An AI collaborator, on paper.

Before anything is built, you should be able to read exactly what an AI collaborator will do, what it can see, and where your people decide.

First, the film. Then a separate example on paper.

See the options. Keep the plan current. · 1 minute 40 seconds, narrated, with captions. A fictional company’s planning decision: a clear comparison of the options, a weekly check that keeps the plan current, a proposed date change with what it leaves open, and the people who decide. Working prototype: recorded AI run, live calculations, simulated delivery.
Read the film’s on-screen text
  1. 0:00 Illustrative excerpt · fictional company
  2. 0:00 Planning meeting, transcript export · Week 1
  3. 0:00 Project team: It’s all in there, but only you can really read that workbook.
  4. 0:00 Leadership: I need to see the choice, not the spreadsheet.
  5. 0:00 Project team: Can we see our options clearly—and keep the plan current?
  6. 0:05 DeepMoat AI working brief. From the request: “Can we see our options clearly—and keep the plan current?”
  7. 0:07 The decision: Six weeks or eight for the agreed scope.
  8. 0:09 What to check: Capacity, commitments and the estimate.
  9. 0:11 Who decides: Leadership, at the next review.
  10. 0:11 Scope agreed with the team.
  11. 0:13 Meeting notes · Planning workbook · 30-page briefing.
  12. 0:15 Planning workbook, as received. Weekly capacity used for the plan: 120 h.
  13. 0:15 Capacity already committed.
  14. 0:18 80 h already committed · 40 h available.
  15. 0:19 Availability confirmed with the team.
  16. 0:23 Planning workspace · Working prototype · Version 1. 6 weeks: 240 hours, modeled capacity over 6 weeks. 300 hours, estimated work. 60-hour shortfall.
  17. 0:24 Working prototype. Designed with DeepMoat AI.
  18. 0:28 Illustrative assumptions: fictional figures; weekly averages, confirmed with the team; committed work comes first; the estimate covers the agreed first scope; a comparison, not a schedule; headroom isn’t approval.
  19. 0:36 8 weeks: 320 hours, modeled capacity over 8 weeks. 20-hour headroom.
  20. 0:43 Team feedback: Compare both options side by side.
  21. 0:48 Refined with DeepMoat AI.
  22. 0:48 Compare side by side. Option A, 6 weeks: 240 hours, 60-hour shortfall. Option B, 8 weeks: 320 hours, 20-hour headroom. Both options use the same inputs: 40 hours a week available, 300 hours of estimated work.
  23. 0:51 Working prototype: recorded AI run, live calculations, simulated delivery.
  24. 0:51 Weekly plan check · AI teammate · Mondays at 07:00 · fictional data.
  25. 0:55 What changed: Training coordinator, 15 → 19 hours a week on other work. From the update: “…from 15 to 19 a week…”
  26. 0:56 Why it matters: The 8-week plan is now 12 hours short. 288 hours against 300 needed · was 20-hour headroom.
  27. 0:58 Who approves: Project lead. Status note drafted. Every number in it appears in the plan or the updates.
  28. 1:00 Approved by the project lead · Ready to send · simulated.
  29. 1:02 Working prototype: recorded AI run, live calculations, simulated delivery.
  30. 1:03 What changed: Training, Tue, Week 6 → Tue, Week 7. Training coordinator, Wed 09:14: “…Tuesday of Week 7…”
  31. 1:06 Why it matters: The pilot needs the training first. Pilot can’t start Mon, Week 7. Owner: Project lead.
  32. 1:08 Pilot start proposed: Wed, Week 7.
  33. 1:10 Who approves: Project lead reviews the proposal.
  34. 1:10 Open: Capacity: 12 hours still to resolve. Open: Wednesday pilot availability: not confirmed.
  35. 1:11 Approve date-change proposal.
  36. 1:12 Date change approved by the project lead · Ready to send · simulated. [Both open items stay on screen.]
  37. 1:17 What matters · What can be automated · What needs your judgment.
  38. 1:23 The same approach fits mid-sized and large companies, and family offices: the meetings, the reports, the decisions.
  39. 1:32 Book a working session · 45 minutes · $375. Includes a meeting summary. Further work is scoped separately.
  40. 1:32 deepmoat.ai · A fictional example. Narration is AI-generated.

Rather talk it through? Book a working session · 45 minutes · $375.

A separate example, on paper.

The film follows one fictional company. The paper below is a different fictional example: Harbor & Pine, a property-services company, and its quotes and follow-ups.

Many people would call it an AI employee. Here is the paper for a fictional company: the summary of a first conversation, the blueprint for one AI collaborator, a sample week and the first month. The company is invented. The questions are the ones we agree before any setup.

It starts with one conversation.

Maya Reyes started Harbor & Pine Property Services with one truck. It now has 140 people in three offices, and quote requests arrive faster than the office teams can answer them. She booked a 45-minute Signal Triage ($375) about quotes and follow-ups. This is the kind of summary she would receive afterwards.

Harbor & Pine · Signal Triage summary

Sample meeting summary · fictional

Quotes and follow-ups are the place to start.

Your goal
Reply to every quote request within one working day, in all three offices, without adding office staff.
What is happening now
About 400 requests a month arrive by web form, email and phone. Each office writes quotes by hand from its own copy of the price list. Follow-ups depend on who remembers, and some requests wait days for a reply.
What would help
One AI collaborator that drafts quotes from a single shared price list and keeps one list of open requests across the three offices, with an office manager approving everything that goes to a customer.
Questions to explore
Is the shared price list complete enough to draft from? Who owns the work in each office? Which requests should never be drafted, such as complaints and large commercial bids?
Recommended next step
Write a one-page blueprint for the role, test it on one office’s requests from last month, then roll it out to the other two.

A real summary reflects your conversation. Nothing on this page describes a real client.

One role, written down.

Before setup, the role, its access, human review and training are agreed in writing with the people who will work alongside it. For Harbor & Pine, the blueprint would read like this.

Role blueprint · Quote and follow-up coordinator

Harbor & Pine · fictional example

Drafts quotes and keeps every request moving. Sends nothing on its own.

The role
Drafts quotes from the shared price list and keeps track of every open request, in all three offices, until someone replies.
What it reads
New requests from the web form and each office’s shared inbox. The shared price list. The crew calendars, read only.
What it may do
Draft quotes and follow-up emails. Suggest visit times. Keep the list of open requests up to date. Flag anything unusual.
What it never does
Send anything to a customer. Change prices or offer discounts. Promise dates. Open payroll, bank or payment details.
Where a person decides
Every quote and message before it is sent. Any complaint, exception or discount. Any quote over $10,000 also goes to the operations director. Any change to what it can see or do.
How it is checked
Each week every office manager reviews a sample of drafts against the price list. A log records what it drafted and who approved it, and the operations director gets a monthly summary.
Who is responsible
The operations director owns the role; each office manager runs it day to day. Maya approves any change to its access. In a real engagement, deepmoat.ai sets it up and trains the teams.
How to stop it
An office manager can pause it for their office, and the operations director can pause it everywhere. Requests then go to the teams by hand, as before.

What a week looks like.

The collaborator prepares. People decide. Nothing leaves the company without a person.

  1. Monday

    Prepared

    Sorts more than 60 weekend requests by office, type and urgency, drafts quotes for the routine jobs, and flags four that the price list does not cover.

    Decided

    Each office manager approves their routine quotes and prices the flagged jobs themselves.

  2. Tuesday

    Prepared

    Finds 14 quotes from last week with no reply and drafts short follow-ups.

    Decided

    The office managers edit a few and send them all.

  3. Wednesday

    Prepared

    A customer writes in upset about a missed visit. The collaborator does not reply. It gathers the job history and passes it to the office manager.

    Decided

    The office manager calls the customer that morning.

  4. Thursday

    Prepared

    Suggests visit times for new jobs from the crew calendars, each marked as proposed.

    Decided

    The crew leads confirm most and move a few.

  5. Friday

    Prepared

    Prepares the weekly log for each office: what it drafted, what was approved and what was changed.

    Decided

    The operations director reads the summary and asks for one change to the quote template.

The first month, step by step.

An illustrative plan for this example. In a real engagement this is Capacity & Buildout, typically $25K–$75K, with timing and scope agreed in writing first.

  1. Week 1

    Agree the role.

    Write and sign off the blueprint. Agree one shared price list, the quote templates and a month of past requests.

  2. Week 2

    Test it on past work.

    Run it on last month’s requests in a test space. The team compares its drafts with what was actually sent.

  3. Week 3

    Use it, with approval on everything.

    Connect the first office’s inbox and web form. Every draft is reviewed by a person before anything goes out.

  4. Week 4

    Review it together.

    Look at the log and the team’s experience. Decide what to keep, change or stop, then train the other two offices.

What the AI collaborator does

Reads new requests, drafts quotes and follow-ups from the price list, keeps the open list current and flags what does not fit.

What stays with people

Prices, promises, complaints, every message that leaves the company, and the decision to keep using it.