Mister Mobile × Claude AI · Internal

The AI Playbook v3

v1 showed what was possible. v2 showed what the team built. v3 is about making it last: using AI safely now that it can act for you, and measuring our progress every month so we keep getting better.

Version 3.0 · October 2026 Usage data: September 2026 Internal · HQ staff & HODs
Why there is a v3

AI now does things, not just answers. So we need two more habits.

Since v2, almost everyone with a seat is using Claude, and three out of four people use Cowork, which reads your files, works in your browser and finishes tasks on its own. That is where the time savings come from. It also means a mistake can now do something, not just say something. v3 adds the two habits that keep this working as we grow: stay secure and keep measuring.

What v2 got right

Copy what works from colleagues. Real, local cases moved more people than any training. Keep sharing them.

What's new in v3

Cowork in the cloud, Google connectors, how to use AI responsibly, clear rules for using agents safely, and a monthly usage report with an engagement score for every member.

What hasn't changed

AI prepares, gathers, drafts and checks. You decide, and you own the outcome, including anything an agent does under your name.

September 2026 · 1 to 30 September

Where we are now

79
Seats (63 in v2)
92%
Active in the month (90% in v2)
1,266
Person-days on Claude
4
Average active days per week
20.9k
Messages, chat and Cowork
5,400
Cowork and Code sessions

Sixteen more seats than v2, and 73 of 79 people used Claude in September, on four days a week on average. That is 286 messages per active person. Claude is now part of the working week for almost the whole office.

How people are working

Share of people using each way of working, from the 30 days to 22 September.

🖥 Cowork 73%
💬 Chat 69%
⌨️ Claude Code 16%
Two or more 57%

The good news

Cowork overtook Chat: 73% of people now use Cowork (65% in v2), ahead of Chat at 69%. People are moving from asking questions to handing over whole tasks, which is exactly the shift v1 and v2 asked for.

The gap to close

Twelve people still only use Chat, and Cowork sessions slipped 8% from v2 (4,854 vs 5,255) even with more seats. Those twelve are the fastest win for October.

New · What Claude can do now

Cowork keeps working when you walk away, and Claude is connected to our Google tools

Two upgrades since v2 change how much you can hand over. Cowork can now run in the cloud and follow you across devices, and the company has switched on connectors so Claude can work directly in Gmail, Calendar, Drive and Sheets.

☁️ Cowork in the cloud Beta on our Team plan

  • Start it and leave. Give Cowork a task, close your laptop, and Claude keeps going on Anthropic's servers.
  • Pick up anywhere. Open the same session on the web, the desktop app or your phone to check progress, answer a question or collect the result.
  • Get a ping. Your phone notifies you when the task is done or Claude needs your input.
  • Schedules run on their own. Scheduled tasks now run with no device online.

Claude can only reach files on your computer, or use your browser, while the desktop app is open there.

🔌 Company connectors are on

ConnectorWhat Claude can do
GmailSearch and read mail, draft replies, send with your approval
Google CalendarCheck availability, create and update events
Google DriveFind and read Docs, Sheets, Slides and PDFs
Docs, Sheets, SlidesCreate and edit files live, beside the chat

Turn them on: in a chat or Cowork task, click + › Connectors, switch on the ones you need and sign in with your Mister Mobile Google account.

Try one this week

Gmail + Sheets

"Find this week's supplier emails about stock arrivals and put model, quantity and ETA into a new Google Sheet."

Calendar + Gmail

"Look at my calendar for next week, find a 30 minute slot with the outlet managers, and draft the invite email for me to review."

Drive + Cowork in the cloud

"Read every SOP in this Drive folder, list where they contradict each other, and notify me on my phone when the summary is ready."

New · Monthly usage report

We now measure this every month

Management asked a simple question: are we using AI more each month? The usage report answers it. It is built from Claude's own analytics, updated on the 2nd working day of every month, and reviewed by HODs and management.

Our target

Every organisation measure higher than the month before: people using Claude, person-days, messages, sessions and the average engagement score.

The monthly usage report showing September 2026: 73 of 79 members active, average score 58, and the Organisation month on month table
The report for September 2026, our first month on record. From October, each measure shows the change from the month before and whether it is on target.

The engagement score, 0 to 100

One number per person that combines how often and how deeply you use Claude. Targets are fixed, so a higher score always means more real usage, not a better ranking against colleagues.

PartWeightFull marks at (per month)
Active days35%24 days
Cowork and Code sessions
Cowork counts as sessions or Cowork messages ÷ 3.2, whichever is higher, so long sessions count fairly		35%300 sessions
Messages, chat and Cowork15%500 messages
Outputs: artifacts, projects, PRs, file edits15%15 outputs

Points rise quickly at first and then level off, so getting started counts for a lot. Full marks means you are among our heaviest users.

Where the team sits today

29
Champion · 75 to 100
13
Regular · 50 to 74
22
Light · 25 to 49
11
Dormant · below 25

Bands from the 30 days to 22 September. See the report for the latest.

Average score: 58. That is our baseline. Lifting the Light group into Regular is the single biggest lever, and most of them only need one real task moved into Cowork.

For everyone

  • Ask your HOD where you stand and which part of your score to work on first.
  • The quickest gains come from trying Cowork for a real task and using it on more days of the week.
  • This is not to check up on anyone. It shows where help is needed.

For managers

  • Review your team's usage every month and recognise the Champions.
  • Pair Light and Dormant colleagues with a Champion for fifteen minutes.
  • Pick one or two BAU tasks a month that Claude or Cowork can take over or speed up.
New · Using AI responsibly

AI does the work. You stay responsible for it.

Claude can now read our emails, files and spreadsheets and act on them. That is what makes it so useful, and it is also why the person using it has to stay in charge. Think of Claude as a very fast new colleague: you would not send their work to a customer without reading it, and you would only show them what they need for the job.

🔍 Check before you trust

Claude can sound sure and still be wrong. Before anything goes to a customer, a supplier or management, check the prices, stock, dates, policies and numbers against the real source, such as POS, WooCommerce or the SOP.

✂️ Share only what the task needs

Most tasks do not need personal details. Remove names, phone numbers, NRIC, IMEI and card details before you paste something in, or refer to an order number instead. Passwords and API keys never go in.

🧑‍⚖️ People decide, AI supports

Hiring, staff performance, refunds, pricing and anything legal: let Claude gather, compare and draft, then make the call yourself and be ready to explain it without saying "the AI said so".

🤝 Be fair and honest

No fake reviews or testimonials, no AI images of real staff or customers without their consent, and no copying a competitor's content. Anything public must still sound like Mister Mobile and be true.

What responsible use looks like day to day

✅ Do❌ Don't
"Summarise this complaint and draft a reply" with the customer's name and phone removedPasting the full customer record, contact details and all
Let Claude draft the quote, then check price and stock in POS before sending Sending an AI quote straight to the customer
Use the Drive or Sheets connector on the folder the task needs Downloading a full customer or staff list and uploading it to a chat
Ask Claude where a figure came from, then open that source Copying numbers into a report because they look right
Use Claude to shortlist and summarise candidates, then interview and decide Rejecting someone because Claude scored them low

What happens to our data

We are on Claude's Team plan, a business plan, so Anthropic does not use our chats or Cowork sessions to train its models by default. One exception: if you press thumbs up or down, that conversation may be used. Do not rate chats that contain company or customer information.

The simple test

Would you be comfortable if the customer, your manager or the PDPC saw exactly what you gave Claude and what you did with the answer? If yes, go ahead. If not, take the personal details out or ask IT first.

New · Using agents safely

Cowork and Claude in Chrome act for you. Treat them like a new colleague.

A new colleague is helpful, fast and keen, and also easy to mislead. Agents read files, web pages and emails, and some of what they read can contain hidden instructions written to trick them. This is called prompt injection. Seven habits keep you safe.

  1. Give Cowork one working folder, not your whole drive

    Make a folder for the task and connect only that. Keep finance documents, staff records and anything with passwords out of connected folders.

  2. Approve anything you cannot undo

    Switch to Manually approve for sending messages or emails, deleting files, posting, paying or changing live systems. Never use Skip all approvals for work tasks.

  3. Stop if it goes off script

    If Claude opens sites or files you did not mention, changes topic, or asks for a password or customer details, stop the task, take a screenshot and tell IT.

  4. Keep Chrome away from sensitive pages

    Claude sees whatever is on screen. Do not open the side panel on banking, payroll, HR or payment pages, and close tabs you do not need before you start.

  5. Only use approved connectors and plugins

    Each one widens what Claude can reach. Use the ones the company has switched on, such as Gmail, Calendar, Drive and Sheets, and ask before adding a new one. Read every email before you approve sending it.

  6. Schedule only low-risk tasks

    Summaries and reports are fine. Do not schedule tasks that send messages, change records or touch sensitive files. Check past runs and delete tasks you no longer need.

  7. Check before you share a link

    Published pages and artifacts can be shared with anyone who has the link. Make sure there is no customer or staff data inside before you send one out.

The one-line version

If an action would be hard to undo, you press the button. If something looks off, stop and tell IT. Nobody gets in trouble for reporting.

Governance · updated for v3

The rules, in short

Rules 1 to 5 carry over from v2. Rules 6 and 7 are new.

  1. Claude is our default, approved platform

    It is the one we have an agreement with, control access to, and can audit.

  2. Another tool is better for a job? There is a path for that

    Request it. It goes on the approved list if the tool is data-safe, or the process keeps sensitive data away from it.

  3. Share only what the task needs

    Remove customer and staff personal details, IMEI and payment details before using them with any AI tool. Passwords and API keys never go in.

  4. No personal AI accounts for company work

    Your own ChatGPT or Gemini login sends company data somewhere with no agreement and no record. That is a PDPA exposure. If you need something, ask.

  5. Access is per person, per tool

    Nothing is enabled by default. Ask IT for what your role needs.

  6. Check before it leaves your hands New

    Anything AI helped produce that goes to a customer, supplier or management is checked by you first: facts, prices, tone and data.

  7. You own what your agent does New

    Anything Cowork or Claude in Chrome does under your account is your action. Approve anything you cannot undo, and stop and report anything that looks wrong.

Your move

Start here this month

Ask your HOD which band you are in, then do the one thing below.

Dormant or Light

Pick one task you repeat every week. Make a folder for it, connect it to Cowork, and describe the task the way you would brief a new colleague. Ask a Champion in your team to sit with you for fifteen minutes.

Regular

Use Claude on more days of the week, not just bigger days. Move one chat habit into Cowork so the work is done for you, not just described.

Champion

Turn your best workflow into something others can reuse, a skill, a project or a template, and teach one colleague. Send it in as a case for v4.

Everyone

Check AI work before it leaves your hands, take personal details out before you paste, give Cowork one folder, and approve anything you cannot undo.

For HODs and managers

  • Read the monthly update in the management channel and review your team in the report.
  • Move at least one BAU task per month into Claude or Cowork, and share it so other teams can copy it.
  • Make sure your team knows the seven safe-agent habits above. If you or your team need help with a workflow or training, reach out to Kai.