🧙🏼 Your brain on AI

SUBSCRIBE | ADVERTISE

Howdy wizards,

In this edition:

• MIT’s new study on what AI does to your brain, and why it’s being misinterpreted

• I’m working on turning my weekend project with Cursor into a real product

All you need now is massive quantities of filtered coffee and a reclining chair.

Here’s what’s brewing in AI.

DARIO’S PICKS

1. MIT study on ChatGPT’s effects on the brain

MIT researchers recently published a study on how using AI affects our brains. It tries to showcase potential long-term effects of being dependent on AI and what happens if it’s suddenly taken away from us.

The experimental setup:

• 54 college students wore brain-monitoring headsets while writing a series of essays over 4 months.

• Students were split into 3 groups: the first group could use ChatGPT, the second could use Google search, and third had to rely solely on their brain.

• Participants were tasked to write an essay in 4 separate sessions.

  • Sessions 1-3: everyone used the tool they had been assigned.

  • Session 4: the script was flipped and the ChatGPT group were denied using AI, while the brain-only group suddenly got ChatGPT access. The Google search group was used as like a middle ground / control and stayed the same throughout the assignments.

In session 4 where the groups switched tools, it got juicy. The brain-only group maintained high brain activity while using AI and strategically improved their work with it—achieving the best overall results.

On the other hand, 83% of ChatGPT users who suddenly lost their access to the tool couldn't quote their own essays minutes later. Overall, they showed weak neural activity and started mimicking how AI writes. The researchers coined this phenomenon “cognitive debt”, where AI replaces effortful thinking while our own abilities atrophy.

‎ Why it matters‎ ‎ The mainstream media loves a good clickbait story, and this study is being widely reduced to "AI makes you dumb".

AI is definitely shaping how our brains are wired through neuroplasticity, but how you use it matters.

From my interpretation, it’s this: anything you leave to AI gets amplified. A recent Tumblr post I read framed it beautifully - AI assistants are mirrors reflecting how we use them. Approach AI as a lazy shortcut, and that's what it becomes. Use it to challenge your thinking and refine your ideas, and it morphs into something else entirely.

Be purposeful about how you use AI and how you don't. For work that matters, apply your own reasoning first. Then bring in Chat or Claude to help you improve it.

IN PARTNERSHIP WITH CODECADEMY

Get hands-on experience using the latest tools in tech

Get ready for new career opportunities with hands-on learning from Codecademy. 

With 600+ interactive courses, Codecademy Pro helps you learn job-ready skills like AI, cybersercurity, and data science. Learners can also practice for technical interviews, build portfolio-worthy projects, and prepare for top industry certification tests.  

Ready to give hands-on learning a try? What’s brewing in AI readers can save 15% on an annual Pro membership when they use code SKILLUP15 at checkout.

Try Codecademy

UP CLOSE

In this mini-series I share different ways I’m using AI from week to week, as well as practical tips & tricks I discover and actually use.

Building an actual product using Cursor

Two weeks ago, I showed you how I built my first web app, an analytics tool, in a weekend using Cursor + Claude. Basically, I had an optimisation problem in running this newsletter which required real and continuous data analysis. Using Cursor, I hacked together a solution that gives me actionable insights to make better decisions. I’ve been thinking that I want to improve and launch it, so others can benefit from it too.

As I chatted to Cursor/Claude about what it would take to actually bring this tool to market, my weekend confidence took a serious hit. I realised what I have here isn’t "finished software", but rather more of a proof-of-concept held together by a lot of duct tape.

I’m currently learning the difference between building something that merely works and building something that would be able to handle real users without exploding.

Welcome aboard my learning curve of learning to code with AI as a non-techie. This week: going from “wow, it works” to “how can I build it properly?”.

Validation of the problem the app solves through Deep Research

I built this app because I had a problem. But I had no idea if other newsletter operators had the same challenge.

Enter ChatGPT's Deep Research. I had it systematically dig through Reddit to validate whether other people are having similar problems, to what extent, and what is the level of “pain” behind it (more pain = more potential). Not only did it confirm that yes, tons of creators are struggling with the same analytics blind spots—it also gave me a solid insights about which metrics matter most and which features would actually move the needle for them.

In 30 minutes and 3 targeted prompts, I completed what would have been days/weeks’ worth of market research not long ago. I'm probably going to write a separate piece about this process because it's turning out super helpful and is widely applicable to anyone building products, whether you're in development, product, or marketing.

With a liveable degree of market validation locked in, it was time to tackle the technical reality check…

The technical side of building a actual product

Cursor loves dumping everything into one massive code block. The problem of that is like those old Christmas lights where one dead bulb kills the whole string.

This became painfully clear as I kept improving the app and my code got bigger. Data points would randomly shift. Input fields, buttons and tooltips would suddenly stop working. Cursor would misunderstand which specific calculation I wanted to update because everything was tangled together.

The problem was a complete lack of code structure. I needed to do what developers call refactoring—restructuring code without changing what users see. Your app looks identical, but under the hood, everything gets organized into proper components. Each piece has its place, and updating one thing doesn't break five others.

So I asked Cursor: "Create a step-by-step refactoring plan to make this ready for real users." It delivered a detailed roadmap I'm still working through.

But refactoring was just the first part of this. With cleaner code, I started thinking bigger: "If I want thousands of users, are we building this right?"

Turns out, we weren't. Claude suggested some fundamental changes: integrating with a proper database that can process millions of data rows quickly, adding authentication and security layers, and moving calculations to the backend instead of exposing everything at the front-end level.

Conclusion: my weekend app needs to put on big boy pants if it’s going to serve real users.

Helping AI see the changes it makes

Cursor was struggling with a particular functionality I was trying to build. One thing I found helpful is sending it screenshots after each iteration, so it sees what I see. However, while it kept improving the app with each screenshot, the whole back and forth between me and AI was taking lots of time. So I asked it to “go online by yourself and take screenshots of the app and iterate until you get pixel-perfect accuracy”. What do you know, it immediately built a script that takes screenshots and informs itself with it recursively!

Watching AI pull up my web browser and screenshotting away feels like living in the future. It doesn't always nail the pixel-perfect accuracy, but it gets significantly closer than flying blind.

What I'm learning about the role of domain expertise

The more I work with AI on this project, the clearer it becomes: AI builds the interface to your solution, but it can't (yet) replace your reasoning about the problem itself.

If I had just handed over the challenge of building the metrics in the app directly to Cursor without deeply understanding both the calculations I wanted to include and what they are supposed to represent, I would have gotten something that looked impressive but solved the wrong problem. My domain expertise—knowing which questions to ask and why they matter—is what's making this tool actually useful.

AI is incredible at execution. But the strategic thinking? That's still on you.

The bigger picture shift

Until very recently, building something like this without technical help would’ve been far out of the reach of non-technical individuals, including yours truly. Importantly, I'm not just lazily typing what I want into Cursor but putting real cognitive effort into understanding the problem at depth and combining that with willingness to learn what software development actually requires.

Something I think is worth repeating: we're witnessing the collapse of the technical barrier between "good idea" and "working solution". Most people these days still stop at ChatGPT. You can get a solid step ahead by downloading Cursor and playing around with ideas.

PS I’m not close to finishing an MVP here. At a minimum I’d need user authentication, payment processing, setting up a secure database, API integration with the user’s data source…the works. But two weeks of part-time hacking on this, I'm starting to see how it could actually work. As I keep building and inevitably face-plant into new challenges I didn't see coming, I'll let you follow the messy journey. Stay tuned for more.

SPONSOR NOTE

Codecademy is sponsoring today's newsletter, and in a genuinely weird coincidence, they happen to have a solid beginner’s guide to Cursor with practical examples. Sometimes the universe just aligns like that. You can create a free account and search 'Cursor' if you're interested.

END OF REPORT

You made it to the footer. Fascinating. You're clearly in the 'actually building things' camp rather than the 'still asking ChatGPT to write emails' crowd. What's landing? What would make this more useful? Hit reply and let me know.

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.