Why the smartest AIs still mess up breakfast

Let me tell you about a friend of mine.

He’s building a robot that makes pancakes.
You say “blueberry,” and it flips a perfect golden circle onto your plate.

(Yes, this is real. Several people I know are working on kitchen robots right now. The opportunity is massive with the rise of new AI tech.)

Except—it’s not working.
At all.

53 blueberry pancakes ended up on the floor.
A bottle of syrup… half-empty… dripping into the carpet.

Why?

Because the robot didn’t know when to stop.
It didn’t understand context.
It just did what it was told—by a human, and an AI language model—over and over again.

(Read that again: a language AI told it to do this.)

As impressive as AI might seem, it still has huge blind spots: a genius parrot that can do complex reasoning but can't make toast.

🤯 Two Missing Pieces in Today’s AI

At the heart of this are two things:

• Context – What the AI understands about your world
• Tools – What the AI is allowed to do in your world

Most AI today?
They have brains but no senses.
"Hands" but no awareness (if you connect to a tool).
Instructions but no understanding.

Impressive.
But clueless.

✨ Imagine AI That Actually Gets You

Let’s dream for a moment.

What if AI could:

• Block out time before you burn out
• Book meetings aligned with your values
• Protect dinner with your family
• Write emails with empathy and clarity
• And yes—remind you of your mom’s birthday 🎂

That’s not a calendar.
That’s a time wizard.

But we don’t have that wizard yet.

(If you know a solution that does all of this, please share and take my money)

Why?

Even the smartest AI doesn’t understand your full context.
And the tools it needs? Scattered across apps that don’t talk to each other. Like having dozens of different cables to connect to each digital aspect of your life: calendar, work, health apps, etc.

Another example.

🧦 The Laundry Problem

AI can write poetry.
It can build complex software.
It can summarize a research paper in seconds.

But it can’t do your laundry. Or cook your dinner. Or clean up your fridge.

(What we call "low value" tasks...)

Not because it’s not smart.
But because it doesn’t understand your house, your clothes, your routines. Additionally, you need to connect it with hardware, tools, dexterous arms, and moving robots, among other things.

It’s always the same two problems:

• Context – Where are the clothes? How should each be washed? What detergent do you use?
• Tools – Arms. Sensors. Appliances. Rules.

The issue isn’t just about intelligence.
It’s about connection.

🧠 Solution? Model Context Protocol (MCP)?

That’s where (what we call) Model Context Protocol (MCP) makes complete sense.

It’s a new, universal way to give AI what it desperately needs:

Understanding and action.

You may have seen it on X, LinkedIn, or Instagram and thought:

“Okay, looks like a new tech hype. Probably for developers.”

But here’s the truth:

This is for everyone.

MCP is going to affect how we live, how we work, and how we trust AI—whether or not you ever write a line of code.

Note: If you’re a developer already building MCP servers, this letter isn’t for you. I won't be explaining technical details or how to build MCP servers. You can read my tech blog for such details.

However, if you’re just a curious mind and want to know what this is all about and what this all means, read on.

📦 So… What Is MCP?

Think of MCP as a universal plug for AI.

It’s not a product.
Not an app.
Not a startup.

It’s a standard—like USB-C—but for connecting AI systems to the world around them.

Right now, developers are building AI-first apps using it.
Companies like Anthropic and OpenAI just adopted it.
It’s happening. Quietly, quickly.

🧠 The Core Idea Behind MCP

There are just two things you need to know:

1. Resources = Things the AI can look at
   (Think: your files, code, processes, docs, calendar, photos, PDFs, even fridge inventory...)
2. Tools = Things the AI can do
   (Think: send emails, make purchases, control smart home devices, write to a database, book a meeting...)

Resources = Context
Tools = Actions

Simple. Powerful.

The magic? Once you define these things clearly, AI can use them autonomously.
It can act intelligently, within boundaries you set. The protocol takes care of the details, including sending information, explaining the tools, and how the AI should utilize them.

Pretty powerful, I should say.

🤖 Example: The Smart Fridge

Let’s say your fridge can talk to your phone and order groceries, and is also aware of your diet, fitness targets, and the number of daily calories and protein you're aiming for, and may be connected to your fitness apps, like Whoop or Apple Watch...

(I think we will have products like this in the near future (a couple of months); the demand is huge.)

We do have the technology to build such a thing NOW. The only missing piece was something like MCP.

The problem:

Well… what happens when your AI doesn’t know what’s already in the fridge?

Or worse—what if it keeps ordering eggs every time you say “I’m hungry”?

With MCP:

• Resource/context = A live inventory of what’s inside + your health app + your targets (a prompt?)
• Tool = A “buy groceries” action with rules for quantity, budget, timing

Now when you say, “Make me dinner,” it knows what you have, what you need, and what not to waste.

🏢 Why Businesses Are Paying Attention

This isn’t just about fridges and pancakes.

Think:

• Healthcare – AI that sees only approved data, books real appointments
• Customer Service – Bots that understand your issue, take safe actions
• Finance – AIs that help you spend wisely but never transfer money without your okay

MCP isn’t a toy.
It’s the plumbing for the AI-powered world.

It’s what turns AI from something clever… into something trustworthy.

Prediction: MCP will become an integral part of every modern company's IT system within months.

💼 Career Hint

Roles like MCP Engineer or AI Context Architect are going to be everywhere.
This is early, foundational technology. Like the internet in the ’90s.

If you’re thinking about what skills matter next—this is one to track.

🧪 What’s Already Happening

People are using MCP today to:

• Build private AI tools that read your notes but not your bank info
• Let AI handle customer requests without breaking anything
• Connect LLMs to real-world systems, from storage to smart homes
• One guy built a Figma (design tool) MCP to generate mockups and designs on Figma within minutes

In a nutshell, MCP is turning working on any app to a natural conversation.

It’s open-source, free, and available now. Anyone can build their own server and add any context or tool to AI agents.

Smarter Isn’t Enough. AI Needs Context & Tools.

I don't like to talk about the "future of AI" — because nobody has a clue of what is going to happen a couple of months from now!

But if I had to say something, I would argue that the future of AI isn't about clever prompts or fancy algorithms.

It's about giving AI the right context to truly understand what makes our lives better.

It's about setting smart boundaries that keep people safe.

And most importantly, it's about making AI work for people, not destroy them, isn't it?

The real question that keeps me up at night isn't "how smart can AI get?"

It's this:

"What happens when AI can do anything... but doesn't understand WHY it matters to humans?"

That's what keeps me motivated to work on this. Because I believe we can build AI that truly gets us.

And hey, I'd love to hear your thoughts on this! What aspect of AI excites or worries you the most? What would you like to see in the next issue of this letter? Hit reply and let me know - I read every email, and your perspective helps shape what I write about next.

This wraps up the beginner-friendly run.

Next issue, we’re getting back to the heavy hitters.

Until next time,

—Charafeddine

P.S. If this letter left you scratching your head about MCP, don't worry - drop me a note and I'll clarify things in the next issue!