Mario De Luca on ESIS: Why I Built an AI Operating System for My Company at 22

Why I needed an operating system in the first place

I'm not a computer scientist who got into this because I love tech. I come from sales, and at 18 I failed to build anything four times in a row: dropshipping, trading, memecoins, my own fashion label. What I took from that is in my piece on those four failed attempts. The short version: I collected tactics instead of building systems. Every new idea was a new trick, and every trick depended on me.

As the company grew, the same problem came back in a different shape. Information was everywhere: in chats, in calendars, in notes, in people's heads. Who reached out to which prospect? What happened today? What's on tomorrow, and what do I need to know before it? I started spending my days searching. That's the point where you either hire more people to search, or you build a system that makes searching unnecessary.

I chose the system. Not because people are replaceable, but because people should be having conversations, not doing copy and paste.

What ESIS is – and what it isn't

ESIS is an always-on machine, a Mac mini in the office, where several services work together. I built it with Claude and Claude Code (more in Claude for business, German). The difference from a chat window: a chat waits until someone asks. ESIS works without anyone asking and speaks up when something matters.

What it isn't: an employee who makes decisions. ESIS prepares, summarizes and reminds. People decide.

What ESIS does every day

1. Listening in, read-only

ESIS is part of my team's WhatsApp groups – as a reader. It doesn't post on anyone's behalf, and it doesn't butt into conversations. That was a deliberate decision: a team that feels a machine is talking over them will eventually move the conversation somewhere else. And then the knowledge is gone again.

2. A daily report nobody writes by hand

In the evening, ESIS sums up what happened during the day. Not as a novel, but as what I need to know in the morning: what's done, what's open, where something is stuck. In my experience, anything that has to be tracked by hand eventually stops being tracked. Anything that shows up on its own gets read.

3. Flagging new leads instantly

When a prospect comes in, ESIS flags it right away. In sales, speed decides whether someone is still warm or already talking to the next person. The conversation itself is handled by a human – more on that in my piece on emotion in sales.

4. The calendar at a glance

Which meetings are coming up, what's new, what got moved? ESIS sums it up so nobody has to open five calendars to know what the day looks like.

5. A memory that never quits

At the core is what I call the "Brain": a knowledge base in Obsidian that serves as the company's memory. Decisions, processes, lessons, project status. ESIS draws on it whenever it summarizes or answers. The idea behind it is simple: knowledge that lives in one person's head goes on vacation with that person. Knowledge that lives in the company's memory stays.

The development team made of agents

The part that surprises people most: ESIS has its own development team of AI agents that builds software. I describe what's needed, the agents build it, and at the end a human decides whether it stays. What an agent like that is and what it can do in a business, I've written up in detail in AI agents for business (German).

I measured what it costs. Two software jobs the team completed came to 52 and 114 US dollars in model costs. That says nothing about what any given piece of software costs – scope and standards decide everything. But it shows the order of magnitude building has moved into, and it's the reason I measure cost per job instead of guessing. How to calculate costs like that honestly is in What does an AI agent cost (German).

Why the voice runs locally

ESIS has a voice, and it runs locally on the machine. There are two reasons. First: what's said inside a company belongs inside the company. The less data has to leave the building, the less you have to explain, secure and hope. Second: a system that depends on the outside goes down when something outside goes down. So wherever it makes sense, I build locally.

Three decisions I'd make again

• Reading, not writing. ESIS observes and reports; it doesn't speak for people. Trust inside the team matters more than an automation rate on paper.
• One memory instead of ten places to file things. Everything important ends up in one place. Ten half-maintained tools are worse than one knowledge base people actually use.
• My own build instead of a pile of subscriptions. I understand every part because I built it. When something breaks, I know where to look.

What I learned

The most important lessons weren't technical. The first: more notifications don't mean more overview. A system that's always saying something eventually gets ignored. A good system says little – but when it does, it matters. Getting that right is harder than hooking up any API.

The second: a memory is only as good as its upkeep. If nobody records decisions, even the best AI can't find them later. The machine takes the typing off my hands, not the thinking.

The third: automation without measurement is just hope. Only since I started calculating what a job costs and what it saves do I know which parts are worth it and which just look good. If you want to go deeper, the basics are in AI automation (German).

Who something like this makes sense for

An operating system of your own isn't worth it for everyone. It makes sense when a team works through chats, when information is scattered and when the boss spends more time searching than deciding. It makes no sense when there are no repeatable processes yet – then you're just automating chaos.

For companies that don't want to build something like this themselves, there's NextCore: we build and run AI agents and automations for businesses, as projects and with monthly support. If you'd rather build it yourself, my piece on the Clone Looping System describes the path I lay out for that.

Why at 22

My father is over 60 and still working. He built an ice cream café in Germany without speaking German, others destroyed it out of envy, and after that both my parents worked full-time. I build so that he can stop. That's the whole reason I'm not waiting for the right moment. A system that frees up time isn't a toy to me – it's the shortest way there.

And ESIS isn't the end, it's the foundation. Jarvis, my own AI clone, is built on top of it too.