On the fear of missing out
How have you been?
I had a great time visiting my family in Vietnam and then taking a holiday in Japan— so I was away for a while, sorry for that! Honestly, I am still in my holiday mood but slowly coming back now 😅
Last week, I finally found the time to learn MCP (Model Context Protocol). It filled the news a couple of months ago, so I’m definitely late to the party, but my imposter syndrome forces me to study it.
And you know what? I think it is pretty cool!
What is MCP and Why It Matters
If you work with agentic systems, you’ve probably heard the terms Function Calling or Tool Use. Let’s say you’re building a research agent to analyze data or documents. You might want to connect your agent to GitHub repositories or Google Drive documents to fetch relevant files. Instead of you writing your own tools for these integrations, MCP allows you to expose available tools to your agents via an MCP server and use them across many different applications.
It is similar to Language Model Protocal, where developers can get consistent Python language support across VS Code, PyCharm and other editors through a single language server rather than separate plugins for each IDE.
Here’s how it works when your agent or application becomes an MCP Host and interacts with MCP Servers: or example:
User: “what are the cheapest options to go to London by train between 1st of August and 10th of August?"
*AI: thinking... let me see which tools are available*
<AI connects to the MCP server>
<AI asks for a list of available tools>
<MCP server returns [list\_time\_table, list\_price, ...]>
*AI: thinking... the user wants to know the cheapest options to go to London, so I will use the list\_price tool*
<AI calls list\_price tool>
<MCP server returns [01/08: 120, 02/08: 110, 03/08: 200, ...] >
*AI: thinking... the list\_price\_tool returned [...] and the cheapest option is 110 on the 2nd of August.*
AI: the cheapest option to travel to London is on the 2nd of August for a price of 110 euros.
The Learning Cycle: FOMO, Imposter Syndrome, and Curiosity
Pretty cool! Okay, now I have learned MCP. And then I thought, what is next?
I have been hearing people talk about knowledge graphs—maybe I should study that.
In this field, you have to keep learning. Whether it’s FOMO, imposter syndrome, or just genuine curiosity, something keeps you going. And honestly, having a healthy relationship with FOMO and imposter syndrome is just fine.
But I realize that this pattern keeps repeating. The pressure to keep up makes me learn one thing and move on to the next thing. I learnt the next thing, and then the next next thing. By the time I have learnt 10 things, I have already forgotten the previous things I learned and maybe never touch them again.
Does this sound familiar?
Sounds familiar? I know!
Ever feel like AI news is just too much sometimes? You know that feeling when you’re scrolling through LinkedIn feed, and every second, you are thinking:
“OMG, there is another AI thing!”
“Wait, what is this framework? I have to study it.”
Well, this isn’t necessarily a bad thing. You don’t have to be an expert in everything. But if you know the concept and keep it at the back of your mind, then when something comes up— a project at work or a consulting service that needs it—you will know which technology to use.
My strategies with learning new things
There is just so much hype in this AI industry. How to know what is hype and what is not? How to make things that you learn stick? My advice?
- If something is not hype, it will stay. Wait for the hype to settle, then dive in
- Start with the “why” before the “how”: Before diving into tutorials & building things, spend time understanding what problem each technology solves
- Build small, build now: Don’t just read or watch— do it! Pick the tiniest possible project and try to build it. The frustration of getting stuck teaches you what you actually need to learn.
- Connect it to something you already know: Tie the new concept to a project you are working on and think how you can use this new concept!
So I built my MCP Server
So I decided to build something small to solidify my MCP understanding. I am planning a short trip to London this summer, but the train tickets from Amsterdam are quite expensive. I already had a scraper that scrapes ticket prices, so I thought: why not expand it further by adding MCP components? It’s a great way to learn the new concept by getting your hands dirty and building something practical.
What about you? Have you been learning anything new lately? I’d love to hear about your experiences—reply and let me know!
What we’ve been reading/watching this week
- I recently finished the AI Engineering book by Chip Huyen. It is a comprehensive guide to building real-world AI applications. It covers all the aspects of building AI applications, from understanding foundational models, dataset engineering to evaluation. There is a whole chapter for each key-concept like prompt engineering, RAG, fine-tuning, agents, and dataset engineering. A lot of focus on AI-as-a-judge, evaluation benchmarks and evaluation metrics. Definitely worth reading!
- Google Deepmind created a new model, alpha evolve, that made a breakthrough in 20 open math problems, including more efficient ways of doing matrix multiplication.
- A collection of leaked system prompts for many AI models.
That’s it for now. Catch you soon!
With love,
Lan and Robert
