I wanted a JARVIS

I wanted a JARVIS. The one from Iron Man. An assistant I could talk to naturally, ask to do things and leave to get on with them. One that could help me manage the house, organise my day and move my projects forward. One that remembered what we had discussed and told me how everything was going.

The suit of armour could wait.

What appealed to me was that familiarity: Tony Stark speaks and his assistant knows what he’s talking about. There’s a conversation in progress, tasks waiting to be done and a shared history. That was what I wanted to bring into my everyday life. Quite an ambitious idea, when you stop to think about it. Say “I want a JARVIS” and the whole thing fits into four words and sounds incredibly easy.

A morning at 06:30

This was how I pictured it: it’s 06:30 and I have a summary ready of the news I’m interested in and the work I’ve asked the assistant to do overnight. While I make coffee, I go over my schedule and ask for some music. I know what’s finished, what’s still pending and where I need to make a decision.

Then I leave home and carry on the conversation from my phone or my car. Back at the computer, I pick up the same subject. The assistant keeps track of the conversation. That was how I imagined a day with my JARVIS.

That continuity mattered more to me than any spectacular demonstration. My day takes me between places and screens. I wanted the assistant to follow along. An idea I mentioned on the phone should still be there when I got to the computer, together with the instructions and decisions we had made.

I was used to chat

My starting point was using artificial intelligence in a chat. Write something, get an answer and keep talking. That was the format I knew. Having an assistant of my own, able to remember things and take care of tasks, was something I still had to learn about.

I didn’t know what a bot was. Nor did I know how to create one or teach it to do what I wanted. I had a fairly clear idea of what I wanted done and very little idea of how to turn it into something that worked. A wonderful combination if you want to start making life difficult for yourself.

I’m not a software engineer. I learn by doing. And in this case, the question driving me was very specific: how far could I get with AI in building that assistant?

The first step was to explain clearly what I expected of it. The word JARVIS gave us a shared starting point, but it had to come down to ordinary things: remember an instruction, carry on with a project, finish a task and let me know. That was where it started to get interesting.

Remembering what mattered

Memory was at the centre of everything. I wanted to be able to tell it something and have that information remain useful later. My projects, my preferences, a decision I had made, a correction. All of that was part of the work I expected to share with the assistant.

I cared particularly about what happened between one conversation and the next. A task can be left half-finished. An idea can need several days. I wanted to return to it without having to reconstruct from scratch what we were doing, why and what I had already asked for.

For me, remembering included knowing what had changed. If I corrected an instruction, that correction had to stay with us. If I chose a direction, the assistant needed to keep it in mind next time. Storing words was only part of it; being able to carry on was what made it useful.

And I wanted to carry on from any device. An instruction given as I left home had to keep making sense when I came back. I wanted to find both the task I had set and the progress made on it. The conversation belonged to the project and to me. Switching screens shouldn’t force me to start a new relationship.

That was what I meant by persistent memory. Being able to come back tomorrow and pick up where we had left off.

I wanted a personality too

I wanted to recognise my assistant by the way it spoke to me. A conversation partner with a name and a recognisable style, someone I could address naturally. Personality was part of the dream from the start.

For me, that familiarity had practical value. If I was going to share ideas, hand over tasks and come back to ask how they were going, I wanted to feel comfortable in that conversation. I wanted to build a way of working together that I could sustain over time.

The division of work I had in mind was simple to explain: I supplied the ideas and the drive; the assistant had to remember, organise and carry things out. I could change my mind, ask questions or adjust the course. It needed to keep enough context to follow me.

And keep me informed, too. Because asking for something and then waiting around to see what happens isn’t much fun, however much personality the other party has.

Getting things done and keeping me posted

I wanted to hand over a task and get on with something else. Even keep talking while the work moved forward. Afterwards, I needed to know what had been done, what was left and which decision was mine to make.

That was an essential part of my JARVIS: giving me the work back in an organised form. Finding the result, understanding its status and being able to decide the next step. That was precisely why the morning summary made sense. It gave me a way to return to unfinished business without chasing each item separately.

That wish was the start of the journey towards what I now call La Machina. The project began on 7 September 2026. The story will grow to include five main bots I talk to, with other helper bots behind them, apps of my own and a control panel. But at this point in the story, I was facing something much smaller: my first bot on a platform called Hermes.

Its name was Alonides.

The next episode starts there: learning what a bot was, how to create one and how to teach it what I expected of it. I wanted a JARVIS. Time to start with the first bot.

La Machina in this episode
Starting point: I used AI in a chat and didn’t know how to create a bot.
Goal: an assistant with memory, personality and continuity across devices.
What’s next: my first bot, Alonides, on Hermes.