Cursor is my new Google Docs

I use Cursor for writing and pretty much everything I do at work.

I use Cursor (the code editor) for pretty much everything I do lately, and not just for vibe coding.

Whether I'm drafting emails, internal documents / communications, LinkedIn posts, blog posts or prompts for other AI tools or tasks, I’m doing it in Cursor.

I still edit everything that it generates, but editing is much easier for me than creating from scratch. Cursor, in combination with a couple of mindset changes, are doing wonders for my productivity.

---

Disclaimer! This post might get old quickly, given how fast AI tools (and my workflow) evolves, but thought I'd share what I'm doing in the hope of learning more from others as well.

---

Before I go into specifics, these are the two key insights that have completely changed how I use AI tools like Cursor are the following:

1. Better context over perfect prompts

Instead of writing prompts as if I was explaining what I wanted to an 8 year old, I treat Cursor’s agent as a super smart collaborator: "this is what I want, this is all the background context you need to figure it out, go take a stab at it". The key to good results for me has been providing the best possible context and not to micro-manage the AI.

This allows me to simply brain dump some basic/disorganised thoughts about what I want to do along with the right context for the task, and let the agent figure it out. Then I edit the output, which again, is much easier for me than starting from a blank page.

For instance, if I'm writing about some new Tinybird feature, I'll pass Cursor the documentation and/or any internal document or prior post about it, and simply refer to it as if Cursor already knew about it. Turns out, it immediately does.

That, along with other Cursor niceties I talk about later, explains also why I'm using Cursor for writing:

everything I do becomes potential context for future tasks, and I can keep it all together, building out a body of work that I can constantly refer to.

2. Work in bulk

Rather than tackling tasks one by one, I've shifted to working in bulk through "agent mode".

In the mornings, I spend about 45-60 minutes doing a brain dump of all the tasks I want to accomplish during the day. I then let Cursor work through them in agent mode while I go take a break. Then I review, edit, and make the results mine.

---

These two things feel exponentially more productive for me. And it's getting better: the more context I generate across related tasks, the better the quality of each individual output feels.

How I use Cursor

I’ve got a folder called writing that looks as follows:

writing/
├── .cursor/
│   └── rules/         # Custom rules for different content types
├── context/           # Folder for random context (e.g. someone else's blog post)
├── linkedin/          # All my LinkedIn posts
├── newsletter/        # Newsletter drafts like this one
├── posts/             # Internal posts and longer-form content
├── prompts/           # AI prompts I've created
├── tweets/            # Twitter content
└── .cursorrules       # Global rules for the project

My .cursorrules File

My global rules file sets the tone and context for all my writing, including some examples of previous writing. This is how it looks like:

You are a helpful writing assistant to <your role here>. 
Your primary task is to help with writing in general, 
whether that is internal or external announcements for the company 
(taking into account the company's brand and tone), 
personal tweets, recommendation letters, essays or emails.

What follows is an example of a previous public blog post, use it as a reference for the tone and style of any writing that you do:

Title: $30M to lead the shift to real-time data

<content>
We are excited to share...

.cursorrules files were deprecated recently in favour of Project Rules which I’m already using for specific folders (see below)

Project Rules for specific tasks

I've also created specific rules for different types of content in the `.cursor/rules` folder:

• linkedin.mdc - Best practices for LinkedIn content. These apply to content in the linkedin folder only.

• tweets.mdc - Guidelines for Twitter threads. These apply to content in the tweets folder only.

• etc...

I generated some of these rules using ChatGPT Deep Research and then I asked Cursor to summarise the key takeaways when it comes to generating content and turn them into a prompt (see the example below).

These rules help Cursor understand the specific requirements for each task/content type without me having to repeat instructions every time.

Daily Workflow

Each morning, I do the following:

1. I start with a to-do list on a a new Cursor notepad everyday. I name each notepad with today’s date.

2. Under each task, I add specific notes about what I want and I gradually expand them with more context and details, referring to other files in the project as needed. You can do that using the '@' symbol to refer to files in the project.

At the top of the to-dolist, I use a prompt that instructs Cursor to break tasks into separate files. At some point, this will probably go to the project rules, but since I'm editing and experimenting every day, I'm simply passing it as the beginning of the prompt for now. This is how that prompt looks like today:

Everything that starts with "# Task:" is something that I want you to do independently in agent mode. It will usually be a writing assignment and you should go about it as follows:

* For each task, unless the content of the Task says otherwise (like correcting an existing one or creating new ones), create a new file.
* Tasks can be Post, Tweet, LinkedIn, Prompt, Newsletter or others. For each, there is a folder where you should create the file: if it's a "# Task: Post", create the file in the posts folder. If it's a "# Task: Tweet", create the file in the "tweets" folder, etc. If you don't find the appropriate folder, create it in the root of the project.
* Always write in Markdown format. 
* The files should have the .md extension. 
* Please, before you start with the tasks, figure out the current date (you can run 'date -I' in the terminal) so that you can use it to name the files with as follows: name them with the current date (YYYY-MM-DD) plus a title you choose for the task. For instance, if on March 6th, 2025, I ask you to write an internal post about restructuring the GTM team, the file could be named "2025-03-06-Restructuring-the-gtm-team.md" or something like that relevant to the content.
* Follow the applicable cursor rules for each project or folder

# Task: Prompt
Write a prompt and add it to new project rules file in .cursor/rules that:

- Applies to @linkedin folder and *.md files within it
- Summarizes all the best practices relevant to writing great content on linkedInd that are described in the file @linkedin_bestpractices.md

# Task: Post
I want to create a new version of the existing post @tinybird forward.md...

# Task: Post
Write an internal FAQ post about Forward vs Tinybird Classic...

# Task: Prompt
...

The tasks go into more detail of course, but the key here is providing great context.

3. When ready, I paste (CMD+SHIFT+V) the entire batch into a new Cursor composer window in “agent mode”. For some reason, it doesn’t pick up the files I reference very well when paste, so I usually have to re-add the files as context to the chat.

   

4. Cursor processes everything and creates draft files for each task

5. I review and refine each file, adding my personal touch and style. I still do this in Cursor, because:

   1. the "TAB" button works great to help me rewrite faster, and,

   2. I want to keep the final version in Cursor, so that I can use it as context for future tasks.

Again, this approach lets me focus on high-level thinking first and get to 80% done very quickly. Then I can edit throughout the day as I do other tasks knowing that that bulk of the work is done.

6. At the end of the day, or at the beginning of each day, I commit the project to git; I want to make sure I don't lose any work as this context is pure gold for future productivity.

An old example

## What else I have in mind

I'm constantly adding more and better context for Cursor to work with, so I'm thinking of how to incorporate more internal documents, internal wiki, etc. to the process. I sometimes wish everyone worked in a shared repo to have everyone’s context to use right away :-)

I keep waiting to start getting wonky results, but it hasn’t happened yet. Supposedly, the more unrelated tasks (and longer context) you give the LLM the worse it does, but I’m regularly passing four or five different tasks at once along with several referenced files and it is doing a hell of a job.

The next step is to start using Cursor to suggest tasks for me, given everything else I'm doing or given what's going on in the company. For instance, I'm doing more public writing these days about what we are building, or about what's going on in the company; it would be awesome if Cursor, given all the company context, would suggest topics I could write about and write a first draft for me.

Anyways, I'm excited to keep experimenting with this and see where it goes -this workflow has transformed my productivity and the quality of what I produce thanks to:

1. Better context

2. The LLM is applying my style across multiple related tasks

3. I can focus my energy on refining rather than creating from scratch

I'd love to hear about how you use AI tools to boost your productivity.