Introducing The Agentic Architect: a weekly newsletter on building AI agents
Starting today I send The Agentic Architect once a week. Every issue carries one AI agent architecture pattern, one case study from production and the AI news that changes how you build. It is free and comes in English or Polish.
I write on the blog rarely but at length, because one post is one problem taken apart all the way down. The thing is, a lot happens between posts. A pattern works for a client. Something breaks quietly for a week and nobody sees it. A model ships that forces you to rebuild half a pipeline. Until now those things ended up in my notes. Now they go to your inbox, once a week, in a form you can read over a coffee.
What every issue carries
Every issue has three parts, always in the same order.
A pattern. I take one way of building an agent, draw it as a flow and write about when it fits, where it breaks and what you need to put around it for it to survive real data. A good example from the blog is an evaluator that grades its own work. When the judge and the author are the same model, it approves almost everything. So someone else has to do the checking.
A case study. A real failure or fix from production, with no client names. I describe what happened, why nobody caught it and what finally stopped it.
AI news. I pick only the releases and papers that are worth changing something in your own system for, that week. The rest I skip. Even when everyone is writing about it.
This is what I do hands-on: advising on AI strategy and building agents that survive the demo.
Who I write it for
I write for the people who decide how an agent is put together. I call that role an agentic architect, although the business card usually says something else: engineer, tech lead, CTO, founder. They share a question that comes back in every project. Which step do you hand to the model, which do you keep in code and which belongs to a person?
More depends on that answer than on the choice of model. Some of the decisions that shape it I wrote down earlier as rules for what code holds and what the model is allowed to do. In the newsletter I take one such decision a week and work through it.
The Agentic Architect
Practical patterns, case studies and AI news. Zero spam, once a week.
You'll get one email to confirm. Unsubscribe with one click. Privacy policy
How it works under the hood
The whole thing runs on my own server. Listmonk, open source newsletter software, manages the list, and Amazon SES sends the emails from Frankfurt. A sign-up has to be confirmed by clicking a link in an email. Without that, nobody lands on the list.
I do not add a tracking pixel, so I do not know who opened an email, and I do not want to know. You unsubscribe with one click in the footer. The English and Polish editions are separate lists, and the language picks itself: you get the one the sign-up page was in.
The form loads no outside scripts. The address goes straight to my own installation, which also keeps count of attempts, so nobody can flood someone else's inbox with confirmation requests.
Where to start
You do not have to wait for the first issue blind. On the newsletter page I compare six agent flows side by side, from prompt chaining to an autonomous agent. Each one has its own diagram and a list of the places where it breaks.
Signing up takes a few seconds. You enter your address, click the link in the email and you are done. And if an issue is not worth your time, just reply and tell me why.
