My AI operating system
I have more things I want to try than evenings to work on them. Agents help, but explaining the next step to each of them can become another evening activity.
Now I give Hermes a marketing goal, and it works out what to do, including what to ask the factory to build. Here is how the parts fit together.
My actual dashboard, captured 15 October. Work and Personal keep separate knowledge and access. Task names are hidden; some activity isn’t tracked here. I’ll follow the Personal side below.
First, somewhere to keep the context
A few months ago I shared Pensiever, my second-brain setup. It quietly became my default operating system: somewhere to keep what I’m working on, what we tried, and why I made a decision.
This is an extension of that story. Context is useful when I can ask a question. It becomes more interesting when an agent can use it to continue the work, without another introduction to my life and all its unfinished projects)
Probably this is still the most useful part of the setup. New agents arrive quite often. My reasons for doing something should survive the change.
Last year, the evenings belonged to Receiptist
In 2025 I built Receiptist, a Telegram bot for tracking expenses, with Cursor. After 15 years away from coding, getting to a working product was quite an experience. I wrote about it in “I haven’t coded in 15 years, but AI helped me build this.”
A couple of years earlier I probably wouldn’t have attempted it. With agents, I could. But I still spent evenings deciding what to change, explaining it, checking results, reading the code (it was needed at least back then), and going back when something didn’t work.
So the agents did a lot of development. I did quite a lot of moving things between agents.
2025: evenings building with Cursor. 2026: marketing goals go to Hermes; product changes go from Hermes to Fabryka.
Now the question is: what should happen next?
Receiptist exists. Now I’m working on marketing: who it is useful for, where to reach them, and which experiment is worth trying.
Hermes is my personal assistant agent. Fabryka is my personal dark factory: agents doing development, checking the result and delivering changes.
I give Hermes a marketing goal. It makes a plan, works out what it needs to execute it, and if something is missing in the product, gives Fabryka the task. The factory delivers the change, Hermes checks the result and continues with marketing.
Get Andrei Matsko’s stories in your inbox
Join Medium for free to get updates from this writer.
Which means I’m discussing marketing with Hermes, and development happens as part of that work. I don’t need to turn every product gap into a separate evening with a coding agent)
I give Hermes the goal. Hermes makes the plan, requests product changes when needed, checks the delivery and carries on.
The Personal side, closer up
Actual dashboard, 15 October. Hermes on the left, Fabryka on the right, Pensiever underneath; some activity isn’t tracked here.
Hermes connects the marketing work with the product work. Fabryka does development. Pensiever keeps the context so the next discussion can start somewhere useful.
Sounds quite straightforward when written like this. Getting the parts to actually work together is another story)
The work comes back to Hermes
Receiptist’s original-image viewer is an example of the delivery part. I asked for it through Hermes in 2026. Testing found a broken image link; the factory repaired it and the next checks passed.
Hermes accepted the tested version on my behalf. The update reached the live product and the result came back to Hermes.
Small begining…
Next, another small product. Probably a few.
I’m planning to try mini-products through Hermes and Fabryka without my participation in development. Let’s see how far this goes)
I don’t know exactly how much time the whole setup saves yet. And getting people to use a product is its own work, of course.
The rise of intelligent agents
It’s still not fully backed, my (well, Codex’s) interventions needed, but already see that probably Muse and Dots of this world can enable all of that just with one ask :)
Probably, the part I want to keep is my own context and knowledge in Pensiever. Maybe that becomes the lasting part, while the agents doing the work keep changing.
