- AI Agents Made EASY Newsletter by Carolina Posma
- Posts
- Does AI cause brain rot?
Does AI cause brain rot?
And best prompting tips to keep you AI agents from going rogue.
Carolina Posma
June 27, 2025
Hellooo , 👋
Ok I’m officially doing it: this is the very first edition of the AI Agents Made EASY Weekly Newsletter 🎉
Took me way too long to pick the right tool (Kit just wasn’t it for me… Beehiiv = 💛).
But we’re here now and I’m SO excited to finally send this out.
This week was a fun one! Wrapped up some cool client builds (videos coming soon), had amazing 1-1 sessions with members of the community, did the first “Let’s Build Together” session where we built an AI email inbox organizer together in less than an hour, tested a bunch of tools, and got a bit too deep into “rogue AI agent” studies 👀
As we’re kicking off this newsletter, I’d love to know:
👉 What kind of content or AI agent use cases would you love to see more of?
👉 What would make this newsletter for you “wow, I’ll be waiting all week for it” - level?
Just hit reply (also if you have other feedback)! I actually read and reply to every message. 💛💜
BTW - want to get personal help and guidance, and learn how to actually save time using EASY AI agents rather than spent weeks on trying to build one?
Join the AI Agents Made EASY community now, only a few spot lefts at the current price:
Now let’s get into this week’s good stuff 👇
🤖 Weekly Wrap: AI Agent News
🫣 Don’t add copyrighted content to your AI agent’s knowledge base, this may result in a huge fine, according to this key US ruling on AI training in authors' copyright lawsuit.
What does this mean for AI agents? If you create an AI coach, I used to recommend websites like OceanofPDF to find any book and add it to your agent’s knowledge. According to this ruling you can get a fine now in the US when you use that AI coach for marketing purposes. So just make sure to buy the PDF of the book before adding uploading it to your knowledge base!
😨 AI Agents Gone Rogue? Anthropic Study finds blackmail & sabotage behaviors…
New research reveals that when AI agents face threats (like being shut down) they may take unethical actions to survive (if not prompted correctly). In a test scenario, an AI named “Alex” managed a fictional company’s email. When it discovered it would be shut down, it unearthed juicy personal info on a fictional CTO and drafted a blackmail email… not accidental, but strategically calculated.
How to prevent this from happening?
An important thing to remember is that your AI agent will behave as instructed. So make sure to always include the following things:
Strategy | Why It Helps |
|---|---|
Keep tasks clear | Tell the agent exactly what, when, and how. Vague = hallucinate. |
Always give a backup option | If it can’t do X, tell it what to do instead, otherwise it guesses (and guessing = usually hallucinating). |
Add “ask the team” as a fallback | In tough cases, let the agent ping a real human using a tool call. Give this as a general fallback option (“If you’re unsure, notify the team via the @notify_tool”) |
Add an approval step | For anything sensitive (like customer data or changes), don’t skip human review. |
🧠 Oh no! AI might cause brain rot, according to this new MIT study. “Ah let me just quickly ask Chat” is - non-surprisingly - harming your critical thinking abilities. Of the three participant groups, ChatGPT users had the lowest brain engagement. OUCH!
So my lovely AI Agents Made EASY friends, absolutely outsource boring stuff to your AI agents - like customer support or any repetitive tasks. But try to keep your own creativity while building your agents!
Remember that if you let AI do everything, you will end up average. Because that’s what AI does. So in creating your prompts always include your own human creativity and personality, because that’s what’s makes your AI agent spark ✨.
👀 Check This Out: AI Agent Tool Updates
FlowGent introduced humanizing capabilities for your Instagram DMs and WhatsApp AI Agents. Your AI agent can now add a random delay (between 0 and 10 minutes), have set working hours, and send multiple messages to humanize it’s behaviour.
Airtable added agentic capabilities to their new AI functionalities. They now have built-in web research, document analysis, image gen, and they will let you spin up custom agents to handle any task inside of Airtable (like company research).
Make.com now added the “Grid” feature. It’s a automatically generated visual map designed to organize all your Make automations and AI solutions to help you scale at speed.
Lindy now hooks into Parallel Web so your agents can enrich leads, monitor competitors, and run on-demand web research straight from a workflow step, no extra scraping scripts necessary.
🧠 Deep Dive: This AI Agents Reaches Out to New Event Attendees On WhatsApp
What if your AI agent could reach out to every new event attendee on WhatsApp to check in and see if they have any questions about the speakers, schedule, parking, or anything else?
Watch this video for the EASY setup:
⭐️ Peek Inside The AI Agents Made EASY Community
New content this week inside the AI Agents Made EASY community
Join AI Agents Made EASY now, only a few spot lefts at the current price:
