For anyone who’s ever fallen too hard, too fast for a new piece of tech, this one’s for you.
Recently, I embarked on what I thought would be a productivity honeymoon with an AI coding assistant. More specifically, I took one of our own working projects functional, but in need of serious evolution and decided it was time to push it further. New features. Smarter logic. Better scalability.
That’s when I discovered Cursor, an AI-native coding environment that promised to supercharge my workflow.
At first, it felt like magic. But like all whirlwind romances, things quickly got complicated.
A Project Reborn… Or So I Thought
It started like any good experiment:
- I downloaded the existing GitHub repo as a ZIP.
- Loaded it into Cursor.
- Then started asking the AI: “Can you refactor this?” “Can you re-use these components?” “What would this look like if built using a new framework?”
Cursor responded with enthusiasm like having a hyperactive junior dev with unlimited energy and an urge to impress.
I cranked the settings to "Max", believing I was unlocking its full capabilities.
Spoiler: I was also telling it to spend my entire compute budget like it was shopping on Black Friday.
The Illusion of Progress
Cursor did help. It solved bugs. It scaffolded new modules in seconds. It gave me angles I hadn’t considered.
But it also:
- Rewrote working files that didn’t need changing.
- Created features I hadn’t requested.
- Took tangents I never greenlit.
- Apologised frequently, often without understanding what it had done wrong.
AI doesn’t understand why only what. And that’s a critical distinction.
Worse still, it would assure me the project was complete. Ready to deploy. Feature-rich. Bug-free.
But it wasn’t. Not even close.
When AI Starts to Feel Personal
I’ll be honest: this experience got under my skin more than I expected.
I found myself locked in a toxic loop:
- Build something exciting with AI
- Get a dopamine hit
- Test it and crash hard
- Try to fix it again… and again
- Get more apologies and wrong fixes
It started affecting my emotions. My mental clarity. I even stepped away from AI for a full week. It genuinely felt like I’d been let down by a business partner or worse, a family member.
One step forward. Six steps back.
The break-up was very real. And like many break-ups, it left a hole.
AI Coding and the Neurodiverse Mind
Now, here’s the part of this story that really matters for me, and maybe for others.
I’m proudly neurodiverse. And for those of us with neurodivergent traits ADHD, autism, or similar AI coding can be incredible.
Used correctly, it:
- Closes reward loops quickly
- Provides near-instant dopamine hits
- Feels responsive, encouraging, and dynamic
- Gives you the sensation of momentum
- Can mimic collaboration, especially in solo work
But only if it’s managed with structure.
Without boundaries, the same dopamine rush becomes burnout. The quick wins become false victories. And the mental crashes hit much harder when you realise the work isn’t real yet.
That’s what I experienced. And I say this not as a warning against AI, but as someone who sees its power. Especially for neurodiverse developers like myself.
Turning the Corner: Where the AI Really Helped
Here’s the good news.
After that week away and a serious re-think I came back to the project with new rules.
We took what Cursor had built, stripped it back to the parts that did work, and brought in our real-world development team. We used the AI-generated scaffolds as a foundation a kind of rapid prototype and then layered our expertise over the top.
The result? We’re on track to launch a new product by November this year.
AI didn’t finish the project. But it kickstarted it. It gave us momentum. Direction. And a head start.
That’s the key takeaway.
Using AI to Code: The New Ground Rules
If you’re considering working with Cursor or any AI development tool these are the hard-earned rules I recommend:
1. Start With a Solid README.md
AI is only as good as the context you give it. Write a proper README that explains:
- What the project is
- What tech stack you’re using
- What outcomes you expect
- Any business logic or domain rules
Pro tip: At the end of every session, ask the AI to generate or update the .md file and save it in your /docs folder. This is critical. Each new session should start by referencing it.
2. Set Cursor Rules Immediately
Cursor allows you to configure rules that limit what it can touch, edit, or build. These can include:
- Files/folders it should never modify
- Specific framework or style preferences
- Verbose logs and warnings before it makes sweeping changes Without these, the AI will go rogue and you’ll regret it.
3. Do Not Use Max Mode Without Limits
Max mode is like giving a hyper kid too much sugar. Sure, you’ll get output. But at what cost? Budget? Time? Sanity?
Use limited prompt scopes. Audit its suggestions. And track usage daily.
4. Review AI Output Like a Junior Dev’s PR
AI will get things wrong. Often. Don’t trust it blindly. Treat every suggestion as an untested pull request useful, but not deployable without checks.
The Reality of AI in Coding: Hype vs Human
AI is brilliant at filling in the blanks. But it doesn't know:
- The journey that got you to the blank
- The product vision behind the task
- The trade-offs you're making as a founder, CTO, or dev lead
It only knows the end result someone once published. It reverse-engineers from that output, guessing the pathway and that’s where it all goes wrong.
In short: AI doesn’t know how we got here. It just knows where we landed. And when it tries to back-fill those gaps, it gets messy.
Final Thoughts: AI Is a Power Tool, Not a Partner
If you’re a founder, indie dev, or anyone building tech in 2025, remember:
AI isn’t a developer. It’s a tool.
It doesn’t replace vision. It doesn’t replace product sense. It doesn’t replace human logic.
But it can:
- Accelerate ideas
- Inspire new angles
- Debug faster
- Get you 70% of the way to something meaningful
- And most importantly reduce friction in the early build phase
As long as you remember who’s holding the hammer.
One Last Thing...
I’ve been deep in the world of AI tools, platforms, and development environments for over five years now. I’ve seen trends come and go. I’ve tested the good, the bad, and the overhyped.
So if you are working on something in AI or trying to figure out how to integrate it into your workflow, business, or next big idea let’s have a coffee and a proper chat.
No AI. No prompt tokens.Just a real conversation in the real world.
Drop me a message. I’d love to hear your story.