# Friday's decision (/claude-and-chatgpt-2/case-4-nishas-month-end/fridays-decision)

---
type: Document
title: "Friday's decision"
description: "The client's owner wants twenty packs a month, sent straight to the bank, for less money. Nisha must decide how far the setup may run alone, and what the work is worth now that preparing it is cheap. It is the same work a Forward Deployed Engineer does."
status: stable
order: 0.84
ksor:
  owner: team:panaversity
  audience: [ public ]
  approval:
    by: process:panaversity
    at: 2026-11-12T13:13:29Z
chapter: "F00G"
part: Front
expert_status: required
concepts: [ "Friday's decision" ]
last_verified: 2026-11-12
sources:
  - id: tag-routine
    title: "Anthropic, Set up routines (Claude Tag)"
    resource: https://claude.com/docs/claude-tag/users/proactivity
  - id: openai-projects
    title: "OpenAI, Projects in ChatGPT (OpenAI Help Center)"
    resource: https://help.openai.com/en/articles/10169521-projects-in-chatgpt
  - id: openai-scheduled
    title: "OpenAI, Scheduled tasks in ChatGPT (OpenAI Help Center)"
    resource: https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt
generated:
  at: 2026-11-12T13:13:29Z
  by: process:panaversity
trust_tier: unverified
build_id: sha256:ee289d75a60f3acc7dc4cab134056c3f9c1ab9c4abdba810061b7f3810eb4d15
dirty: true
ksor_version: 0.0.60
---

## The call

At 6 p.m. on Friday, October 8, Nisha took a call from **Walter Brennan**, who owned Lakeshore Supply. In Cleveland it was 9 a.m.

He was friendly and direct.

*"Dana showed me what you did this week. I own three more companies, all smaller than Lakeshore. I want the same pack for each of them. And my bank would like a short one every week, not only at month-end. Send those straight to Tom Becker, so that Dana does not have to forward them."*

Then he asked about the price.

*"Dana says your computer did it in half an hour. So I expect each one costs a good deal less than nine hundred dollars."*

He wanted an answer on Monday.

## Two questions inside one

After the call, Nisha wrote down what he had really asked.

1. **How far may it run alone?** Each of the four companies would get one month-end pack and four **weekly packs** a month: twenty packs in all. A weekly pack is like the weekly cash file Lakeshore gets today. Walter wants the sixteen weekly packs sent straight to the bank. Could the setup prepare and send them with no person in between?
2. **What is the work worth now?** The firm's price had been two days of labor. In a normal month the labor was now about two hours.

## How far may it run alone?

Hamza made the case for letting it send. *"Twenty approvals a month, and most weeks nothing changes. If we read twenty packs in a hurry, we are not really reading them. Let the setup send a pack when the status is READY and the figure is above the minimum. Then we can spend our reading on the ones that are not, and read a few of the others."*

It was a fair argument. Sana did not work Fridays, so Nisha phoned her.

*"Tuesday's pack was READY, above the minimum, and wrong, because the bank's file was not complete yet,"* said Sana. *"Hamza's rule would have sent it."*

She was fair about it. *"Nisha's approval sent it too. The approval did not catch the error. The morning check did. Then a person read the new pack before it went."*

*"That is what I want to keep. When something changes, a person reads it before the bank does. And the bank has a name it can ask. The setup has run three times in its life. Ask me again when it has run thirty."*

*"Then the morning check protects us, not the approval,"* said Hamza.

Nisha wrote down both points. Neither of them was wrong.

Nisha thought about what a routine is. It runs without anyone watching. It can do whatever the worker can do, including sending email.[^tag-routine]

She drew a line with three places on it. The same line is a table in [Exhibit 11](exhibits.md#exhibit-11-three-answers-for-the-owner).

![How far may it run alone? An arrow runs left to right: the setup does more, and a person reads less, and later. Three cards on the line. Prepare only: the setup may read, match, build and post ready for review; a person must check, approve and send; the firm today. Send if READY: the setup may also send when the status is READY and the figure is above the minimum; a person must review every NOT READY pack and a sample of the others; Hamza proposes. Send always: the setup may prepare and send every pack; a person must read what went out, afterward; Walter asks. A band at the foot says Tuesday's pack was READY, above the minimum, and wrong.](img/how-far-may-it-run-alone.png)

The firm's habit sat in the first place: nothing goes to a client without a named approval. The owner was asking for the third.

## What is the work worth now?

Nisha put numbers on three ways to charge ([Exhibit 11](exhibits.md#exhibit-11-three-answers-for-the-owner)). All figures are made up.

**1. Charge for the hours.** About two hours of people's time for each month-end pack: perhaps $110.00. A weekly pack is smaller: perhaps $40.00. For each company: $110.00 + 4 × $40.00 = $270.00 a month. It is simple, and the owner would agree at once. But if the firm gets faster, it works fewer hours, so it earns less. And nobody would pay for the time spent building the setup.

**2. Charge for the pack.** Keep today's prices for each company: $900.00 for the month-end pack and $1,500.00 a month for the weekly work. That is $2,400.00 a month for each company. The client pays for a checked pack with a name on it, not for hours. The firm would have to explain what the name is worth: the four tests, the review, the correction on Wednesday. Walter had already said he wanted a lower price.

**3. Charge for the service.** One monthly fee for each company, perhaps $1,200.00. It covers the month-end pack, the weekly packs, the morning check and any correction. The firm earns the same in a quiet month and a bad one. It must decide how many clients three people can be responsible for, even when the setup does the work.

Today Lakeshore alone pays the firm $2,400.00 a month. For all four companies, multiply by four: the three answers would bring in about $1,080.00, $9,600.00 or $4,800.00 a month.

![Three answers for the owner. Three cards. Card 1, charge for the hours: about 270 dollars a month for each company; it is simple and cheap; the firm earns less each time it gets better. Card 2, charge for the pack: 2,400 dollars a month for each company; the client pays for a checked result with a name on it; the firm must explain what the name is worth. Card 3, charge for the service: about 1,200 dollars a month for each company; it covers the packs, the morning check and corrections; the firm must decide how many clients it can answer for. A band at the foot says the setup changed the cost of the work, and it did not change who is responsible.](img/three-answers-for-the-owner.png)

## What the setup changed, and what it did not

Nisha counted. Four companies, a month-end pack and four weekly packs each: twenty packs a month. By hand, that was impossible for a firm of three. With the setup, the building was easy: one new project for each company, the same skill, the same kind of schedule.

The reading was not easy. Twenty packs a month meant twenty approvals. Sana worked four days a week. And what if three clients had a problem on the same Friday, when Sana was away? Someone would have to read each one before the bank did.

*"He is right that the work got cheap,"* said Nisha. *"He is wrong about which work. Preparing got cheap. Being responsible for it did not."*

## The same parts in ChatGPT

Dana's own team worked in ChatGPT. Nisha checked whether the answer would be different there. It would not. ChatGPT has the same parts under similar names: projects with instructions, skills, plugins for connected apps, scheduled tasks, and files through its desktop app, as [ChatGPT 2.0](../tour/chatgpt.md) and [Set up the apps](../new-way-of-working/set-up-the-apps.md) show.[^openai-projects],[^openai-scheduled] The questions on her paper did not depend on the product.

## Friday evening

It was seven in the evening in Karachi. Nisha had until Monday.

She had two decisions to make, and they were connected. If the setup did more alone, the price could go down. But an approval means less when nobody has read the pack.

**What should Nisha tell Walter?**

Use the [questions for discussion](questions-for-discussion.md). The case does not say what she chose.

## This is the work you are preparing for

A **Forward Deployed Engineer (FDE)** works inside a company to solve its business problems with AI Workers.

> **It is the same work a Forward Deployed Engineer does.** Look at Nisha's week. A client brought a **real problem**: a report for a bank, every month. There were **real limits**: the bank's minimum, a reviewer who works four days, a Team plan, a deadline in Cleveland. **People wanted different things**: Walter a lower price, Sana her name only on what she had read, Hamza less reading, Dana one pack she could trust. And Nisha had to make **a decision about what to build**, and how far it may run alone.
>
> Every company is a new case like this one. Later in this book, you learn to do this work. The cases prepare you for it, because **an FDE's job is a case that never ends.**

![The work of a Forward Deployed Engineer. Every company is a new case, and Nisha's week had all four parts. A real problem: a report for a bank, every month, that took two days by hand. Real limits: the bank's minimum, a reviewer who works four days, a Team plan, a deadline. People who want different things: Walter a lower price, Sana her name only on what she read, Dana one pack she can trust. A decision about what to build: which parts to join, how far the setup may run alone, and what it is worth. A band at the foot says an FDE's job is a case that never ends.](img/the-work-of-an-fde.png)

This is why this part of the book ends with cases. [How this book teaches](../overview.md#why-cases-fit-this-book-so-well) said it at the start. Each case is one practice run.

[^tag-routine]: As verified on November 10, 2026, Anthropic says a routine in Claude Tag "uses that channel's connections with the same permissions as a typed request." [Case 3](../case-3-nishas-firm/overview.md) describes the worker and its channel. Set up routines, Claude Tag docs, Anthropic.
[^openai-projects]: As verified on November 11, 2026, OpenAI says project sharing "is available on ChatGPT Free, Go, Plus, Pro, Business, Enterprise, and Edu", and that "Project instructions apply only within that project". Projects in ChatGPT, OpenAI.
[^openai-scheduled]: As verified on November 11, 2026, OpenAI says "Scheduled tasks can use supported apps, including Gmail, Slack, and GitHub, when those apps are available for your account or workspace." Scheduled tasks in ChatGPT, OpenAI.
