Journal - Week 2

Introduction

This week there is a lot to mention. I’ve been diving deeper into AI agentic architecture. Actually, I am proud to announce that I will be presenting next week in my adopted city, Logroño, a workshop on this topic.

But life is not only about AI. Other topics like Shopify are also worth mentioning.

As last week, let’s start.

Agentic Architecture

All good, this is the topic that has been blowing my mind in recent weeks.

You know, when AI exploded 4 years ago, there was such a lack of knowledge about what was going to happen that each of us started running around like headless chickens, screaming and trying to guess what the next step was to avoid becoming obsolete.

The very first thing that came to our minds was to learn about Machine Learning and what exactly an LLM was. We learned about convolutional networks, deep learning, transformers, encoders, Python, and Pandas.

From the very first moment, I felt that it was not related to what my career had been until that point. I had always built digital products and web apps.

But having control over how to build your own model seemed to me the logical next step.

Of course, being a Machine Learning Engineer is a career path in itself and far enough from software engineering to make me rethink whether I was doing the right thing.

Now everything is sorted out in my head. I know that ML Engineers / Data Scientists are other job profiles I can work with, but we serve different domains, and what I need to master is how to use those black boxes they build to power the apps I have always built.

And the word that summarizes it all is “agentic architecture”.

Agentic architecture is neither more nor less than a (web) app powered by specific LLMs and properly orchestrated.

This is a new layer on top of what we had before, with some new concepts and ways of working to learn, but nothing far from what I have been doing my whole life: engineering.

Semantic Kernel & Langchain4j Frameworks

I will not go through a long explanation in this post of what these two frameworks are. They serve the same goal: building agentic apps.

They are very similar. One is powered by Microsoft and looks a bit more modern, and the other one is a Java open-source project.

I have been building some demo apps using both of them, and I don’t yet have a solid opinion on which one is better.

What I can tell you is that, for Java, L4J looks more mature, allowing basic features such as streaming responses, which is still in development for SK.

It is also true that SK is mainly being developed for C#, as it is a Microsoft product, and Java is coming as a later option.

I will keep an eye on both of them (and maybe others), and I will keep you updated.

Ragas

Besides building agentic apps, or maybe closely related to that, I kept learning about RAG, and there was a framework for RAG evaluation that caught my attention. The name is RAGAS. (https://docs.ragas.io/en/stable/getstarted/quickstart/#step-3-set-your-api-key)

Evaluating the fidelity of a model or RAG architecture is very important. It is still unclear to me how to standardize this, but RAGAS looks promising, and I will add it to my tech stack and ask the team to try it.

Safety Rules for AI

Also, I took a small step to guarantee that AI development is safe. I read about skills, and I really think they can be a security hole. Therefore, I told my clients and teams to avoid using skills without knowing very well whether they are 100% safe.

Also, internally I have created an internal marketplace where we publish the curated list of skills we want to use in projects.

Therefore, we know we are safe, and we also homogenize our tooling. Let’s see if it works.

Others

Besides learning a lot about AI, agents, and agentic architecture, I have to say I did an MBA in Shopify user management and Meta Business.

I had to jump into this topic to help two of my clients: first, to clean up the Shopify user / roles chaos they had internally, and second, to start advertising programmatically on Meta.

And yeah, although it is not a 100% CTO duty, I enjoyed gaining functional knowledge of these tools. You never know when you will need to use them. In the end, Shopify is the market eCommerce leader, and Meta Business is one of the most important platforms for digital ads.

Note: NOT AI-generated at all. I only passed it, after writing, through Claude to correct grammar errors as I am not a native English speaker. BUT the main ideas and ways of expressing them remain untouched.