompts for top weekly tasks this week.
Common mistakes
- Orphan chat apps nobody opens after week two.
- Speed praised; careful verification unrewarded.
- No disclosure norms when AI drafted external work.
Practice (10 minutes)
- Write five checks before sending AI-assisted work.
- Create three shared prompts for top tasks.
- Use both on the next three outputs this week.
Check your understanding
Key takeaways
- Reusable prompts beat one-off cleverness.
- Verification checklists prevent silent quality decay.
- Team norms multiply individual skill.
- Protect deep-thinking time; don’t automate reflection away.
Practical applications
Build a personal library of 5 prompts for recurring work.
Standardize job description drafts with inclusion review steps.
Use structured comparison tables for multi-vendor responses.
Create issue-list extractors for first-pass contract review.
Checklist before publish: claims, names, numbers, confidentiality — pin three creative prompts.
Library materials for this week
Extend this week’s decisions with briefs, cases, and primary sources that matter for your next meeting.
Output Verification Checklist
A pocket checklist before you send AI-assisted work into the world: facts, numbers, tone, confidentiality, and accountability.
Deloitte Australia — AI errors in a A$440,000 government report (2025)
Published 6 October 2025
In October 2025 Deloitte agreed to a partial refund after a A$440,000 report for the Australian government was found to contain AI-related errors, including fabricated citations. A professional-services case on verification, disclosure, and reputation — not a chatbot gimmick.
Optional focus hour
After the core lesson (~40 min), spend about 40 minutes on one long-form source matched to this week’s level.
HBR IdeaCast — leadership / AI episode (Apple Podcasts show)
Harvard Business Review
Published Ongoing series (pick a recent AI episode)
Why this week
Daily-work week: peer executive audio on ownership and culture, not tool features.
What to take away
One habit to adopt; one phrase on verification for your team.
Check your understanding
Week 9 · 3 short questions · no grades shared outside this device
Select an answer for each question.
Hands-on
~20 minBuild three power prompts
This is the compound-interest skill of AI literacy.
- Choose three recurring tasks (meeting prep, email, analysis, coaching notes).
- Write each with Goal / Context / Constraints / Format.
- Add a final line: 'List assumptions and uncertainties separately.'
- Save them where you will actually open them next week.
Reflection
- Which hour of your week is most full of low-judgment drafting you could accelerate?
- What quality standard will you refuse to lower for speed?