On the small moment when you stop checking, and why it always arrives on a full day.
There is a half second before you send a message you did not write, when something in your stomach pauses.
The text is accurate. The tone is right, the name is right, even the time it suggested makes sense. And still there is a catch in your shoulders and in your breath, half a second before your hand carries on.
In the first week you feel it every time. By the third week, on an evening with seven other messages in it, you do not. That moment does not arrive on a day you choose. It arrives when you are tired.
I run my whole practice with an AI assistant. It sees my calendar and my inbox, it prepares drafts before I get to them, and in the evening it writes down what actually happened that day. The three rules below were not written out of caution. Each one came from a mistake that had already happened.
Almost everyone who talks to me about AI describes the same thing. I opened it, I asked, I got a good answer, and next time I started from nothing. That is a chat, and a chat is a clever stranger in a lift who knows nothing about you.
An assistant is a different object, and there are only four differences. Memory, meaning one document that says who you are and how you work. Access to real things, a calendar and an inbox. Routines, things that happen without you asking in that moment. And rules, a list of what it may never do without you.
Notice what the first three create. The moment there is memory and access, there is a path for something to leave your desk without passing through you. Not a catastrophe, simply a situation that did not exist before. The fourth item is what makes the other three safe, and it needs writing before it is needed rather than after.
The tendency to lean on an answer that is already prepared, and to check it less than you would check a colleague, has a name in the literature: automation bias.
A systematic review by Kate Goddard, Abdul Roudsari and Jeremy Wyatt screened close to fourteen thousand papers and analysed seventy four of them. Two of its findings matter here. What increases over reliance is workload and time pressure. What reduces it includes clearly assigned accountability, meaning a structure where it is obvious who signs off at the end.
On the source and its limits: the review looked at clinical decision support systems rather than personal assistants, it was published in 2012, more than a decade before the models we are discussing now, and a large share of the studies inside it were run in simulated settings rather than in live practice. I read the abstract and not the seventy four studies, which is why there is no percentage anywhere on this page. Carrying the finding across to a one person practice is my reading, not a result. Goddard, Roudsari and Wyatt, JAMIA, 2012.
Translate that into your own week. You will not stop checking on a day you decided to be careless. You will stop checking on the day with seven sessions in it, which is exactly why the rule belongs in writing rather than in memory.
Rule one. Nothing goes out without me. It drafts, you send. Always, including when the draft is perfect and the step feels like a formality. This is the only rule I ask people to copy word for word, because it puts the signature back in your hands, which is precisely what the review above calls accountability.
Rule two. Other people's material stays out. The name of someone who sits across from you, what was said in a session, a diagnosis, a case description. No. The assistant runs the business around the work: enquiries, scheduling, money, content and the shape of the week. It does not run the work itself. If you hold a licence, read what your own code says before widening this, and do not take my article as the authority.
Rule three. It gets things wrong. Not often, but it does: an invented detail, a missed context, and the same confident tone whether it is right or not. So every number and every name gets checked before it leaves, the way you would check behind someone in their first week. Not out of suspicion. Out of order.
Your rules page, fifteen minutes, written once
One page, plain language, the way you would explain your work to someone joining you on their first day. Not a policy document.
When to stop: if you have more than six rules, stop there. A list you cannot recite from memory is no longer protecting you, it is only reassuring you.
A good assistant does not decide for you.
It makes sure the decision reaches you while you still have the energy to make it.
There is one test, and it does not happen in a good week. It happens in the week where you were late to two sessions and twenty messages are open. If everything that left your desk still passed through you, the rules are holding. If something went out and you cannot remember approving it, that is not carelessness. It is a rule that lived in your head rather than in a file.
There is also the other side. Some people write four pages of rules and then never start, because the system became too heavy to touch. If that is you, three sentences are enough, and the only way they improve is by being used.
It costs money. There are free tiers to begin with, and serious use needs a subscription. The only test I know is what one hour of your own work is worth.
And order does not cure burnout. If your real load is emotional rather than organisational, a good assistant will simply make you more efficient inside the same difficulty. That distinction took me a while to see in my own week, and it is the first thing I look at when someone tells me they are drowning in admin.
No. The rules page is written in plain language, and the first step is an ordinary chat window that you paste it into. Connections to a calendar or an inbox come later and are made with approval clicks.
In my practice the answer is no, and it is a rule rather than a preference. If you hold a licence or work under a professional code, read what your own code says before widening this.
Memory, access, routines and rules. The fourth is what makes the first three safe, and it is the one most people skip.
The page takes fifteen minutes and the first task can run the same evening. The only measure worth using is whether one recurring task moved out of your head and actually happened.
Or book a free 30 minute intro call, no pressure.