Case 3: Nisha's Firm

What changed when a company of one hired two people and gave its AI Worker a name of its own

Panaversity case study · November 2026

Where this case fits

This is the seventh section of Claude and ChatGPT 2.0, and its third case study. Case 1 showed a helper: Nisha managed two AI tools while she sat in front of them. Case 2 showed a personal AI Worker: it ran while she slept, under her name, with every key in her hands. This case climbs one more step, to a team AI Worker. It has its own name and its own accounts, and three people can hand it work.

Case 2 ended on a question it could not answer. If the worker emailed the client, the email would come from Nisha's address, and the client could not tell what she had read. This case answers that question. Then it shows the new problem the answer creates.

• A worker with its own badge. Setting up the desk: keys, buildings and badges introduced a worker that acts under its own accounts. Here you see one at work.
• Saving work somewhere lasting. Why the borrowed desk gets cleared showed that the desk can be cleared. Watch where the firm keeps each week's file.
• The second signature. Staying safe showed that a person makes the final decision. In a firm of three, the question becomes which person, and how the worker can tell.
• Words are not locks. Case 2 separated what a setting enforces from what a rule only asks. Here, an instruction is followed word for word and still fails, and a rule holds only because the worker reads it well.
• Agent skills. The Tour introduced skills as saved instructions that teach the AI how to do one kind of job your way. Here, an expert writes her method down as a skill, and the firm learns what a skill can fix and what it cannot.
• Identity and the business layer. From personal worker to business worker listed what a business adds around the worker. On Monday, Nisha has to decide how much of it she needs.

About this case

This is a fictional teaching case, prepared for class discussion.

Made up for teaching: Nisha, Hamza, Sana, Lakeshore Supply, its controller and its bank, every figure, every cost, the flight, the people's messages and everyone's thoughts. No real company or client data is used.

Checked against the documentation: how Claude Tag handles identity, shared credentials, channels, instructions, routines and logs, and what OpenAI says about its workspace agents in Slack. Each fact has a footnote with the date it was checked. Claude Tag was in public beta, so these facts can change.

From a chat test, not yet from the product: The worker's Friday file and its answers come from a chat test in November 2026, in which a small Claude model stood in for the worker. Case 3's practical on Claude Tag has not yet been run. The exhibits that would come from running it (the Friday thread, the three records and who caught what) are each labeled. The practical run sheet says exactly how to run and record the exercise.¹

One thing the documentation does not say. It does not say whether Claude asks for a confirmation before it sends an email. In our chat test the worker refused to send without an approval. See the note on The Friday send.

Not accounting advice. A bank reconciliation is a standard task, but the one in this case is a short teaching example. It needs a qualified accountant's check before anyone relies on it.

A discussion case, not a validated demonstration. Read what the worker did as what one model did in a chat test, not as what the product always does.

The case follows the style of a Harvard Business School case but is not connected to Harvard. It does not say whether anyone's choices were right or wrong.

Doing the lab? Here is the order. Read the first three story pages: A firm of three, A worker with its own name and Two routines and one rule. Then do the lab. Then read the last three story pages and the exhibits. In the lab you take Hamza's place on Friday, so reading the later pages first spoils it.

Words from earlier sections. You will meet these again. Each one is explained where it first appears.

• AI Worker: an AI system you hand a whole job to, inside limits you set.
• The borrowed desk: the temporary computer where the AI does its work. It is cleared afterward, so anything worth keeping must be saved somewhere else.
• Connector: a link that lets the AI use another tool, such as a mailbox or a shared drive.
• The second signature: a person's final yes before something with real effects happens, such as sending a file to a client.

How this case is organized

The story comes first, in six short parts. Read them in order.

1. A firm of three: the staff, the client, the weekly job, and why Nisha's project could not come with her.
2. A worker with its own name: the worker joins the firm's Slack, with its own accounts and a key nobody can see.
3. Two routines and one rule: the weekly schedule, the firm's rule about sending, and four things the setup did not say.
4. The Friday send: an approval on Wednesday, a flight on Friday, a file that does not leave the firm, and what Sana finds on Monday.
5. Sana's skill: Sana writes her method down as an agent skill, and the firm learns what a skill can fix.
6. Monday's decision: the client asks who can release a file when Nisha is away. Nisha must decide how files leave the firm from now on.

After the story, The case at a glance sums it up in ten sentences. Then come the questions for discussion and the exhibits. For instructors, the teaching note gives a discussion plan. The lab needs no team plan: you map the channel, decide as Hamza, check the file with and without an agent skill, and test whether a sentence passes as an approval. The practical run sheet says how to run the exercise and record it. Last, the final assessment gives you a new team, a new incident and a gap to find.