Introduction
First of all, why you should use AI? Easy, because it makes you output stuff faster and easier. How? Let's see together:
First of all, you need to have at least one AI model always accessible, with a shortcut. For this, you can use Raycast (which is my favorite choice), or you can change your search engine and use OpenAI, Perplexity, or any other model. You don't know how to do it? Awesome; ask your favorite Chatbot.
That's the mood change you need to have. In this AI era, you cannot not-know something... or better, gathering easy-to-understand knowledge is as easy as ever before. SO, whenever you don't know something, just ask AI. For real... don't bother your co-workers, don't bother your friends, (don't bother Google... because it's trashy). JUST ASK AI.
So after you understood this, let's see where we stand with AI.
How to make a Great Prompt?
What are some tips for writing effective prompts?
Some piece of advice to write good Prompts:
- Read the prompts you see online and test them yourself.
- Try to test the model on stuff that the LLMs shouldn't be able to do. Always push their boundaries.
- Be Clear and Specific
The more specific your prompt, the better the response. Avoid vague instructions and clearly state what you want.
Example: - ❌ "Tell me about stars." - ✅ "Explain the life cycle of a main sequence star like our Sun, including key stages and timescales."
- Specify Output Format
Clearly define how you want the response structured.
Example: "List five renewable energy technologies in a table with columns for: Technology, Efficiency Range (%), and Main Advantages."
Sources:
Create a Prompt Automagically!
Try Claude's Prompt Creator tool to build, test, and refine your prompts with a user-friendly interface. It helps you structure prompts effectively and see real-time results.
Context Rot - Be Gentle to your AI Memory
Why Your AI Assistant Gets Worse Over Long Conversations
Ever notice how an AI assistant starts out sharp, then gradually seems to... forget things? You find yourself repeating context, explaining the same stuff, or watching it contradict what it said an hour ago.
That's context rot.
What's Actually Happening
AI assistants have a memory limit. Think of it like a whiteboard — there's only so much space. As your conversation grows, older stuff gets erased to make room for new stuff.
The problem? Sometimes what gets erased is important.
Example — The Vanishing Requirements:
Message 1: "I'm building a recipe app. It needs to work offline and support metric/imperial conversion."
Message 47: "Can you add a feature to save favorites?"
AI: "Sure! I'll add favorites. Users can save recipes and access them anytime they're connected to the internet."
You: "...I said it needs to work offline."
The AI didn't ignore you. It just ran out of whiteboard space.
Common Symptoms
The déjà vu conversation:
"We should use PostgreSQL for this."
*(30 messages later)*
"I'd recommend MongoDB for your database."
"...You told me PostgreSQL an hour ago."
The endless re-explanation:
"Remember, this is for a kids' educational app, so keep the language simple."
*(10 messages later)*
"Here's the copy: 'Optimize your learning outcomes through our pedagogical framework.'"
The forgotten constraint:
"Budget is $500 max."
*(Later)*
"This $2,000 solution would be perfect for your needs!"
Why It Happens
Long conversations
The longer you go, the more gets lost. A 200-message conversation is practically begging for context rot.
Think of it this way: If you had a 4-hour meeting with no notes, how much would you remember accurately by the end?
Topic jumping
Mixing unrelated things in one chat muddies the waters.
Example — The Context Collision:
"Help me write a stern email to a vendor who missed a deadline."
*(Done, moving on)*
"Now help me write a birthday message for my mom."
AI: "Dear Mom, I'm writing to express my disappointment in your failure to meet expectations this quarter..."
Extreme example, but you get the idea. Context bleeds.
No external notes
If everything lives only in the chat, it's all vulnerable.
Bad approach:
"Remember, the API key is xyz123 and the endpoint is api.example.com/v2"
This will disappear into the void eventually.
How to Avoid It
Write things down outside the chat
Keep a simple doc with the essentials. Nothing fancy needed.
Example — A Simple Project Note:
# Recipe App - Quick Reference
## Core Requirements
- Offline-first (must work without internet)
- Metric/imperial conversion
- Budget: $500
## Decisions Made
- Using SQLite for local storage (Jan 15)
- Going with React Native (Jan 16)
## Current Status
- Basic UI done
- Working on search featureNow when the AI forgets, you just paste the relevant bit.
Start fresh when it makes sense
Don't let conversations run forever. When you start a new chat, give it a quick handoff.
Example — Good Handoff:
"Starting a new session on my recipe app. Quick context:
- React Native app, offline-first
- SQLite for storage
- Just finished the search feature
- Next up: implementing favorites
Here's the current code structure: [paste or link]"
The AI now has everything it needs without 200 messages of baggage.
Keep projects separate
One conversation per project. Don't do this:
Bad:
Chat: "Website redesign + Tax questions + Recipe app + Gift ideas for spouse"
Good:
Chat 1: "Website redesign"
Chat 2: "Recipe app"
Chat 3: "Quick questions" (for one-off stuff)
When It's Already Happening
Notice the AI getting confused? Don't push through.
Example — The Recovery:
Before (pushing through):
"No, I said PostgreSQL, not MongoDB. And it needs to work offline, remember? Also the budget is $500, not $2000. And it's for kids, so simple language please."
You're now wasting half your messages on corrections.
After (clean reset):
*(New chat)*
"Hey, picking up a project. Here's where we're at:
- Kids' recipe app, offline-first
- PostgreSQL database
- $500 budget
- Simple, kid-friendly language throughout
Currently working on the favorites feature. The save button should..."
Fresh start, no baggage, no corrections needed.
The One Thing to Remember
Don't rely on the AI to remember. Write it down somewhere else.
Think of yourself as the AI's external hard drive. The chat is just working memory — temporary. Your notes are long-term storage.
A 30-second note now saves you 10 minutes of re-explaining later.