This is a very interesting concept: using a programming tool to manage non-programming tasks. Of course, some people use Git for that sort of thing.
Personally, I'm not interested in Cursor or Windsurf - or anything that costs me money on a monthly subscription basis.
What I find more reasonable, given the low cost of API access to these tools, is to spread a few $10-20 API accounts for the main services, then access them using GUI front ends like MSTY or LMStudio.
DeepSeek, for example, costs very little. I put $10 in my API account in January. I ask it questions from MSTY almost every day. When I last checked my account a couple of weeks ago, it still had $9.97 in it. I'd used 3 cents in almost three months.
The latest Google Gemini is also incredibly cheap, not to mention the incredibly useful Google NotebookLM.
But the idea of having one AI tool that can manage everything is sort of my Holy Grail. Right now, I use the Obsidian note-taking app with a couple embedded AI tools like SystemSculpt, and outside that I use MSTY. I work with the AI in MSTY (or online), then copy what I get into Obsidian. MSTY can do RAG with an Obsidian vault, but I haven't set up a workflow for that yet.
I turned to Cursor after struggling to generate code on common AI platforms without success - the code I provided was too long for any free tools to handle.
As a beginner coder, I found Cursor incredibly helpful. It delivered the correct, edited code based on the changes I requested. I use it often for generating code for web design and highly recommend it!
I have been using Cursor/Windsurf mostly for coding. While I read similar articles about using it for non-technical work, I have not made the leap. I think one challenge is that most of my files are not in markdown, which makes working on them in Cursor harder.
Curious, what is the most important use case for you that you cannot stop doing now that you have done it in Cursor?
I’m a big fan of using it for drafts. Having my entire library of written works available to it for referencing + its ability to search the web is invaluable. It’s truly a second brain
I agree that Cursor is fun to write with. It has become indispensable as writing environment for me as well, especially the diff view. But it feels like a hack because it isn't designed towards this.
Since you mentioned open-endedness in your piece, my take on ”the modern AI workspace“ is this: I think it’s worth stopping to frame everything from a productivity and goal-oriented perspective and rather think about it like this: AI workspaces should be about sparking curiosity and enabling new ways of exploring, connecting ideas and ultimately thinking, not just optimizing existing workflows.
I mainly love writing with AI because it extends and complements my perspective, not because it makes me more efficient (actually I spend _more_ time writing because I’m exposed to a broader variety of thoughts). Anyway, here comes the plug – I wrote a bit about this here:
This is a good point. I'm pretty technical, so I often wonder if there's an opportunity for a Cursor-lite application for those that just want to use it for things like writing or simple file management.
Jake, thanks for sharing all this. I've been a heavy Claude pro user but after seeing this post, I'll try Cursor. Two questions...
1. I'm wary of giving Cursor Agent authority to modify my (entire) local filesystem ("Move all my research PDFs into a references folder and rename them with today's date."). Is there a way to restrict its write permissions to only certain directory trees? Then I might create a top-level "Cursor/" directory into which I move the things I want to operate on (and leave the rest of my filesystem "safe").
2. How much of my local documents (that I have it operate on) would remain private & local? e.g. if I tell it to "batch edit documents...", do those documents first get transmitted to Cursor's central server (and/or LLM / MCP service like Claude or whatever) for processing, and thus perhaps remain online part of some cloud-hosted chat log?
1 - Yep, just load up a folder like you were starting a new repo. This will restrict the agent to operate only within that folder (and you can add as many subfolders within that as you want)
2 - Turn "Privacy Mode" on in Cursor and they (and any third parties) will never collect your usage and general data. Outside of that, they use Claude/OpenAI/etc APIs so the privacy policy will be whatever is for each corp. Typically, the APIs for LLMs are more secure and default to ditching input data after it's processed
Re: 1 - That's something I have been looking into. Windsurf can actually access everything on my computer and can read and create files outside of the folder (even if I turn off access via my Mac's system settings). After reading your comment, I went to try the same prompt with Cursor and found that Cursor suggested running a code to list files or create a file, instead of doing it automatically for me, which feels a bit more secure.
So technically, Cursor can still operate outside of the folder but it is programmed to ask for permission first.
This was a great idea - I got some really good results. However, Cursor seems to have a 2Mb file limit which is very low if you are using a lot of PDF files for some of the research/source data to help generate stuff.
Definitely an unfortunate current restriction. A (albeit hacky) workaround is asking the agent to write a script that splits the file into chunks under 2MB, and then process it that way.
But I would imagine at some point they’ll offer a way to up that file size limit.
Glad you’re getting good use out of an AI workspace!
After writing an article about using the coding editor Zed for non-coding tasks:
https://news.aidful.net/i/149508805/zed-from-coding-editor-to-universal-ai-assistant
I also explored the use of Cursor and Windsurf. These are in many AI integration aspects superior to Zed.
Very inspiring article which triggerd many thoughts I want to implement.
Is there a way to let the editor perform tasks periodically?
This is a very interesting concept: using a programming tool to manage non-programming tasks. Of course, some people use Git for that sort of thing.
Personally, I'm not interested in Cursor or Windsurf - or anything that costs me money on a monthly subscription basis.
What I find more reasonable, given the low cost of API access to these tools, is to spread a few $10-20 API accounts for the main services, then access them using GUI front ends like MSTY or LMStudio.
DeepSeek, for example, costs very little. I put $10 in my API account in January. I ask it questions from MSTY almost every day. When I last checked my account a couple of weeks ago, it still had $9.97 in it. I'd used 3 cents in almost three months.
The latest Google Gemini is also incredibly cheap, not to mention the incredibly useful Google NotebookLM.
But the idea of having one AI tool that can manage everything is sort of my Holy Grail. Right now, I use the Obsidian note-taking app with a couple embedded AI tools like SystemSculpt, and outside that I use MSTY. I work with the AI in MSTY (or online), then copy what I get into Obsidian. MSTY can do RAG with an Obsidian vault, but I haven't set up a workflow for that yet.
I love it!
Thank you for the insightful read and great tips!
I turned to Cursor after struggling to generate code on common AI platforms without success - the code I provided was too long for any free tools to handle.
As a beginner coder, I found Cursor incredibly helpful. It delivered the correct, edited code based on the changes I requested. I use it often for generating code for web design and highly recommend it!
Thanks for writing this, Jake!
I have been using Cursor/Windsurf mostly for coding. While I read similar articles about using it for non-technical work, I have not made the leap. I think one challenge is that most of my files are not in markdown, which makes working on them in Cursor harder.
Curious, what is the most important use case for you that you cannot stop doing now that you have done it in Cursor?
I’m a big fan of using it for drafts. Having my entire library of written works available to it for referencing + its ability to search the web is invaluable. It’s truly a second brain
I assume you write in markdown? Thanks for sharing!
I do! But it’s taken a bit to get used to.
Worst case scenario just write in txt and then ask Cursor to convert it to md
Wow I wonder why I didn't think Cursor could do this too. Great post!
I agree that Cursor is fun to write with. It has become indispensable as writing environment for me as well, especially the diff view. But it feels like a hack because it isn't designed towards this.
Since you mentioned open-endedness in your piece, my take on ”the modern AI workspace“ is this: I think it’s worth stopping to frame everything from a productivity and goal-oriented perspective and rather think about it like this: AI workspaces should be about sparking curiosity and enabling new ways of exploring, connecting ideas and ultimately thinking, not just optimizing existing workflows.
I mainly love writing with AI because it extends and complements my perspective, not because it makes me more efficient (actually I spend _more_ time writing because I’m exposed to a broader variety of thoughts). Anyway, here comes the plug – I wrote a bit about this here:
https://j0lian.substack.com/p/auto-associative-workspaces?r=68k57
This is a good point. I'm pretty technical, so I often wonder if there's an opportunity for a Cursor-lite application for those that just want to use it for things like writing or simple file management.
Jake, thanks for sharing all this. I've been a heavy Claude pro user but after seeing this post, I'll try Cursor. Two questions...
1. I'm wary of giving Cursor Agent authority to modify my (entire) local filesystem ("Move all my research PDFs into a references folder and rename them with today's date."). Is there a way to restrict its write permissions to only certain directory trees? Then I might create a top-level "Cursor/" directory into which I move the things I want to operate on (and leave the rest of my filesystem "safe").
2. How much of my local documents (that I have it operate on) would remain private & local? e.g. if I tell it to "batch edit documents...", do those documents first get transmitted to Cursor's central server (and/or LLM / MCP service like Claude or whatever) for processing, and thus perhaps remain online part of some cloud-hosted chat log?
Hey Scott! Thanks for commenting.
1 - Yep, just load up a folder like you were starting a new repo. This will restrict the agent to operate only within that folder (and you can add as many subfolders within that as you want)
2 - Turn "Privacy Mode" on in Cursor and they (and any third parties) will never collect your usage and general data. Outside of that, they use Claude/OpenAI/etc APIs so the privacy policy will be whatever is for each corp. Typically, the APIs for LLMs are more secure and default to ditching input data after it's processed
Re: 1 - That's something I have been looking into. Windsurf can actually access everything on my computer and can read and create files outside of the folder (even if I turn off access via my Mac's system settings). After reading your comment, I went to try the same prompt with Cursor and found that Cursor suggested running a code to list files or create a file, instead of doing it automatically for me, which feels a bit more secure.
So technically, Cursor can still operate outside of the folder but it is programmed to ask for permission first.
Excellent! Thanks.
This was a great idea - I got some really good results. However, Cursor seems to have a 2Mb file limit which is very low if you are using a lot of PDF files for some of the research/source data to help generate stuff.
Definitely an unfortunate current restriction. A (albeit hacky) workaround is asking the agent to write a script that splits the file into chunks under 2MB, and then process it that way.
But I would imagine at some point they’ll offer a way to up that file size limit.
Glad you’re getting good use out of an AI workspace!
great read. would love to collab as well.
Yep cool take I dig