Every kind of work Point does carries a level of its own. At the bottom, suggest-only: Point stays out of that kind of work. In the middle it prepares and waits for you. At the top it handles that kind of work outright, without asking. A new account arrives with everything sitting in the middle.
That is the fourth of the six questions, and it is the only one whose answer you write yourself. What gets read, what can leave and what gets written down are facts about the product, true before you touch a thing. This one is a decision, taken one kind of work at a time, and taken again whenever you like.
Which is where careful people go wrong, and not in the way they are braced for. The mistake is rarely that somebody raises a setting too early. It is that the whole arrangement gets read as one question about nerve, answered once, at the bottom, and never opened again.
Not a slider on your nerve
Picture what almost everybody pictures. A single control running from cautious to reckless, and moving it a statement about how far you have come to trust the software in general. On that reading there is nothing to do but wait until you feel braver, and feeling braver is not a thing evidence produces. Two years pass.
It is not shaped like that. There is no one position you are in. Filing has a level, replying has a level, scheduling has a level, and they have nothing to do with one another. A single sentence you type can straddle two: ask for a thread to be dealt with and answered, with the filing raised and the answering not, and the filing happens while the reply is prepared and left sitting there for you. One request, two levels, and Point is not in a mode at all.
That changes the question being asked of you, which is the useful part. You are never asked how much you trust this. You are asked, about one narrow kind of work at a time, whether you would rather do it, check it, or hear about it afterwards. That one has an answer.
Three places to put your check
The three levels are three places to stand in relation to an event.
At the bottom the work stays yours. In the middle your attention sits in front of the action: the thing is prepared, you look at it, it goes ahead or it does not. At the top your attention sits behind the action: it happened, and you find out by reading what was done.
The last chapter drew the line those three sit on. A check in front of an action can stop it. A check behind one can only tell you about it. Both are worth having, they do different jobs, and the level is nothing more elaborate than which of the two you have chosen for a given kind of work.
Put that way, a good deal of the fear drains out. Raising a setting is not removing a safeguard. It is moving one, from the costly side of the event to the cheap side. Whether that is a good trade depends entirely on the work, which is the next thing.
The middle position is not free
Review looks like the position with the risk taken out, and it is where a new account is handed to you for sound reasons. But it has a price, nobody prices it, and that is how people end up worse off than they would be at either setting on either side of it.
Something prepared and waiting is still an item. It sits somewhere, it wants a decision, and the decision falls to you at a moment you did not pick. For work that comes up twice a week, that is nothing. For work that comes up ninety times a day it is a second inbox, and the only thing in it is things you already agreed ought to happen.
You know what becomes of that queue, because you have watched it happen to every approval queue anybody has ever built. It gets cleared rather than read. By the fourth week you are going through it in blocks, at speed, on the sender and the shape of the thing rather than on its contents, which is precisely the reflex the first chapter named as the failure already running quietly in your building.
And that is worse than either honest position, for a reason worth being blunt about. A queue you clear without reading protects nobody, and it does one thing a raised setting never does: it produces a record saying you approved. You have taken on the whole substance of having checked while doing none of the checking, and when something goes wrong you cannot even say you did not see it.
So the safe position is not the low one. It is the one you will still be keeping properly in November, on a bad week, with a filing deadline on Thursday. A setting you comply with sloppily is worse than the honest setting either side of it, and that test alone will sort most of the rows without any reference to how you feel about AI in general.
You are raising a family, not a task
Levels are held per kind of action, and the kinds sit in families. Raise a family and everything beneath it inherits. Set the sorting and filing side of things to run on its own and all of it runs on its own, including the members you were not thinking about at the moment you moved it.
Worth slowing down for, because what you had in mind is not what you raised. You were thinking about the receipts. The family also holds whatever else counts as sorting, and that is not a list you carry in your head or can usefully guess at from outside.
Which gives the fortnight after any change a job to do. The record is where a family becomes visible, one row at a time, and reading it closely for two weeks is less about catching errors than about finding out what you actually agreed to. Most of what goes past will be dull. Dull is the finding. The row that is not dull is the one telling you the family was wider than the intention.
There is a second condition that is easy to miss. Asking Point for something in plain words is itself a kind of action and carries a level like everything else, so what you hand over that way runs on its own only when that has been raised too. Both have to be open. A raised setting somewhere in the tree is not a gap that anything can be pushed through, which sits alongside the strangers held at the door rather than replacing them.
What one notch too high costs
Now the thing the last chapter promised: the cost of a setting sitting one notch above where it belonged.
Inside your own account, less than you fear, and the shape of it is specific. Something was done that you would have done differently. You find it on a Tuesday, four weeks in, with a time against it and the reading it was made on. It has a location. The filing goes back, the level comes down, and nothing else in the arrangement is implicated, because the levels do not talk to each other. Turning one down is as cheap as turning it up. A setting you are living with is not a commitment you entered into.
Outside your account it does not have that shape, and that is why the row covering what you say to people earns the slowest movement of any of them. A message that went out under a raised setting is sitting on hardware belonging to somebody with no obligation to you, and nothing described in this series fetches it home. What you get is the record, quickly, which is worth a great deal on the day and is not prevention.
The other direction costs something too, and hands you nothing to look at. A setting left too low produces no event, no row, no complaint. It produces a year of approving things you had already decided were fine, and the benefit never arriving, and no moment at which anybody mentions it. Which is the same asymmetry that makes a blind spot expensive: the error that announces itself is the cheaper of the two.
The rows that stay where they are
None of which argues for the top as a destination. Some kinds of work are finished where they are, and leaving one on review permanently is a decision reached rather than a decision postponed.
The reason is not that the writing is weak. It is what the mailbox does not contain. Every judgment Point makes is made out of the correspondence, and correspondence is a partial account of a relationship: the call, the ten minutes in the corridor, the thing about that family you have known since 2014 and would never put in an email. Work whose correctness depends on the part nobody wrote down cannot be handed to something reading only the part that was. That is a fact about the document rather than about the model, and no better version of the software repairs it.
In a practice that ceiling stops being a preference and turns into something written down, since who may commit the firm to what has never been a matter of individual comfort. A firm’s own policy is where it belongs. Point holds the level per kind of action across everything a mailbox can be handed, and what each level does, task by task, is set out on the dial’s own page.
A setting that cannot see where it is
One thing this chapter has quietly assumed all the way through, and it is the assumption that breaks first for the people most likely to be reading.
Every level here attaches to a kind of work. Filing. Replying. Scheduling. That holds if a kind of work means one thing to you, and for most of a week it does. But you may not be one person with one working life. If you run the practice and something else besides, replying to a client is not one kind of work at all. It is a careful thing on one side of a line and an ordinary thing on the other, and no single setting is right for both unless something underneath it already knows which side you are standing on.
Which is a question about the wall between the two, and it is the fifth of the six.