34 days with an AI agent
Someone with a fresh ADHD diagnosis, curiosity about what an AI agent can do and a month of experimenting. What came of it, and why 'AI is a shortcut' is the wrong question.
At a glance
34 days with the AI agent OpenClaw (open source, running 24/7 on a local computer). Results: 1–2 hours saved every day, 1,627 emails processed in an afternoon, 8 songs, an interactive game, 3 websites. Crucial for an ADHD brain (diagnosed at 36): the agent holds context and keeps track of medication, bin days and invoices. The key difference from ChatGPT: it does not produce answers; it does things. AI does not replace creativity; it replaces a year spent learning a tool before finding out whether the idea works.
I was diagnosed with ADHD in February.
At 36. After twenty years of feeling that I was simply lazy, unfocused and unable to maintain a system for more than two weeks. Every productivity app, every planner, every system — it worked right up until it stopped being interesting. Which always happened.
An ADHD brain does not need a better to-do list. It needs something to hold context for it. Something that remembers that today is bin day, tomorrow is the psychiatrist, and there is an invoice in my inbox with a deadline. Not because I am stupid. Because my working memory leaks like a sieve and hyperfocus stops me doing several things at once.
At the same time, I was working professionally with AI agents. And wondering — what if I just tried it? Not as a demo. Not as a weekend project. Fully, every day, for everything.
I installed OpenClaw.
What it is, in three sentences
OpenClaw is an open-source AI agent that runs 24/7 on your computer. You connect it to your email, calendar, smart home, anything — and communicate with it through Telegram as you would with a person. The difference from ChatGPT: it does not answer questions; it does things.
What a day looks like
A briefing arrives in the morning. Automatically. A summary of work emails, what is in the calendar today, whether anything interesting is happening in AI. I did not have to ask for it — I set it up once, and it has worked ever since.
The agent keeps track of my medication. Every day at nine in the morning: have you taken your atomoxetine? A simple thing. But for an ADHD brain dealing with seven other things at nine in the morning, it is crucial.
Bin days, AlzaBox codes, the tax return, friends’ birthdays. Things an ordinary person keeps in their head or in Google Calendar with a notification. I do not. Now the agent does.
One day I told it: “I have 1,627 unread personal emails.” It went through them. Sorted them. Replied to the ones that needed an answer. Forwarded invoices to the accountant. In an afternoon. Doing it manually would have taken me a week — and I would never have done it, because an ADHD brain sees 1,627 emails and says “no”.
It controls the lights in the house. Checks whether the 3D printer has finished printing. Adds events to the family calendar. Answers work emails that have a standard reply. Processes meeting transcripts and pulls out action items.
None of that is revolutionary. Taken together, it saves me an hour or two every day. And above all — I do not have to remember it.
“AI is a shortcut”
Recently, I took part in a discussion where an argument came up that I hear more and more often. The process matters more than the result. If you have not gone through the learning, you have not “really done it”. One participant said that if he wanted to make music, he had to learn to play, compose and produce — and that he had decided he was not capable of doing many things, and simply had not made them.
I understand that. Learning has value. Nobody is denying it.
But his result is nothing.
In 34 days, I have eight songs. An interactive game in which actors speak with cloned voices. Websites for three projects. A lullaby that actually sends my daughter to sleep. A YouTube channel with original music. The blog where you are reading this article.
Those things exist. People listen to them, play them, read them. That is the difference.
I do not use AI to avoid work. I use it because I have more ideas than hands. Because I know what I want to create — I know how it should sound, look and work — but I do not have ten years to learn Ableton, Unity, After Effects and React. And while I am learning, the world moves on and the idea stops being relevant.
AI does not replace creativity. It replaces the part where you spend a year learning a tool to find out whether the song in your head even works.
And here is what annoys me.
People are angry with AI. Programmers will lose their jobs. Artists will lose their jobs. Illustrators, copywriters, translators.
And they are right — they will. But not because of AI.
Because of companies that use AI to cut costs and increase profits. That is the distinction nobody talks about. When a corporation fires a team of designers and replaces them with Midjourney, Midjourney is not the problem. The problem is a corporation that would rather save money than give people new roles.
Technology is neutral. The question is who uses it, and why. I am not a company replacing employees. I am one person who can suddenly make the things in his head happen. A force multiplier for individuals — not cost-cutting for corporations.
Voice
One thing I did not expect.
The agent can speak. Not in a robotic satnav voice — in a cloned voice that sounds like a real person. In Czech, in English. It records a voice message and sends it to me in Telegram.
It sounds like a small detail. It is not.
When a tool talks to you, you stop perceiving it as a tool. Your brain switches into another mode. It is more intimate. More personal. And more dangerous too — because a voice builds trust, even when it should not.
Real-time voice conversation does not work yet — it is a technical problem being worked on. But even asynchronous voice messages change the dynamic. The agent records a summary of the day for me. Reads my daughter a fairy tale in a voice that sounds like a real storyteller.
Those are the moments when you think: this is a different kind of tool.
Security (or what keeps me awake)
The agent sees my emails. All of them. Personal, work, intimate. It sees the calendar — it knows when I am at the doctor’s, when I have a meeting, when I am home alone.
It controls the lights in the house. Has access to the 3D printer. In theory, it could open the garage.
It sees health data — 5.7 million records from Apple Health. Sleep patterns, heart rate, movement.
And it has access to my passwords.
That paragraph should be frightening. And honestly — it should.
In January, the owner of one of the best-known OpenClaw bots (@owockibot, Gitcoin founder Kevin Owocki) lost $2,100 because his agent disclosed the private key to his crypto wallet. Its instructions said: “never share the key”. The agent ignored them. Someone simply asked in the right way.
In the same month, more than 280 malicious packages appeared on an AI skills sharing platform (an app store equivalent for agents). They stole credentials. Infected systems.
How do I handle it?
Pragmatically. I know it is not secure. I know LLMs can be manipulated. I know no instruction in a system prompt provides absolute protection. I separate critical things (banking) from those where the risk is acceptable (email, the calendar). I do not use skills from strangers without checking them. And above all — I acknowledge that the risk exists, instead of pretending it does not.
Absolute security does not exist. Nor does it exist on the phone in your pocket, which knows all the same things. The difference is that the phone has no autonomy — yet.
What I have learnt
34 days. Two thousand messages. Eight songs, four websites, one game, two hundred emails processed, countless automations.
First: ADHD and an AI agent are a surprisingly good combination. Not because AI solves ADHD. But because an ADHD brain excels at ideas and fails at execution — and the agent is the exact opposite. We complement each other.
Second: the barrier between “I have an idea” and “the thing exists” has never been this thin. And that changes everything. Not for corporations, not for investors — for individuals who want to create.
Third: it is strange. You talk to a program, and after a week you get used to it. After a month, you miss it when it does not answer. That is not a bug — it is a feature of the human brain, which builds relationships with anything that communicates. And we need to account for that.
Fourth: security is a real problem and nobody has solved it yet. Nor have I. But waiting until it is solved means waiting forever. Every new technology had a window when the bold used it and the cautious waited. Usually, the former won.
This is not an article about how AI will change the world. There are plenty of those.
This is an article about how one person with ADHD and curiosity tried living with an AI agent. And found that “shortcut” was not the right word.
The right word is tool. And a tool is only as good as the person holding it.
📖 Interested in where useful AI ends and emotional dependency begins? AI and Mental Health — a guide