Systems in the Loop
In Systems in the Loop, I document what happens when AI becomes part of the day-to-day work behind a real project.
I’m interested in what still works once the novelty has worn off: which systems I keep, which I change, and which problems only become apparent when they’re put to use in actual work.
What you'll find here
I write about the systems I use to make AI useful beyond a chat window.
AI agents and the workflows around them
context, memory and knowledge systems
local models, permissions and human review
the experiments, mistakes and decisions that come out of building ArtMakers
Some posts will be fairly technical. Others will be about decisions, mistakes, or changes I made after realizing that an approach I liked on paper was not holding up in practice.
ArtMakers is my main testing ground, but Systems in the Loop is not an ArtMakers blog. It is also not tied to a particular AI tool.
I currently use Hermes Agent heavily because it fits the way I work. I also use ChatGPT, have experience with Claude and Gemini, and expect the tools to keep changing. I would rather focus on ideas and working methods that can survive those changes.
Why I’m writing this
ArtMakers did not start as a software project.
I first got into the NFT space as someone experimenting with art. An artist I met around a Solana NFT project encouraged me to try making something myself. I had no art background, but I sat down with my iPad and started drawing.
That turned into abstract work, collages and photography. I minted my work for several years and sold some of it.
During that time I ran into a problem that seemed to have very little to do with making art.
Getting people to see the work often took more time than creating it. There was X, Instagram, networking, community building and the constant question of what an algorithm would show to whom. I saw other independent artists dealing with much the same thing.
ArtMakers grew out of wanting to do something about that.
It was much smaller at the beginning. At one point I imagined something closer to a Discord bot. Then the ideas kept growing, the problems became more connected, and the project became considerably more complex.
AI gradually became one of the ways I tried to manage that complexity.
Eventually, simply opening a new chat and asking questions was not enough. I started experimenting with specialized agents, different responsibilities, project context, memory, review steps and ways of keeping a human involved when an AI should not make the final call.
Some of that worked well. Other parts introduced problems I hadn’t expected.
Systems in the Loop is my attempt to keep track of what I learn while working through them.
About me
I’m Chris Eberhard. I live in Berlin and I’m the founder of ArtMakers.
My path into technology has been somewhat indirect. I trained and worked in retail at Deutsche Telekom, later spent several years in hospitality, and completed a web development course in 2021 after becoming increasingly interested in building things myself. Much of what followed came through self-study and smaller projects. Today I study computer science at FernUniversität in Hagen.
I’ve been experimenting with AI for roughly five years.
I don’t come to this as an AI researcher, and I’m not trying to pretend that I have a finished framework for how everyone should work with these systems. Most of what I know has come from using them on something I actually care about, running into problems, changing the setup and seeing what happens next.
That is the perspective behind this publication.
How I make this
AI is part of the way I research, organise and edit my work, so it would be strange to hide that here.
I use it to explore ideas, challenge my assumptions, structure material and sometimes help with early drafts. The experiences, opinions and final editorial decisions are mine. I review and rewrite what I publish, and I do not want AI to smooth every uncertainty into a confident answer.
If my view on something changes after I have used it longer, I would rather write about that than quietly pretend I had it right the first time.
Who this is for
Systems in the Loop is for people trying to move beyond occasionally prompting an AI and make it part of how they actually build, research, organize or run something.
You do not need to use the same tools I do. In fact, I hope many readers don’t. The interesting part is finding out which ideas transfer.
