🧠 Prompts < Lists

I used to start my Cursor projects dropping one huge prompt:

“Build a Flask app that scrapes a website and saves data to Postgres and X, Y and Z.”

The prompt usually had lots of tokens and was prone to not follow instructions properly.

I got better results when I started to break down the prompts to lists:

“1. Create a basic Flask server with one endpoint.

2. Add logic to fetch data from a URL.

3. Parse the data using BeautifulSoup.

4. Connect to a Postgres DB using SQLAlchemy.

5. Store the parsed data in a table.”

Now agents can work on each step individually and the outputs are usually much better.

On agentic architectures, lists are a much better way to steer the workflows in the right direction.