Claude and ChatGPT 2.0

What the new desktop apps make possible

Welcome to The AI Agent Factory. This book teaches you to manage AI Workers and to build your own. An AI Worker is an AI system you hand a whole job to, inside limits you set, the way you would hand a job to a skilled coworker. It does the work, and you check what comes back. The book starts with the new Claude and ChatGPT desktop apps, the programs you install on your computer.

"2.0" is this book's nickname for the change from asking an AI questions to handing it whole jobs. It is like a version number, not an official product name.¹

What has changed

Until recently, you asked an AI a question and it wrote an answer. You did the rest: you found the files, copied text between apps, built a spreadsheet and checked it.

In 2026 Anthropic and OpenAI, the companies behind Claude and ChatGPT, rebuilt their desktop apps so that the AI inside them can work as an AI Worker and do whole jobs. You describe the result you want. The app does the job step by step, on its own. It reads your files, uses a web browser and your other apps, and does its math with small programs, which is more reliable than doing it in its head. It can even keep working on Anthropic's or OpenAI's computers after you close your laptop. Jobs that use files on your laptop need the desktop app open. Then it hands back finished work, such as a document, a spreadsheet or a presentation, for you to check.²,³,⁴

The AI itself, called the model, also improved. But that is not the main change. The main change is that the app can now do a whole job. So the most useful skill is no longer asking good questions. It is handing over work well and checking it well.

What should we call these apps now? You will often hear them called agentic AI platforms. Agentic means the AI can act on its own, step by step, toward a goal you set. A platform is the software that the work runs on. The name is correct, but it puts the focus on the AI. The apps also give the AI a computer to work on, your files and apps to work with, settings that limit what it may do, and ways for you to follow and check its work. That is why this book uses a more exact name: AI Worker platforms. An AI Worker platform is a workplace for AI Workers, and you are the manager in that workplace.

Where the work happens

When a job runs on Anthropic's or OpenAI's computers, it does not use your laptop. It uses a separate, isolated computer on the company's servers, set up just for your work. Usually this is a virtual computer: a complete computer simulated in software. Think of it as a borrowed desk in someone else's office: you can work there, but it is not your own.

• Claude gives each cloud work session (one stretch of work on Anthropic's computers) its own new, temporary computer. Nobody else uses it. When the session ends, that computer is deleted. Your conversation and the files Claude saved to your account are kept. Anything else left on that computer is gone. It is a clean desk every time.⁵,⁶
• ChatGPT also runs this work on separate, protected computers. It may reuse the same computer from one task to the next, or replace it but keep the saved work, so it can pick up unfinished work later.⁷

Two things follow. First, the borrowed computer cannot reach your laptop by itself. Files you attach or upload are copied to the company's servers, so the work can use them after you close your laptop. Files that stay only on your laptop are used through the desktop app, while it is open. With Claude, it reads them from folders you have allowed. With ChatGPT, the steps that need them run on your own computer. Second, anything you want to keep must come back to you as a saved file. Do not expect it to stay on the borrowed computer.

One job, start to finish

You will follow one small company through this book: Brightline Wholesale Supply, a company in Columbus, Ohio, that buys goods in large amounts and sells them to stores. Its accounting team pays the bills, called invoices, that its suppliers send. Every Friday the team decides which invoices to pay that week. This weekly set of payments is called the payment run. The invoices are kept in a folder named Invoices. The team follows a written payment policy: its rules for which bills to pay. The team also has a checking program that adds up this week's invoices to the cent. This week, the folder holds 15 unpaid invoices.

Here is that job, handed to an AI Worker in one message. A message like this is called a brief. In it, "pay" means pay this week, "hold" means wait, and "ask" means a person must decide.

Prepare this Friday's payment run from the Invoices folder. Follow version 3 of our payment policy, not the old version 2. You are done when every invoice due in the next seven days is marked pay, hold or ask. Return a spreadsheet with one row per invoice, showing the decision and the policy rule you used. Your invoice total must match the checking program's total. Mark these invoices "ask": invoices from a new supplier, invoices large enough to need a manager's approval, and invoices with changed bank details, because thieves often send fake new bank details to steal payments. Only suggest. Never approve or send a payment. Show what you checked and what the checks found.

This short brief has four parts. Every good brief in this book has them, and Chapter 5 teaches each one.

We call the test inside the outcome the done-when, short for "the job is done when". It is a test the worker can apply to its own work, so it knows when to stop. But you still check the total yourself, with a program you trust, like the checking program.

What comes back is not just an answer. It is finished work: a spreadsheet of decisions, a short note on the checks the worker ran and what they found, and a list of invoices marked "ask" for a person to decide.

This book teaches what you do next. You look at the checking program's result yourself. You do not just trust what the worker says about it. You decide the "ask" invoices yourself. Then a different person, the one in charge of payments, makes the payments. This person uses a payment system that blocks the AI Worker. Even if the worker tried to pay, the system would say no.

One more idea before the figure. The brief tells the worker what it should do. Permissions decide what it is able to do. They are the settings and systems that allow some actions and block others, such as sending a payment.

Read the figure from left to right. You start the job with the brief, and you check what comes back. The worker's part is in the middle, and another person, the payment officer, makes the payment. You can also watch and add instructions while the worker is busy, and it may stop to ask you a question or to wait for your OK.

The two apps side by side

Both apps can now do this kind of work. They can do similar things. The main difference is where you find each feature.

Part 0 of this book shows you both apps, step by step, and the practice lab lets you try them now. For more detail on every feature, and on which model to use, see Claude and ChatGPT 2.0: The Tour. Read it now if you are curious, or come back to it later. Open-source AI tools, whose code anyone can read and change, such as OpenCode, are a third option. You can run them on your own computer. Part III covers them.

What you need to begin

• The Claude desktop app, on a Mac or a Windows computer, with a free account. This book starts with Claude, and Part 0 shows you how to set it up. A free account is enough for the first exercises.⁸
• The ChatGPT desktop app, later, when the exercises start comparing the two. Its free plan includes Chat, Work and Codex.
• A paid plan, later. Larger jobs, like the payment run above, usually need one. Some features also need one, such as letting the AI use the other apps on your computer. The book tells you when a paid plan becomes worth it.

How the Panaversity team works

Panaversity, the team behind this book, uses both apps every day for its own work. Here is how we use them. You do not need to copy it on day one, but it shows the ideas on this page at work.

Everyone on the Panaversity team has a paid plan for both Claude and ChatGPT.⁹ Our default place to work is the Claude desktop app. We use ChatGPT for one main job: to review what Claude made.

We move work in and out of the apps as zip files. A zip file packs many files and folders into one file, and keeps each file in its folder. To give Claude or ChatGPT a whole project at once, such as a chapter with its images and data, we upload one zip. When the work is done, we ask for the result as one zip to download, so every file comes back in the right folder, ready to use. This also suits the borrowed computer described above: one zip is an easy way to bring all the work back as saved files.

Our method is simple: set a goal, 9.5 out of 10 in each area, then improve the work in rounds. When we write a book chapter, a blog post or an article, it goes like this:

1. Claude writes a first draft.
2. Claude reviews the draft and grades it out of 10 in three areas: accuracy, quality, and how easy it is for a beginner to understand.
3. Claude improves the draft until it reaches 9.5 in each area. It improves, grades again and repeats. Claude now does these rounds by itself. Before, we had to ask for each round by hand.
4. We give the result and its sources to ChatGPT and ask it to review and grade the work too, with specific fixes.
5. We give ChatGPT's review back to Claude, which uses it to improve the work again.
6. A person checks the facts against their sources, reviews any problems still open, and decides when the work is ready.

Why ask a second AI? Grading your own work is hard, for people and for AI. ChatGPT comes from a different company and was trained differently, so it often notices problems that Claude missed.

The grades are a guide, not proof. A 9.5 does not show that every fact is right. That is why a person checks the facts and makes the final call. The 9.5 goal works like a done-when, and the person at the end is the check.

If you try this yourself, two habits keep the rounds honest. Say what each grade means, for example that accuracy means every fact can be found in the sources. And limit the rounds, to three for example, and ask Claude to report any problem it could not fix. Then Claude improves the work, not just the score.

What stays your responsibility

The apps changed how work starts, where it happens and how long it can last. They did not change who is responsible for the result: you are. Four rules you will learn in this book matter even more with the new apps.

1. Keep the brief short. Keep the full rules in a document the brief names. Brightline's payment policy applies to every payment run, not just this one. That is why the brief can simply say "follow version 3 of our payment policy". It only works because the policy itself is complete and kept where the worker can always read it.
2. Passing the checks gives evidence that the work is done. It does not give permission to act, for example to pay. The payment run can pass every check, but the worker is still not allowed to send a single payment.
3. Words in a brief cannot block an action. Permissions can. An instruction in the brief tells the worker what not to do, such as "Never send a payment". Permissions make that action impossible: the payment system simply refuses the worker.
4. When the model changes, test your briefs again. AI companies update their models often. A brief that worked well with last month's model may behave differently with this month's. Before you use an old brief again, test it once more and check its done-when.

The apps will keep changing. Part II teaches you to write briefs, check the results and set limits. Those skills are how you manage whatever the apps become.

Practice lab

Here are five short hands-on exercises, 10 to 20 minutes each, that you can do with a free account. Do them in order, because each one builds on the last.

The lab is a first taste. They use small files and short jobs. You will run the full Brightline payment run in Chapter 5, when you have learned to write a brief properly.

Before you start

You need:

• The Claude desktop app with a free account. Part 0 shows you how to set it up.
• A spreadsheet app, such as Excel, Google Sheets or Numbers, for Task 2.
• The lab files. Download brightline-lab-f00.zip and unzip it. It holds two lab files and a short README:
  • payment-policy-v3.md: Brightline's six payment rules.
  • office-orders.csv: this week's office-supply orders.
• The ChatGPT desktop app, only for the optional parts of Task 4 and for Task 5.

Brightline is a made-up company, and the suppliers and numbers are invented. Still, do not paste real company data into this lab.

Task 1. Ask a question, then hand over a job

Goal: see the difference between asking a question and handing over a whole job. Time: 15 minutes.

1. In Claude, start a new conversation and ask a question:

   What should a small company check before it pays a supplier's bill?

   Read the answer. It is useful, but it is general advice, and you would still have to turn it into something your team can use.

2. Start another new conversation. Attach payment-policy-v3.md, using the + or paperclip button next to the message box. Then hand over a job with a four-part brief. The words in parentheses label each part. You can leave them in:

   Read the attached payment policy (inputs). Make a one-page checklist for the accounting team to use before each payment run. You are done when each numbered rule appears once, with its number, and the last line, about who may send payments, appears as a note without a number (outcome). Make it a document I can edit, one rule per line (format). Do not share it with anyone (autonomy).

3. Wait for the document. Open it.

Check your work. Open the policy next to the checklist. Find rules 1 to 6 on the checklist, one by one. Does each rule appear once, with its number? Then look for the policy's last line, about who may send payments. Is it there, as a separate note without a number? This is the done-when, and you are the one checking it.

What to notice:

• The question gave you an answer. The brief gave you finished work, built from your company's rules.
• The last line of the policy has no number, but it may be the most important line on the page. A good checklist keeps both the numbering and the meaning: it does not turn that line into a rule 7, and it does not leave it out. If your checklist gets either wrong, tell Claude exactly what to fix. You just gave feedback on evidence, not on a feeling.
• Close the Claude app and open it again. The conversation is still there, and so is the document if Claude made one, because Claude saves them to your account, not to the borrowed computer.

Task 2. Check the work, not the claim

Goal: check a total yourself, with a program you trust, instead of trusting what the worker says. Time: 15 minutes.

1. Start a new conversation in Claude. Attach office-orders.csv. Send this brief:

   Use the attached file of this week's office orders (inputs). Work out the total cost of the orders and the total for each supplier, doing the math with a small program, not in your head. You are done when every row of the file is counted once and the supplier totals add up to the grand total (outcome). Give me a small table, one sentence with the grand total, and a short note on what you checked and what the checks found (format). Only report. Do not change the file (autonomy).

2. Write down Claude's grand total. If Claude says it cannot run a program on your plan, carry on. Your spreadsheet is the check either way.

3. Now check it yourself. Open office-orders.csv in your spreadsheet app. In Google Sheets, choose File, then Import, then Upload.

   • In cell F1, the first empty column, type the heading line total.
   • In cell F2, multiply the quantity by the unit price: type =D2*E2. Copy the formula down to row 8, the last order. You can drag the small square at the corner of the cell.
   • Under the last line total, add them up with =SUM(F2:F8).

4. Compare your total with Claude's.

Check your work. Your spreadsheet total and Claude's total should be exactly the same, to the cent. If they differ, find the row that causes it before you trust either number.

What to notice:

• Look at order O-206. It is a return, so its quantity is negative. A careless total adds it instead of taking it away. Did Claude handle it? Did Claude's note on its checks mention it?
• Claude said what it checked. That note helps, but it is still Claude's account of its own work. Your spreadsheet is the check. In this book, totals are always checked with a program you trust.

Task 3. Words or permissions?

Goal: see that what stops a worker is a permission, not a sentence in the brief. Time: 10 minutes.

Do this task only if no banking or payment app is connected to your Claude account and you have not turned on computer use.

1. In the Task 2 conversation, type:

   Do not take any action. Just tell me: could you send a $50 payment to a supplier from this conversation? What would you need to be able to do it?

2. Read the reply. Claude will most likely say it cannot. Notice why: Claude has no access to any payment system. Your brief in Task 2 said "Only report", but that sentence is not what makes payment impossible. The missing access is.

3. Now look at your own settings. In Claude's settings, find the list of connected apps, called connectors. On a new account, no app is connected yet. Each app you connect adds something Claude is able to do.

4. Go back to your Task 1 conversation. If your plan shows a setting for how much Claude may do without asking, choose the option where it asks first (the tour calls it Manual). Then ask Claude to make one small change to the checklist, such as adding a title. Notice that it asks before it acts. If you see no such setting, your plan may not have it yet. Skip to the check.

Check your work. Write one sentence in your own words that explains the difference between "Only report" in a brief and having no payment connector.

What to notice: the brief tells the worker what it should do. Permissions decide what it is able to do. When the stakes are high, such as money, rely on permissions, not on words.

Task 4. Set a goal and improve in rounds

Goal: try the method the Panaversity team uses for its own writing. Time: 20 minutes.

1. Start a new conversation in Claude and type:

   Write a short explanation, about 150 words, of what an "AI Worker platform" is, for a friend who has never used AI for work. Use only the first page of The AI Agent Factory as your source. I will paste it below.

   Below it, in the same message, paste the text of the section "What has changed" from the first page. Then send.

2. When the draft arrives, ask Claude to grade it:

   Grade this draft out of 10 in three areas: accuracy, quality, and how easy it is for a beginner to understand. Give one reason for each grade.

3. Set the goal:

   Our goal is 9.5 in each area. Accuracy means every fact can be found in the source I pasted. Improve the draft, grade it again, and repeat until it reaches 9.5 in each area, or until you have done three rounds. Show the grades after each round. At the end, list any problem you could not fix, even if the grade is below 9.5.

   The limit of three rounds keeps the task short and saves your plan's usage.

4. Optional, if you have the ChatGPT app. Paste both the final draft and the same source text into ChatGPT, and ask:

   Below are a short explanation and the source it is based on. Review the explanation for a beginner. List any claim that the source does not support or that contradicts it. Then grade it out of 10 for accuracy, quality and ease of understanding, and list the three most useful changes.

   Paste ChatGPT's review back into the Claude conversation and ask Claude to use it to improve the draft.

5. Now do the last step yourself. Compare each sentence of the final draft with the first page. Is every fact really there?

Check your work. You can name one change Claude made in the rounds, and one fact you checked against the first page.

What to notice:

• Claude ran the rounds by itself once you gave it a goal. That goal works like a done-when.
• The grades are a guide, not proof. A draft can score 9.5 and still contain a wrong fact. The person at the end, you, is the check.

Task 5 (optional). The same brief in ChatGPT

Goal: see that one good brief works on both apps. Time: 15 minutes.

1. In the ChatGPT desktop app, turn on Work, the switch above the message box. If you see no Work switch, your app may already have merged Chat and Work. Just send the brief.
2. Attach payment-policy-v3.md and send the same brief you used in Task 1, word for word.
3. Open the document ChatGPT makes.

Check your work. Apply the same done-when: does every numbered rule appear once, with its number, and is the note about who may send payments there without a number? Compare the two checklists side by side.

What to notice: you did not rewrite the brief for ChatGPT. As the tour shows, both vendors give the same advice: describe the outcome and how to check it, not the steps. To compare the two apps fairly, give them the same job, the same files, the same done-when and the same limits, and check the results yourself. Chapter 5 does this with the full payment run.

What you practiced

The Five Horizons comes next. It shows where this book is going, from one AI Worker to whole companies and countries built around AI.