Module 4 · Page 2 of 8 · 20 minutes

Set up your information environment

Agents work better when the information they need is organized and easy to find. A good project folder gives the agent consistent context and reduces the amount of searching and re-reading it has to do.

Recall the workflow

Module 1 introduced the loop behind most useful AI work:

1. Give context.
2. Specify the task.
3. Inspect the output.
4. Revise the request.
5. Verify the result.

When you work with an agent on a larger project, the first step — giving context — becomes especially important. Instead of repeatedly explaining the same project from scratch, you can organize the information so the agent can find and reuse it.

Start with a project folder

A simple way to create a stable information environment is to give each project its own folder. Put the files that belong to that project together, and use clear folder and file names.

For example, a market research project might look like this:

market-research-project/
├── data/
│   ├── customer-survey.csv
│   └── sales-by-region.xlsx
├── research/
│   ├── competitor-notes.pdf
│   └── industry-report.pdf
├── outputs/
└── project-guide.md

The exact structure doesn't matter. What matters is that a human — and the agent — can quickly understand where different kinds of information live.

Tell the agent what the files are

Once the files are organized, you can tell the agent what each file contains and which ones are relevant to the current task.

Example 1: Explain the files directly

Prompt

We are working on a market research project.

The data folder contains:
- customer-survey.csv: responses from our customer survey
- sales-by-region.xlsx: quarterly sales by geographic region

The research folder contains:
- competitor-notes.pdf: notes on three major competitors
- industry-report.pdf: an external report on industry trends

For this task, focus on customer-survey.csv and
competitor-notes.pdf. I want to identify the three most
important customer complaints and see whether competitors
appear to address them better than we do.

Example 2: Ask the agent to explore first

You don't always need to describe every file yourself. If the project folder is reasonably organized, you can ask the agent to inspect it and explain what it finds before starting the task.

Prompt

Before doing any analysis, explore this project folder.

Tell me:
1. What files and folders are here
2. What each one appears to contain
3. Which files are likely to be relevant for analyzing
   customer complaints

As projects grow, create a guide to the project

Small projects may only contain a few files. Larger projects can contain many folders, datasets, documents, scripts, and outputs. At that point, asking the agent to rediscover the entire project every time can be slow and wasteful. It can also reduce performance: when too much irrelevant information fills the agent's context window, important instructions or details can become harder for the model to identify and use reliably.

A useful solution is to create a short project guide: a document that explains the structure of the project and tells the agent where to look for different kinds of information.

Example: project-guide.md

# Project guide

Goal:
Analyze customer feedback and develop recommendations for
improving the customer experience.

Key folders:
- /data: raw survey and sales data
- /research: competitor and industry research
- /outputs: reports and charts created for the project

Important files:
- /data/customer-survey.csv: main source for customer feedback
- /data/sales-by-region.xlsx: use when comparing feedback with sales
- /research/competitor-notes.pdf: use for competitor comparisons

Notes:
- Do not modify files in /data
- Save new analysis outputs in /outputs

Now, instead of asking the agent to inspect the entire project from scratch, you can tell it:

Read project-guide.md first. Then use it to find the files
relevant to this task.

The guide acts like a map. The agent can understand the project structure first, then open only the files it actually needs.

Tip If you only need the agent to work with one or two files, you can also drag those files directly into the conversation and explain what you want done. The general principle is the same: make the relevant context easy to find.

Exercise

Create a project folder for a simple task (such as a homework assignment, or an extracurricular project). Then:

Put two or three relevant files into clearly named folders.
Ask the agent to explore the project and explain what it finds.
Create a short project-guide.md that explains the project structure. (Try asking the agent to do this for you.)
Start a new conversation and ask the agent to read the guide before completing a task.

Next · Page 3 of 8

Specify the task

→