I've started working on the MPC in JavaScript book. The more I dig into it, the more I find myself reflecting on timeless AI engineering tech - the kind of foundational knowledge that pays dividends for years to come. Alongside MCP, I've identified: LLM evals and observability , AI Agents management, RAG and context engineering, the inner workings of models.
What else would you add to this list?
In other news:
-
🎙 I had the pleasure of interviewing Ishan Anand on how he implemented GPT2 in Excel, learning AI while having a full-time job, and why it's important to demistify the magical AI black box and how LLMs work under the hood. You can watch the full episode here:
-
🎓 If I had to recommend just one book on AI Engineering, it would be AI Engineering by Chip Huyen. Marina Wyss did an excellent job summarizing the book in this video. By the way, you can imagine how information-dense the book is, given that the summary itself is one hour long.
-
📚 Speaking of book recommendations — The Second Mountain by David Brooks has almost become a book I study rather than simply re-read. It feels like a real Lindy book, as Nassim Taleb would call it.
-
🛋 And after relocating to the south of Spain, I suddenly had a lot more time to read. One of the books I finally finished was Tomorrow, and Tomorrow, and Tomorrow by Gabrielle Zevin—and wow, did I enjoy it! Definitely worth checking out.
-
🎧 I’ve always admired how Matt Giovanisci runs his lean business - $1M a year with just a team of three. His podcast episode on ideas in the info products really shifted my perspective and influenced how I’m approaching the products I’ll be building next.
-
🤖 This is an example of a real production prompt used for an AI agent. Check out the high level of complexity and detail involved. The prompt comes from the Bolt.new app. As with code, LLMs prompts are constantly updated and evolving.
-
🗃 Modern RAG is likely the most economically valuable form of Gen AI. Cursor or Claude Code are basically RAG apps. One of the real performance gains Cursor has made is using embeddings to figure out where your cursor should jump into for the next tab action. Jason Liu gave a great talk about this.
-
🎰 One of my bets is that, soon, most applications will have an integrated AI/LLM layer, just like they have a database layer. The AI layer could be used for stuff such as NLP capabilities, to enhance the UX, make sense of unstructured data, and even communicate with other AIs.
-
👔 A week ago, I did a talk on MCP in front of 30+ devs. I love writing and hate speaking. So much energy goes into how you speak instead of what you say. When writing, you can edit and refine. When speaking, it's all real-time. But being good at speaking will always give you a huge edge in so many levels.
-
👨💻 When developing your app, start by testing your prompts with smaller AI models. They fail quickly and often - which is exactly what you want. Larger models are so capable that they can solve some of the mistakes, making debugging much harder. On top of that, smaller models are cheaper and faster, giving you a tighter feedback loop.
-
💡 Learned this one, the hard way! You can't combat "burnout" or mental fatigue through rest only. You need wild curiosity and some level of intellectual vagabonding, while doing stuff just for sake of pleasure it gives you, and not to some material - or prestige - reward.
-
✍️ Started the work on the MPC book. I've noticed that writing a technical book feels so close to writing software; PR's and incrementing in versions, using VS Code as my editor, refactoring a lot, using code snippets and autocomplete. I love love this process!
-
📦 If you want to build a portfolio of AI Engineer projects using JavaScript you can check out the GitHub profiles of Hrishi Olickel and Hassan El Mghari. Many nice projects over there! Btw, some time ago I interviewed Hrishi Olickel here.
-
📘 Speaking of AI I think LLM orchestration via code is much more powerful (and fun) than relying on just pure prompt engineering. This is one of the reasons I've written Build AI Agents with LangGraph.js. As Harrison Chase, founder of LangChain, says: "It is all about communication!". Based on the growth chart of LangChain on NPM there are high chances that it will be the next star of the JS word.