The list of what you will NOT automate in your pharmacy

Reading time: about 10 minutes

The situation. You have reached the end of the level. You have sheets that check themselves, reports that arrive and an agent running. And from here on, every week you will think of one more thing that could be automated.

This is deliberately the last case, and the only one that builds nothing. It is a written list — half a page — of what in your pharmacy a person decides. And it has to be written now, while no specific temptation is on the table, because a boundary written with the temptation in front of you is not a boundary any more: it is a justification.

What you need to hand: Half an hour and honesty. This case has no technical check: the check is whether in six months the list still says the same thing.

Step by step

The four that are not negotiable and need no thought

Start with these, which are a given and hold for any pharmacy:

Pharmaceutical advice. What you recommend to a specific person in front of you. It is the professional act, it is yours, and you answer for it.
Referral to a doctor. Deciding this is not for the pharmacy is probably the most valuable decision you make all day, and it depends on things no data holds: the face, how long it has been going on, what they are not telling you.
What somebody is told about their treatment. A text can help you explain it better. It cannot decide what is explained or to whom.
Anything written into a clinical record or an official register.

These four are not my opinion: they are the perimeter of the professional act. Write them down anyway, because a list that starts with the obvious is easier to continue.

Yours, which are the hard ones

Now the ones that depend on your pharmacy and on how you want it to be. Nobody can give you these and they are what really defines the list.

Think of three specific places: what a customer is told (does a person answer messages or can an automation?), what affects the team (is the rota proposed by a machine or by somebody who knows whose mother is ill?) and what commits money (up to what amount does something get bought on its own, if at all?).

There are no correct answers here and not every pharmacy answers the same. What does not work is not having thought about it and finding out the day it is already built.

Use AI to argue with the list, not to write it

Ask it for the list and you get a reasonable, generic one that is not yours and that you will not respect because you did not think it through. What it does do well is find the holes in the one you wrote.

I have written the list of what I will NOT automate in my pharmacy.
Here it is. Do NOT rewrite it and do NOT give me another one.

Paste your list here

What I want from you:
1. Tell me which points are written so vaguely that in six months I will
   be able to read them however suits me.
2. For each point, describe the concrete situation in which I would be
   tempted to break it. The most realistic one you can think of.
3. Tell me what things in a pharmacy are NOT on my list and probably
   should be. Just name them, do not decide for me.
4. Point out whether any of my points is really TWO different things
   mixed together, one automatable and one not.

Be uncomfortable. A list that does not survive this review will not
survive the first busy week.

Point 4 pays best. "I do not automate customer service" is two things: telling somebody an order arrived (which you do, and already automate) and answering somebody asking about their medication (which you do not). Unseparated, the rule cannot be followed and ends up ignored entirely.

Put it where it gets read, not where it gets filed

A list in a drawer has never helped anybody. In the back room, printed, where whoever works with you sees it too.

And that last part is not decoration: the list belongs to the team, not to you. The day somebody builds something with an AI — and it will happen, with or without you — the question "is this on the list?" only gets asked if the list is known.

If you have the induction protocol from the block 2 case, that is its natural home.
The rule that keeps the list worth something

A list of prohibitions that can never change gets ignored the moment it is inconvenient. So it can change, on one condition: it changes cold, not hot.

Specifically: never on the day it would suit you to change it. If on a Tuesday you find that dropping one point saves you two hours, that Tuesday it stays. You note it, and you decide another day, calmly.

It is the same idea as an agent’s turn cap: a constraint is worth something precisely because it applies just when you feel like breaking it. One that gives way at that moment was not a constraint, it was an intention.

And give it a review date — twice a year is fine. Not to loosen it: to check it still says what you mean, now that you know a good deal more than when you wrote it.
And what this list gives you, which is not caution

It may look as though a case ending in "not this" is a brake at the end of a level devoted to automating. It is the opposite.

Having written down what does not get touched is what lets you automate everything else without thinking twice. Somebody without the list hesitates over every idea, because every decision is new; somebody with it just checks whether it is inside or outside and moves on.

And there is one more thing, and it belongs to this job: what your pharmacy has and a machine does not is exactly what is on that list. Automating well means clearing away everything else so there is more time for that.

A full example, and what is wrong with it

A list written by an owner in twenty minutes, run through the step 3 prompt. This is what the model returned — and the interesting part is not the list, it is the review.

What it replies

On "I do not automate customer service": "That is written very broadly. It covers things you already automate — telling somebody an order arrived — and things you never should — answering somebody asking about their medication. As written, either you break it daily or it forbids something you already do."

On "I always review anything leaving the pharmacy": "What does 'review' mean? Reading it fully, or glancing at it? In six months, in a hurry, 'I looked at it' will count as reviewed."

What is missing and should be there: "Nothing about who may build automations. Nothing about what happens when one stops working. Nothing about employee data, only patient data."

The most realistic temptation: "A Friday in August, a person short, fifteen prescriptions pending and an unanswered message from somebody asking whether they can take ibuprofen with their treatment."

And now, what is wrong with that reply:

This is not a failure: this is the case working. Worth saying, because it is the one example in the school where the model’s reply is better than what you gave it. And for a specific reason: you did not ask it to write, you asked it to find holes — and that it does well, because it does not require knowing your pharmacy.
Point 4 pays best, and you can see it here. "I do not automate customer service" is two different things mixed together, one you already automate and one you must not. A rule broken daily stops being a rule within a fortnight, and it drags the credibility of the other five down with it.
And the temptation it describes is the right one. A Friday in August, a person short and fifteen prescriptions. That is the situation where the list breaks, and it is not hypothetical: it happens every August. Knowing in advance what the moment looks like is what lets you recognise it when it arrives, which is half of being able to hold the line.
The one thing to watch: that it does not write the list for you. The prompt says three times not to rewrite it, and even so it sometimes closes with "here is an improved version". If it does, ignore it — not because it is badly written, but because a list that has not passed through your head is one you will not honour on that August Friday.

Half an hour: twenty minutes writing and ten reading the review. What comes out is a shorter, more concrete list with two points you had not thought of.

And what makes it worth something is not the content: it is that you wrote it cold. The same list written on that August Friday would have been a different one — and that is the entire reason this case is the last of the level and not the first.

And if your pharmacy is not like that

If you work alone with no team to show it to. The list still matters, more so: with nobody to ask "is this on the list?", the only brake is you in six months. Pin it up where you see it anyway, and give it a review date — in your case the calendar plays the second person.

If an external provider builds things for you. Then the list is for them too and it is worth sending. Not as an imposition: as information about what can be proposed and what cannot, which saves them work and you arguments. The automations that end worst are the ones somebody built without knowing where the line was.

If you feel you automate too little to need one. Then it is the best moment. A list written when nothing is built is a decision; written when six things are already running it is a justification of what you already did. And if you end up automating nothing, it cost you half an hour.

If somebody on the team has already built something alone. It happens, and it is not bad news: it is initiative. What to do is not ban it but bring it into the conversation — have them show what they built and check it against the list together. An automation somebody hides for fear of being told off is far worse than one that should not exist.

When it does not work first time

I end up with a twenty-point list.

Too many: a twenty-point list is neither remembered nor consulted. Merge the ones of the same kind and settle on six or eight. If one does not fit any group, it was probably not a point but an example of another.

I cannot tell whether a specific thing is on the list.

That is a sign the point is badly written, not that your judgement is lacking. Rewrite it until the answer is obvious in ten seconds. A point that needs interpreting will not be applied in a hurry — and a hurry is when it is needed.

The temptation came and I broke the rule.

It happens, and what matters is afterwards: note it, and decide cold whether the list was wrong or you were. Both answers are legitimate and both are useful. What does not work is never revisiting it, because then the list has broken silently and you will not honour it again.

This feels like paperwork for its own sake.

If it does, write it in five lines and be done. The value is not the document, it is having thought it through once without pressure: that is what makes you recognise the decision when it arrives in six months instead of improvising it. Half a handwritten page in the back room does the job.

Before you use what it gave you

The list is written down, not thought about.
Each point is one thing, not two mixed together.
No point can be conveniently reinterpreted in six months.
It is where the team reads it, not filed away.
It has a review date.
You wrote it: the AI only looked for the holes.

It writes you a beautiful list if you ask. Which is why you do not ask for one. A generated list reads well, gets filed and is not respected, because it never passed through the one head that has to agree with it.

It tells you everything is fine with supervision. It is the comfortable answer and it is nearly always false: supervision of something that works ninety-nine times does not exist on the hundredth. Some things are not automated with supervision, they are done.

It talks you out of a point as excessive. Ask "is this too strict?" and it will give you reasonable arguments for relaxing it — because that is what you asked for. Do not ask that. Its job here is to find holes and describe temptations, not to opine on where your line goes: that is not decided by whoever does not answer for it.

Now, with your own. Write it today, even if it is five lines and even if it feels obvious. In six months, when you have four things running and the fifth occurs to you, those five lines written cold will be worth more than the rest of the block put together.

← All practical cases Back to the school →

The list of what you will NOT automate in your pharmacy

Step by step

The four that are not negotiable and need no thought

Yours, which are the hard ones

Use AI to argue with the list, not to write it

Put it where it gets read, not where it gets filed

The rule that keeps the list worth something

And what this list gives you, which is not caution

A full example, and what is wrong with it

And if your pharmacy is not like that

When it does not work first time

Before you use what it gave you