APAL 03 · AI in Action · Session 1 of 5 · Think

From documents to decisions.

The leadership job: walk into every room better prepared than the person presenting.

Sunday 20 Sep 2026
3:30–6:30 PM IST
Zoom

Printed for: · lane

5 × SundayThink · Do · Repeat · Finance · HR, then Scale

The series

Claude is the core every week. ChatGPT, Gemini and Grok are referenced where each is strongest.

18 Oct is off for Durga Puja. No session that Sunday.

A parallel thread, with its own ProjectThe Chief of Staff is one dedicated Project, started today, that grows every Sunday: hired, gets hands, gets a rota, learns finance, learns people decisions. It is a thread, not the syllabus.

3:30Set yourself up

Before we start

Pick your function in the bar above. Every prompt, example and table on this page changes to fit it. Then pick your lane. Your choices are remembered on this device. We begin with a live demonstration at 3:35. For those ten minutes, just watch.

The one rule on names and numbers, all dayLabels, not names. Rounded numbers, never exact figures, never anything price-sensitive. Policy says no: give no numbers and use the fictional packs. Policy is silent: stop at labels and rounded numbers, and send IT the four-line note this week.

What today moves for you

One use case, one number, and how to take the "before" reading. You log it in the Workbook during Build 1 and post it in chat at 6:15.

Today's row of the Function Outcome Matrix

Use case today

The number it moves

Baseline it like this

Three lanes

Choose the one that matches your set-up. The Phone lane adds phone notes under both builds. The Advanced lane adds a stretch task in every block.

Most of the room

Standard

Laptop browser at claude.ai. Any plan.

Office laptop blocks AI sites, or you work from a phone

Phone or locked laptop

Personal phone: the Claude app or the phone's browser. Nothing to install on the office laptop. Ideal set-up is two devices: Zoom on one, Claude on the other. Use only the fictional Nivara documents or a public document. Phone notes sit under both builds.

Daily AI users

Advanced

A stretch task in every block, marked Stretch. None of them is filler.

Phone or locked laptop The lane, as one card

Zoom stays on the office laptop. Claude runs on your personal phone, in the app or the phone's browser.
Before 3:30, download the three practice PDFs and your function's short extract to the phone. Open the Activity Pack on the same device, so copy and paste works.
Dictate. Do not type long answers on glass.
Post to Zoom chat from the laptop. One short line is enough.
Log your numbers in the phone's notes app. Fill the Workbook tonight.
Any Project step the phone refuses: do it tonight on any browser. The chat-only route is written under Build 2.
Fictional or public documents only.

Voice and language

Answer the interview in short phrases, in whatever language you think in. Claude's own dictation supports English and Hindi (other languages are in beta). For Gujarati, Marathi, Bengali, Tamil and others, use your phone keyboard's microphone, or type. Claude reads and replies in all of them. The standing instructions are always written in English.

Your company's approved tool is something else?

Microsoft Copilot, ChatGPT Enterprise or Google Gemini at work: the method transfers. The prompts are plain text, so try H1 to H3 and D1 to D3 there. We have not tested every tool. Practise today on Claude with fictional documents. Do real-document work in the tool your company has approved.

Stuck?

Send STUCK + step number as a private message to the host in Zoom chat. Nobody else sees it. The facilitator checks private messages at 4:10, 4:50 and 5:50. If an IIMC programme coordinator is on the call, they watch for these and reply. The facilitator clears the common ones at the checkpoints. During Build 1 the facilitator runs the drill live on screen, so anyone who hits a plan limit can follow along and finish tonight when the limit resets.

Free plan or Pro, prompt by prompt

Shown on screen before Build 1 and again at 6:05. Rows are taken from Anthropic's help pages, read on 18 Sep 2026. Free and paid plans both have usage limits that reset on a rolling five-hour window. Pro gives at least five times the free allowance.

Step	Free	Pro	If free blocks it
Prompt Lab, CRISP briefs	Yes	Yes	No action needed
Project to hold them	Yes, up to five Projects	Yes	No action needed
Upload the 30-page Nivara pack, run D1–D3	Yes, but this uses much of a free allowance	Yes	Use your function's short extract (8 to 13 pages). Or follow the facilitator's live run and finish tonight
Upload the 9-page Owner Pack or HR policy, run D1–D3	Yes, comfortably	Yes	No action needed
On free, protect your allowance: in the Lab run only your rewritten prompts, not the weak ones. Use your function's short extract in Build 1. Skip the A1 try-it	Your allowance is limited	Not needed	Then Build 2 is safe
The whole afternoon on one free account	The facilitator tests this on Saturday, start to finish, and shows the result on screen	Yes	If the allowance runs out before 5:38, do Build 2 tonight. It is the same four steps
R1 research with sources	No. Research is for paid plans. Free has ordinary web search	Yes	Watch the demo. Or ask the same question in a normal chat with web search on
Scheduled tasks (Session 3)	No	Yes	Not needed today
H1–H3 Chief of Staff interview, instructions, Context File	Yes	Yes	No action needed
H4 hand over a day, with a document	Yes	Yes	No action needed
A1 interactive one-pager	Yes, once file creation is on under Settings → Capabilities	Yes	No action needed
M1 morning brief, daily	Yes	Yes	No action needed

3:534 minutes · everyone does it

Two settings and one rule

Training off, file creation on, labels not names. We do it together.

Setting 1
Settings → Privacy → Help improve our AI models. We recommend off for work use. On the phone app, Settings is in the side menu. The facilitator shows both.
Setting 2
Settings → Capabilities → switch on file creation. You will need it at 5:13.
The one rule
Labels, not names. Rounded numbers, never exact figures, never anything price-sensitive. Until 4:16 you type nothing about your own company. The Lab runs on the fictional case.

For daily users of another tool

Find three answers for the tool you already pay for: is it trained on, how long is it kept, which plan changes that. Log them in Model Triage after 4:16.

3:57Module 1 · 19 minutes · hands on keys

Brief it: CRISP and the Prompt Lab

Most people use an AI assistant like a search box. A few use it like a capable colleague. Same tool, same price. The difference is method.

CContext

RRole

IInstructions

SSpecifics

PPain points

The two lines that do the most work

Open with

Ask me questions before you start.

Close with

End with what you would verify first.

Your before-and-after

One example is on the facilitator's screen. Your own function's version is here.

The Prompt Lab

15 minutes. Individual, on Zoom, no breakout rooms. Fictional case only. Three weak prompts, P1 to P3, five minutes each. Each is a one-line prompt a busy leader really types. Two good rounds beat three rushed ones.

Paste the Nivara context block first, so you do not need your own company.
Owner or MD: your block says "I own a ₹150 Cr components maker that supplies Nivara." It is already set below.
Run the weak prompt.
Rewrite it with CRISP. Run it again. Compare.
On a phone: dictate your rewrites.
After each round, paste your best rewritten line in chat. The facilitator reads two aloud and says what made them work.

Advanced lane, from 3:57

Skip P1. Do P2 and P3 in two tools, on the same tier, and note which answer you would send. After 4:16: Workbook, tab Model Triage. Take the last task where your usual AI tool disappointed you. Rewrite it with CRISP, using labels and rounded numbers only.

4:1612 minutes · questions in chat

Is my data safe?

We answer this before anyone types anything about their company. You switched training off at 3:53. Now the rules. Said aloud: the three-plan table, the traffic light, the four plain lines, one labels example and the incident line. Everything else in this block is here for reading later. Questions go in chat. Anything needing more than a minute is answered in writing to everyone by Tuesday.

Three plans, three answers

Plan	Used to train models?	How long kept	Source, and date checked
Personal: Free, Pro, Max	You choose. The setting is Help improve our AI models, under Settings → Privacy. Incognito chats are never used	On: kept, with your identity removed, for up to five years. Off: your chats stay in your history until you delete them, then leave Anthropic's systems within 30 days. A chat flagged for safety is kept up to two years. Off does not mean nothing is stored	privacy.claude.com, articles 12109829 and 10023548 · read 18 Sep 2026
Company-bought: Team, Enterprise	No, by default. Feedback you submit is the exception	Kept until deleted, unless an Enterprise administrator sets a retention period (30 days minimum)	privacy.claude.com, articles 7996868 and 7996866 · read 18 Sep 2026
Your company's AI policy	Overrides both	Not applicable	Your IT or compliance head

Where is it kept? Anthropic states that data is stored in the US, and traffic may be routed through the US, Europe, Asia and Australia. India data residency exists for software developers building their own applications, not for the claude.ai app you are using. (privacy.claude.com article 7996890, read 18 Sep 2026.)

The traffic light

Check every document against this before it goes in.

Green

Fine on any plan

Public documents
The three fictional packs
Your own writing that holds nothing personal or confidential
Your day described as meeting types ("board prep", "plant review")

Amber

Company-approved plan only

Internal documents that are not confidential
Internal policies
A calendar export with names and subjects
Your plant's or region's own performance numbers

Red

Never on a personal plan

Customer personal data
Anything about a named or identifiable employee
Ledgers and bank statements
Board, committee and asset-liability committee papers
Unpublished financial results
Customer drawings, customer audit findings, supplier price lists
Unlaunched products, media plans, research verbatims, agency commercials
Appraisal notes, even your own
Anything under NDA or a confidentiality clause

Four plain lines

DPDP

The Digital Personal Data Protection Act makes your company answerable for every person's data it holds. That includes your own employees. Pasting it into a personal AI account moves it outside the company, and you did the moving.

Listed companies

Unpublished price-sensitive information stays out. SEBI's insider-trading rules have no AI exception.

RBI-regulated entities

RBI's Managing Risks in Outsourcing Directions, 2025 and its Cybersecurity, Technology: Risk, Resilience and Assurance Framework Directions, 2026 replaced the 2023 directions. They reach AI tools through your own outsourcing and information-security policies. (Read on rbi.org.in: the NBFC outsourcing direction, id 12941; the NBFC cybersecurity direction, id 13592; the commercial-bank cybersecurity direction, id 13643. The commercial-bank outsourcing direction is from press. Confirm all four with your compliance head.) A personal AI account is not an approved system. Board, committee and asset-liability committee papers are Red. Lenders: nothing about any borrower's account or repayment record.

Owner-managers

You are the policy, so decide once. Red stays Red even for you: bank statements, employee records, customer personal data. Your own sanction letter: yours to decide, after redaction. A customer contract with a confidentiality clause is Red, even redacted. Learn the method on the fictional Owner Pack, then take the real contract to your CA with sharper questions, or onto a company plan.

Labels and rounded numbers, out loud

If your company's policy allows it, this is what speaking in labels and rounded numbers sounds like.

Sales

"Dealer A, about ₹6 Cr a year, payments slowing."

No individual's name or phone number. No ledger or CRM exports.

Operations

"Plant 2, on-time delivery down from the mid-90s to the mid-80s since May."

No customer names, no part numbers.

Policy says no: fictional packs or public documents only. Policy is silent: this is as far as you go, and you send IT the four-line note this week. Owner-managers decide this for themselves. The note is in the Monday Card section, ready to copy.

RBI-regulated and listed entities: treat yourselves as "policy says no" until compliance tells you otherwise.

Amber is about the document and the exact figures. Speaking in labels and rounded numbers is the most a silent policy allows.

If someone does paste personal data by mistake

Delete the chat, and tell your Data Protection Officer or IT head the same day.

Project files and memories

Treat them like chats. They stay until you delete them. No Anthropic page states a separate retention rule, so this is unverified.

Redact in Word

Not said aloud in the session. No IT team needed.

A digital PDF · about 15 minutes

Open it in Word. Use Find and Replace to turn each name into a label.
Delete account numbers, addresses, prices and drawings.
Save as a new file.

A scanned letter · about 40 minutes

It will not open as text. Retype only the four sections you need: covenants, fees, security and guarantees, events of default. Use labels as you type.

A company plan (Team)

It needs a minimum of two members. Press reports put Team Standard at about ₹2,399 a seat a month on annual billing, so roughly ₹4,800 a month for the two-seat minimum. Check claude.ai/upgrade.

A reminder, not a control

Later today you will put seven red lines inside your first Project (H2). That is a reminder, not a control. By the time it objects, the data has already left. The control is you.

Questions that need more than a minute are parked and answered in writing to the whole cohort by Tuesday. Team rules come back in depth in the Scale close of Session 5. A Manager's Rule Card is in the Workbook today, and its text is in the Monday Card section below.

4:28Module 2 · 12 minutes · watch

Documents to decisions

Back in the demo chat, with the fictional Nivara board pack already uploaded. The memo, the push-back, the page check, and where it fails. These are the prompts you see on screen.

1. The memo

D1, in the CXO version: what is being decided, five facts with page numbers, where the pack contradicts itself, three questions. Your own version of D1 is in Build 1.

2. The push-back

Then the PDF is opened to the cited page, on screen, and one number is checked together.

When you repeat a finding in a meeting, cite the document and the page. Never cite the assistant. Save the memo beside the document as a working paper, with the date and the tool used.

3. Where it fails

Pushed past its sources, it can invent.
It sounds equally sure when right and wrong.
Errors hide in numbers, dates and names.
It reflects patterns in what it has read, so never let it rank, score or screen people.

The rule

It drafts, you sign.

4:40Build 1 · 25 minutes · hands on keys

The dreaded-document drill

Memo, push-back, meeting brief. The facilitator runs the same drill live on screen throughout, one step behind the room: the board pack first, then five minutes on the Owner Pack from 4:45, so a stuck owner has a matching screen. Lenders: read the Owner Pack's sanction letter from the bank's side.

Your route through the drill

Safe documents for your function: .

The Owner Pack is fictional: a supply agreement and a bank sanction letter. It is in the case pack, and it is the right document for you today.

The six steps

2 min
Pick your document. Default: the Nivara board pack. On the free plan, use your function's short extract (8 to 13 pages), not the full pack. HR uses the Nivara HR policy. Owners use the Owner Pack: a supply agreement and a bank sanction letter. Operations: if you run plants, use the board pack. If you run procurement or supplier quality, use the supply agreement in the Owner Pack, read from the buyer's side. Or a public document from your function's safe list. Check it against the traffic light.
9 min
Run D1 for your function. HR's default is the HR policy with D1-HR, which produces a manager FAQ. The board pack is HR's stretch. At 4:50 the facilitator shows a saved D1-HR output for one minute. Finance with a circular runs D1-REG.
5 min
Run D2 on the finding that matters most. Open the document. Check one number yourself.
6 min
Run D3, the meeting brief for your most important meeting this week. Sales runs D3-S instead, because your real meeting is a dealer review, not a Nivara board. HR: run D2 on the most ambiguous clause, then D3 for the policy roll-out meeting.
2 min
Log it. The Workbook is an Excel file. It opens in Excel, Google Sheets and the Excel phone app. Tab Baseline and Target: minutes this prep took you last time, and what it found that you would have missed.
then
Post in chat: one thing it found.

On a phone

Download the PDF first, then upload it with the plus icon. For the page check, open the PDF in your phone's viewer and jump to the page. For the log, note the two answers in your notes app and enter them in the Workbook tonight.

Hit a plan limit?

Use the short extract for your function in the case pack, or follow the facilitator's live run and finish tonight.

Stretch

Run the same D1 in ChatGPT, Gemini or Grok, on the same tier, and log what each found in Model Triage. Marketing: start R1 on your own category in a second tab. On the free plan, upload a published category report and run D1 instead.

The prompts for Build 1

Square brackets are yours to fill. Every prompt has a Use-when and a Verify. D1 below is already set for your function. The prompt made for your function opens by itself. The others stay folded, and you can open any of them.

D1 for every function, at a glance

Useful when you brief a colleague from another function.

Function	Sceptical reader	Look especially at	Safe documents

Break at 5:05. Back at 5:13. File creation must be on, under Settings → Capabilities. Not finished? Run D2 now. It takes three minutes.

5:13Module 3 · 8 minutes · watch, then try

Artifacts: your memo becomes something you can click

Claude does not only describe things. It builds them, beside the chat. In the demo, the decision memo becomes a clickable one-page brief with tabs. Then one minute for owners and procurement heads: from the Owner Pack's supply agreement, "build me a days-to-cash calculator". Refine it in plain English: "Make the risks a red, amber, green table." It edits itself. It can be shared as a link.

What to ask it to build

Stretch Build this artifact from the Nivara pack. Then build the same one in the other tool you use, on the same tier, and compare.

Try it · 2 minutes

Paste A1 under your own memo from Build 1. On the free plan this works once file creation is on, under Settings → Capabilities.

5:21Module 4 · 7 minutes · watch

Research: a sourced brief, not a confident guess

Research is for paid plans, so this is a demo. Free users can ask the same question in a normal chat with web search on. In the demo the memo flagged a motor supplier change, so R1 asks: "Who supplies small-appliance motors in India, what changed in the last 12 months, with sources." It plans, searches, reads and reasons across sources, visibly.

Do-along

The facilitator pastes one cited link in chat. Everyone opens it. Is the claim on the page? Is the page dated inside the last 12 months? Answer yes or no in chat. Trust is built per claim, not per tool.

The prompts

R1 carries your function's lens. R1-F is the same brief in a normal chat with web search on, for the free plan.

Stretch

Run R1 or R1-F on your own category, vendor or customer, from public information only, in two tools on the same tier. Count the working links in each.

5:28Module 5 · 10 minutes · listen, then do

Projects, memory, connectors

Stop re-introducing yourself every conversation. A saved working relationship.

Deliberate and controllable

Projects

A Project is a room for one role or one goal. It holds instructions (the standing brief, written once), files (what it should always know) and every conversation about that work. Free plans can hold up to five Projects. Rule of thumb: the third time you paste the same context, it should have been a Project.

Across chats

Memory

Claude also remembers across chats: your preferences, your role, your past work. You can read every memory, edit it and delete it, under Settings. Projects are the deliberate, controllable version. Incognito chats are remembered nowhere.

We only look today

Connectors

Claude can be connected to the tools you work in: Google Drive, Gmail, Calendar, Microsoft 365, Slack, Notion and many more. Connecting company accounts needs your company's approval, and we do it properly on 4 Oct, in Repeat.

Your first three Projects, after today: your Chief of Staff, one key account or vendor, your voice.

What good instructions contain

Good instructions have four parts: role, context, output format, constraints.

"I already have custom instructions in ChatGPT."

Good, the method transfers. Custom instructions are one global paragraph. A Project is a room: instructions, files and every conversation about one role. ChatGPT has Projects too.

Do-along · 2 minutes

Create an empty Project now

Name it Chief of Staff — [first name]. You will fill it in the next block.

No Projects on your phone? Open a new chat and name it the same.

Stretch

Tell Claude three standing preferences: amounts in ₹ Cr, one page, decisions first. Ask it to remember them. Then open Settings and read the memory it saved.

5:38Build 2 · 27 minutes · hands on keys

Your first Project: the Chief of Staff thread

A 7-minute demo, then 20 minutes to hire your own. In the demo: a new Project, H1, an interview answered partly by voice, standing instructions it drafts itself, the red lines with H2, then a Monday handed over with H4: the day written as meeting types, and the board pack.

From the demo

Look at how Aditi wrote her day. Meeting types and roles. No names. She left her one-to-one out completely, because it is a people matter. That is how you will write yours.

Hire your own: four steps

20 minutes. The clock marks tell you where you should be.

by 5:48
Open the Project you created at 5:36. Copy H1 below. It is already set for your function, and the copy button adds the word batch at the end, so it asks all ten questions at once. Paste. Fill the square brackets. Send.
On a phone: Projects sit in the app's side menu. No Projects on your phone? Do steps 1 to 3 in one chat. Copy the approved instructions into your notes app. Create the Project tonight on any browser.
by 5:57
Answer all ten. Dictate. Short phrases. Any language. Labels, not names. Rounded numbers, never exact figures, never anything price-sensitive.
On a phone: tap the microphone on your keyboard.
by 6:02
Read the instructions it drafts. Delete anything it assumed. Add the red lines with H2. Paste the result into the Project's instructions box.
On a phone: long-press to copy. Project → instructions → paste. If the app will not let you edit, do this step tonight.
by 6:05
Test the red line with the made-up "Rohan Mehta" line under H2. It should decline to judge him. Type HIRED + your function in chat.

Policy says no, or you would rather not describe your role today?

Hire a Chief of Staff for Aditi instead. Aditi Rao is the fictional COO of Nivara Appliances. Her interview answers are in the participants' folder, and below. You learn the whole method and type nothing about yourself.

Aditi's prepared answersPaste H1, send, then paste these. Open to view.

The prompts for Build 2

H1 below is already set for your function, with the word batch added at the end.

The red lines inside the Chief of Staff are a reminder, not a control. By the time it objects, the data has already left. The control is you.

Tonight, not now

Prompt H3, the one-page Context File, saved into the Project. Then the full one-question-at-a-time interview if you want a sharper brief.

Hand over a day

The prompt from the demo. It returns the three things that matter, the clash, the meeting you are not ready for, and what you are putting off.

Stretch, pick one Advanced

Diff

Paste your existing ChatGPT custom instructions. Ask: "Compare these with the standing instructions you just wrote. What does each know that the other does not? Merge them."

Board Sceptic

A second Project. Upload the Nivara pack into it and paste the H2 red lines. Instruction: "You are a retired CFO on my board. You have the pack. You believe nothing without a page number. Grade any memo I paste from A to D against the pack, then list what is unsupported."

Marketing: the Brand and Category File

Run prompt H3-B. Public information only.

Sales: a tool for your regional managers

Ask for a one-page account-plan template and the five questions you will ask in every account review. This goes on your Monday Card.

6:0510 minutes · watch

The other three

One pack, four tools: the identical D1 on the Nivara pack in Claude, ChatGPT, Gemini and Grok. If the facilitator ran the test on Saturday, the slide shows the dated result, whatever it is, including if Claude loses. It is a demonstration, not a benchmark, scored by the facilitator against his own key. You can run it yourself tonight. For any comparison, all tools must be on the same tier. If you ran the stretch, post your own counts in chat.

Pick per job, not per headline

Claude rows: Anthropic's own pages. Grok row: xAI's developer pages. ChatGPT and Gemini rows: press reports. All read 18–19 Sep 2026.

The job	Start with	Try this tonight
Read a long document and argue with it	Claude	Run D1 and D2 on the same document in two tools
Research a market or a company, with sources	Claude Research, ChatGPT deep research, or Perplexity	Ask all three the R1 question. Count the working links
Keep context about me and my role across weeks	Claude or ChatGPT	Both have Projects and memory. Paste H1 into each and compare in a week
Talk out loud, hands-free	ChatGPT	Hold a five-minute spoken conversation in each app tonight and compare
Make an image or a short video	Gemini or ChatGPT	Claude does not generate images
Work inside Gmail, Google Docs and Google Sheets	Gemini	It is built into those apps
Know what people are saying in public, right now	Grok	It searches live posts on X as a built-in tool. Ask it what people said about a launch today, then open two of the posts it cites

Where Claude loses today

Image generation, voice, live public conversation, and the size of the free allowance.

Which Claude model

Use the default. Nothing in this session needs you to choose another one.

One caution on Grok, said plainly, with its sources

X states that public posts and Grok chats are used for training by default, until you switch that off. Check the setting yourself, under Privacy in X. The European Commission's website records a formal investigation into X over Grok, opened on 26 January 2026. Press reports say India's IT ministry served X a notice in the same month. Use it to read public mood: a launch, a policy announcement, a brand flare-up. Do not use it for regulated or board-grade work.

Free or paid: three honest answers

Salaried leader

Check claude.ai/upgrade for the price you are shown. Press reports put Pro at about ₹2,399 a month, GST included, by card. The formula: monthly price ÷ the value of your hour = the hours it must save. At ₹2,000 an hour that is about 72 minutes a month.

Owner-manager

Write down what your CA or lawyer charges to read one contract, and how many days you wait. Illustration only: at ₹10,000 a review, arriving prepared and halving two reviews a month is worth ₹10,000 against ₹2,399 on Pro, or about ₹4,800 on the two-seat company plan. Put your own figures in the Workbook's ROI tab. Compare that with reading it tonight and going to your CA with three sharp questions. It does not replace your CA's signature. It makes you a better-prepared client. For real confidential contracts you would need the company plan: two seats minimum, so roughly ₹4,800 a month on press-reported prices. Do the same arithmetic with that figure. If that happens twice a month, pay. If you have not used it on five working days in a month, cancel.

Already paying for ChatGPT Plus

Run both on the same real work for four weeks. Keep the winner. Cancel the loser.

New to this? Stay on free this week. Already paying for another tool? A fair comparison needs a month of Claude Pro: paid against paid.

6:1515 minutes · hands on keys

Your Monday Card

The function round, three actions and the 30-day habit. None of the actions needs IT permission.

Function round, in chat

One line: function · the metric this will move · my baseline guess. Your metric: .

Three actions

Action 1

One dreaded document before your next review

D1, then D2, then D3, on a public or fictional document, or on a real one inside your company's approved tool. Log minutes before and after. Save the memo beside the document as a working paper. No approved tool yet? Send IT the four-line note from the Workbook. It is also below, ready to copy.

Action 2

Brief, do not ask

Every request to Claude this week gets CRISP, and ends with "what would you verify first?" Ten minutes a day.

Action 3

Finish the Chief of Staff thread

Regulated or "policy says no"? Keep it at role level: no company name, no numbers, meeting types only. All of that is Green. Red lines tested, Context File saved with H3. Then the morning brief, M1, every working day: you say only what differs from your usual week. Meeting types only. Leave out PoSH, disciplinary, medical and legal matters entirely.

Your function's version of action 1

For you

Regulated or locked down

A listed peer's annual report, an RBI or SEBI circular with D1-REG, a published tender, or an OEM's public supplier manual. Search "[OEM name] supplier quality manual PDF".

Advanced card

One real brief, delivered to a real person by Friday, produced in two tools on the same tier. Log which one you sent, and the minutes it took against your baseline. If you already pay for another tool, compare paid against paid: that is the fair test. Then run X1.

Tonight, in this order

Finish Build 1 if your plan ran out.
Finish the hire: red lines tested.
Run H3 and save the Context File.
Set tomorrow's morning brief, M1.

The 30-day habit

One habit a week. Tick each week as you finish it. Your ticks are saved on this device.

The prompts for the week

Pass it on

One page your managers can copy, with the rules on top. Translate the six rules into Hindi, Gujarati or your team's language if that is how your managers read: ask Claude to do it, then check it yourself. The page opens with the traffic light, the DPDP line and the Manager's Rule Card. Then three things: the CRISP card, D1 and D2 together on public or fictional documents, and the M1 morning brief. Sales leaders also pass on D3-S and the S1 template. Where the company has an approved tool, managers use that. Where it does not, they stay on the free plan with Green documents only. Your company's AI policy overrides this page.

Manager's Rule Card

Six rules for anyone on your team using an AI assistant this week

Public or fictional documents only, until the company approves a plan.
Nothing about any identifiable person. Not a customer, not a colleague, not a candidate.
No ledgers, board papers, drawings, prices, or anything under NDA.
Check every number at the source before it leaves your desk. Cite the document, never the assistant.
It drafts. You decide, and your name is on it.
Pasted personal data by mistake? Delete the chat. Tell the Data Protection Officer or IT head the same day.

It decides nothing and signs nothing

When your CEO or your team asks what this is, say this.

On time saved

The facilitator's estimates, not a study. The Workbook asks you to measure your own, and to log issues it found that I would have missed. We report the cohort's real numbers in Session 2.

Reading a 30-page pack properly

60 to 90 minutes. Memo: about 10 minutes to produce, 15 to check.

Operations review prep

Plant teams often spend 6 to 10 hours a month building the deck. Today's gain for operations is findings caught and minutes to a first risk memo. The hours come back in Session 2, when the deck's first draft is delegated. We will measure the after then.

Sales review prep

30 to 45 minutes per key account, mostly chasing facts. The brief does not remove the chasing. It turns ten vague calls into three exact questions. Your own numbers arrive in Session 2.

Marketing briefs

A category brief often takes a team three to five working days. A sourced first draft arrives in minutes. The checking still takes an hour.

Owners

The saving is days, not minutes. You stop waiting to understand your own contract.

Finance

A first list of challenge questions on a monthly pack often takes half a day across the team. D1 gives a first list in ten minutes. Checking it at source takes an hour. The judgment stays with you.

HR

A manager FAQ for a policy often takes two to three working days. D1-HR gives a draft in minutes. HR and legal review still takes its usual time.

The commitment form

Name · function · the document I will run · minutes my last review prep took (sales: and the number of chasing calls) · one thing Claude found today that I would have missed · Chief of Staff hired, yes or no · the day I will do action 1 · one question I want answered by Tuesday.

5 × SundayThink · Do · Repeat · Finance · HR, then Scale

Your function across five Sundays

Each card shows the use case and the number it moves. Every people use case works on aggregates only: cohorts, templates and role-level patterns, with no group smaller than ten people. No individual is ranked, scored or screened.

18 Oct is off. No session that Sunday.

27 SepSession 2 · Do

Next Sunday

Next Sunday is Do. Claude gets hands.

Session 2 pre-work

We supply a fictional Nivara sales workbook and monthly management report. Use those. Do not bring company data.

Your row for 27 Sep