One of the things that you will see on many articles about how people use AI is either something called “the second brain” or just connecting your email and calendar and using AI to track all your tasks.

My problem with the majority of these, including the more recent personal agents like Muse, is that you are giving away not just your own personal private information, but also leaving an always-on connection to information from the people that you're interacting with.

In information security circles, this is called the “surface area” of information that you are willing to expose to possible security issues.

Now, it’s true that there isn't a whole bunch of news about prompt injection, or other security hacks for personal accounts. There are enough news articles about agents escaping testing environments and people's personal agents giving far more information than they realized they were going to give. There are many who gave access to everything to agents like Muse and came to regret it later.

❝

By the way, prompt injection is when someone sends a prompt through, say, a calendar invite or an email that otherwise looks safe, but might have in white text that you can't see, something that tells the AI that you're using that is connected to your systems to send along passwords or send along connections to other systems on your computer.

I used to have a Claude system that was connected to my business calendar and email inbox until reading about the data security threats. And considering my business calendar was tied to my personal calendar which has a bunch of family information, I decided to disconnect everything which left me without a task tracking system for a while.

The tracking system I use

Being a neurodiverse business owner, I found myself very quickly needing some sort of tracking system. Thankfully, I found one that was simple - SimpleBrain. (Link takes you to Github, where the files are stored that I copied for this system.)

The premise is simple. Here are the main folders:

  • raw: in here go files with notes from meetings, brain dumps, PDF prints of articles I want to save…anything that I want to capture that might have tasks or information I want to refer to later.

  • wiki: the translate.md file included in the “prompts” folder tells the AI to read everything in the “raw” folder and parse it out into “wiki” and “projects”

  • projects: this holds subfolders, one per project, each with a Readme file that explains what the project is, background, any needed information about the project so that when data is put into “raw,” the system can add relevant information to the subfolder.

  • archive: everything from “raw” gets moved here after the AI processes it.

  • prompts: contains the instructions for AI to use to make the system work.

NOTE: there is a connection from the AI to the folder where this lives. For me, it’s a single folder that I’ve allowed AI to access on my system. So far, it’s worked well for me. I’ve noticed that I’ll occasionally get requests from either ChatGPT or Claude to access more folders when working on other things, which I very rarely do (only other instance I’ve allowed access was for my newsletter), and both LLMs still have been able to complete tasks well. More on this later in this article.

The setup

Let me explain how I set it up. First, I downloaded the system that you see listed in Github by going to Code and then using “Download Zip.”

Then, I opened the contents of that file; it will be a folder called “SimpleBrain-main.”

On my computer, I created a folder called “brain.” I opened SimpleBrain-main, copied everything in there and pasted it into my “brain” folder.

Then in Claude, I went to Projects, created a new Project, and as context, I added the folder called Brain. (For ChatGPT users, you can take similar steps with the same results.) To test out the system, I then opened a new chat and asked it to go to my website, review it, and then pull the data from the website into the “raw” folder. That resulted in this on my hard drive, in the “brain” folder:

  • brain

    • archive (empty)

    • projects (empty)

    • prompts (no changes)

    • raw

      • mrxplorer.md with information about my business gathered from my website

    • wiki (empty)

Once it did that, I then asked it to process what was in the “raw” folder using the translate.md file in the “prompts” folder. Once it was done, here's what I ended up with:

  • brain

    • archive

    • projects

      • mrxplorer

        • Readme (what the business is based on the website)

    • prompts (no changes)

    • raw (empty)

    • wiki

      • a couple of files based on the contents of my website

Adding the task list

Now that I knew that the basic system was working, I decided it was time to expand and add a new folder to this system called “tasks.” So, under the brain main folder, I now had:

  • archive

  • projects

  • prompts

  • tasks (new that I told AI to add)

  • raw

  • wiki

With the folder came new rules for the AI to follow when translating information from “raw” into these various folders. When it was processing any of the data from “raw”, not only did it need to update the wiki and the projects, but it also needed to create tasks.

However, I didn't want the system to always be running and constantly adding new tasks to a file. So instead, I scheduled a new task for AI to run at the end of each workday, take everything that was in “raw”, review today's task list for any changes that had been made, then combine all of that information to create a new task list for the next day.

I now have this running Monday through Friday, updating at the end of each day. I find myself tending to do a brain dump on the task list itself, which I have set up by project so that I instead of just having a super long task list, I can focus on a set of tasks associated with a particular project.

The drawbacks

Admittedly, not having a system connected to email and calendar has its drawbacks. The primary drawback is not having a way to keep track of emails that I might have sent and need to follow up on but lose track of, or emails that come through and get buried.

The second drawback is losing out on a way to connect email and calendar in a way that makes meeting prep a lot easier.

To be fair, because I use Google Workspace through a business license, theoretically, I should have a little bit of leverage with Gemini, but so far, I haven't really seen it work as well as I might hope. I do get prompts to follow up on emails that have been sent after a few days and haven't had replies, or emails that I have received and haven't replied to, but the connection between calendar and email has yet to emerge.

The benefits

The benefits, however, I personally think outweigh the drawbacks. The first benefit for me is not having that always-on connection to my email and my calendar. And because of that, things like prompt injection aren't going to happen because there isn't a way for a system to be reading an email that might have that white text prompt that is asking for access to a system. The information that the AI is processing is information that I am specifically putting into a folder and nothing else.

You may very well say, hey, but you have connected AI to a folder on your computer. And that’s a valid and fair pushback. But so far, I‘ve only seen it connect to that particular folder and not expand out to other folders. It's also working through Claude Cowork and connecting to my hard drive as opposed to connecting to a cloud folder on Google Drive or OneDrive. And so for me, limiting it to the hard drive and not including yet another cloud service in the process decreases that surface area for a privacy invasion. It can also only run if my computer is on. If my computer is off, the prompt can't run.

Overall, this system works for me because it:

  • decreases my concern about prompt injection via my calendar and email.

  • gives me a way to track my tasks in a single location and have them split out by project.

  • is a system simple enough for me to get set up and working within half an hour, including having added my task folder and adding to the initial prompt for what data was being processed and how.

  • runs on a daily basis to help me get prepared for the next day.

  • works really well with my neurodiverse brain due to the simplicity and having it in one place.

While this isn't necessarily 100% market research specific, I do think this system is approachable and flexible enough to be used not just by a freelancer to manage the different projects they might have, but also for a brand side researcher who is trying to manage multiple stakeholders, or even by someone who is looking for a simple yet flexible system to use for project management.

I can also see this system being used by a small team to manage a collaborative project together.

I’d love to hear if you try this system out and how it works for you. Simply reply to this and let me know how it goes, or if you have questions! And if you know someone looking for a system that helps them stay organized without having to connect their email and calendar to the LLM, share this with them!

❝

Interested in coaching on using AI tools to support your work? We start from wherever you’re at, with whatever LLM you’re using. One hour a week, tied to your work, so you can see the progress in your work. It’s a safe place to explore tools and learn how they work for you so you can take the lead in your career. Contact [email protected] to learn more!

Recommended for you