Run it yourself

DHUM / DATA 70600 ยท Class 2

Getting started

Make an empty folder and open it as a project in the app.

If you get stuck or see a question you do not understand, share your screen.

Three settings

MODEL

Which model does the work. Models differ in speed and capability.

THINKING

How much reasoning effort it uses. Higher settings can take longer.

APPROVALS

When it needs your permission. Some actions can run without asking.

Tell it how to talk to you

Before making changes, briefly explain your plan.
When you finish, explain what changed and which choices you made.
Explain technical terms the first time you use them.
Show me the commands you ran and their results.
Tell me when something fails and how you plan to handle it.
Ask before deleting a file or saving a version in project history.

Ask it: "Save these as my instructions for every project, not just this one. Tell me where you put them."

The same request on each computer

Build a small web app for New Yorkers to explore the most recent full week of NYC 311 service requests from NYC Open Data. Let people compare complaint types and boroughs. Show where the data came from and which dates it covers. Start it so we can open it in a browser.

Use last week's prompt exactly. The most recent full week has changed.

While it runs

What is it doing right now?

What did it decide that the prompt did not say?

What did it ask permission for?

Find these in your run

The folder

Where did it inspect your folder? What did it find?

New files

Which files did it create? Find them in your folder.

Commands

Find a command it ran and the result that came back.

A server

Did it start a server? Which address did you open?

Did it ask permission? What was it about to do?

Where the numbers come from

The agent writes a program that counts the city's records. We can follow a number back to those records and check the calculation.

The city's records reads The program the agent wrote keep last week's rows group them by borough count each draws The page shows the counts

Check the page against the data

Pick a borough. Does the count change?

Spot-check five records. Are their dates within the week you asked for?

Pick one count. Check it against the data file or the city's site, using the same dates and filters.

Narrow the browser window. Can you still read and use the page?

Ask it to explain its work

What did you build, and which choices did you make?

Which commands did you run, and what came back?

If you read code, ask it to show and explain the changes.

For a second opinion, start a new conversation. Ask it to find problems and show evidence for each one.

Compare the choices and results

Share your screen and show one choice your agent made.

How did it differ from another run? Show one thing you checked and what happened.

Ask for one change and test it. Keep the folder and conversation for next week.