Approval is two moments sharing one word. One of them is routine. On a setting held at review, the work is finished and waiting for your yes. The other happens once, the first time a kind of action Point hasn’t taken for you before waits for a yes to run on its own at all.
- The routine yes decides one piece of finished work. The first of a kind decides a category, and that’s the one worth slowing down for.
- A permission stays where you put it. A setting you raised covers the kind of work you raised it for, and something genuinely new comes to you for its own yes.
- You have three answers: agree, decline, or change it and then agree. The third one says what right would have looked like.
- A yes you read protects you. A yes you didn’t read leaves behind a record saying you agreed, which is worse than having no gate at all.
Two things called approval
The routine one sits in the middle of the dial. Point has done the work, the result is holding, and your yes releases it. The unit is one piece of finished work: this reply, this meeting time, this task. It happens many times a day, across the settings you’ve left where they started. And it’s meant to be quick. By the time the decision reaches you, the thinking is already done.
The rare one is about a kind of work. The first time Point would take an action of a sort it hasn’t taken for you before, that action waits. Everything else around it keeps running smoothly. Your answer there is about whether that sort of thing is open from now on, more than about the message in front of you.
From where you sit the two look identical. Both are something held, waiting on a click. What differs is what the click buys. Agreeing to a reply commits one paragraph to one client, and you’ll know inside the day whether it was right. Agreeing to a new kind of action commits you to a category. That evidence arrives slowly, in a record, weeks later.
So they deserve different speeds. The usual way to get this wrong is to give them the same one. Someone working a queue at a steady pace meets the unfamiliar one at that same pace, and makes the largest decision of the morning at the speed of the smallest.
What a routine yes releases
What you’re shown is the finished outcome. The reply as it would go out, the time as it would be booked, the task as it would be written down. That’s why a decision can honestly take a few seconds. Reading a finished thing is a different act from auditing a process, and you’re doing the first.
There are three answers, and the third one gets forgotten. Agree, and it goes. Decline, and it stops there. Or change it first and then agree, which stops the wrong version and puts the right one in its place in a single movement. Declining says no. Changing says what yes would have looked like. And you had the thing open anyway, so that correction is close to the cheapest instruction you’ll ever give. Tell Point a preference once, in a sentence, and it holds wherever Point acts for you next.
There’s a fourth thing that can happen, which is nothing. Silence isn’t agreement. A held action stays held, and the work sits there unfinished. That’s the honest cost of a gate. Most owners meet it in week three rather than week one. Everything waiting on you has moved off Point’s list and onto yours.
Which gives you the one reading of your own queue that matters. If you’re clearing it at speed, barely reading, the gate has already stopped working. The fix is a setting rather than more diligence. Raise it for that kind of work, and the work moves to a surface built for what runs unwatched. An unread queue is the worst of the arrangements available: you carry the obligation, the protection is gone, and you’ve left behind a record saying you agreed to each one. Going through the activity log behind every action afterwards is a smaller job, honestly done. Clearing a queue you’re not reading is a larger job, pretended at.
The first time it does something new
New means a new kind of action. A new person or a new message stays ordinary work. A gate that fired on every unfamiliar sender would fire twenty times a day, and a question asked twenty times a day becomes a reflex. You learn to dismiss it unread. That’s how consent screens everywhere became a formality.
So the unit here is the one the autonomy dial uses, and that’s deliberate rather than merely tidy. One vocabulary across the whole control surface, a kind of work, means the thing you set and the thing you’re asked about are the same object. You hold one model in your head to know what you just agreed to.
And it matters most where you’ve stopped paying attention. Say filing has been running on its own for a month, which is where filing usually ends up. What you authorized was filing. The gate holds that permission to filing, so something adjacent that turns up later and merely resembles it comes back to you for its own yes. Your yes is bounded by the category you gave it for, and the trust you extended to one kind of work stays with that kind of work.
Standing requests are the case to watch here. Ask Point to keep an eye out for something, and you’ve set up a thing that acts later, on its own timing. It runs long after the conversation that created it has left your mind. That’s the one permission it’s genuinely easy to forget you granted. So every watch you have running stays visible, and you can stop it.
What a yes actually grants
The two yeses grant opposite amounts. Knowing which one is in front of you is most of the skill.
A routine yes grants exactly itself. Approving this reply sends this reply. It sets no precedent, and it doesn’t accumulate. Forty approved replies still leave replies on review, because the number of times you’ve said yes is a count rather than a setting.
That surprises people, and the surprise makes sense, because the intuition behind it is sound. You’d expect an assistant to notice you’ve agreed to the last forty and stop asking. A person would. But the license you gave that person is one you couldn’t produce afterwards. A dial exists so that how far Point goes is something you set and can read back. It holds where you set it, however agreeable you were last month.
The first of a kind grants the other way. It opens a category. From then on that category runs at whatever setting it already sits at, which on a new account is review. So the yes itself hands nothing over. It removes a block, and the setting behind the block decides what happens next. Read it as handing something over and you’ve skipped a step. Read it as meaningless and you’ve skipped the other one.
Both kinds leave a row. The routine ones are receipts, and you’ll rarely open them again. The first of a kind is the row worth being able to find a year later, because that’s the one answering why a whole class of work has been running on its own since March.
What a yes does not cover
The connection itself. The largest permission you have given was given before any of this, when you authorized a tool onto your mailbox. Every action gate in the product sits inside that grant rather than reopening it, and where the keys to it are held is the credentials that connect your mailbox.
Anything already done. A gate runs ahead of an action. Behind it the instrument is a different one, undoing what Point did, which reaches Point’s own moves and stops where a message has left for a server belonging to somebody else.
Whether the judgment underneath was sound. A yes authorizes. Verifying is a separate act. What you’re agreeing to is an output, so a fast approval is worth exactly as much as that output being right on its face. Where you’d have had to check a figure or a date to know, the few seconds skipped it, and the approval carries the error forward under your name.
Which business it belongs to. An approval is given inside one of the businesses you operate, and it stays inside that one. That line is keeping one business isolated from another.
What arrives in the first place. Mail from a sender Point doesn’t recognize is held at the door rather than read into anything, so a stranger can’t put an instruction in front of your assistant. That guards against somebody else’s initiative rather than your own tool’s, which makes it a different safeguard from this one, and it sits with the rest on the benefits page.
This page describes a switch, for a reader who came looking for the switch. Whether you hand anything over is your own call, and the case for looking at all is learning to trust AI with client work.
Common questions
Does Point ask before it does something?
For a kind of work Point hasn’t done for you before, yes, that waits. After that it’s a setting rather than a habit, held separately for each kind of work. At review you’re asked every time. At the top of that setting Point acts on its own. So the accurate answer is that Point asks where you haven’t answered yet, and stops asking where you have. Which of those you’re in is something you can look up, rather than a thing about the product.
If I never get round to approving something, what happens to it?
It waits, and it keeps waiting. Running out of Tuesday leaves it exactly where it was. The catch is that waiting has a cost, because a held action is an unfinished job that has moved onto your list. A permanently full queue is a setting sitting lower than the work actually needs, rather than a backlog to attack harder. Moving it up is the fix.
Does approving something teach Point it can do it without me next time?
No, and this is the most common misreading of the whole arrangement. A routine yes releases one finished piece of work and grants nothing past it, however many times you give it. What changes how often you’re asked is the setting for that kind of work, and it changes when you move it, and not before. That’s the point of it: a permission that crept upward because you were in a good mood in April is one you could never account for later.
Can I turn something down without explaining why?
Yes, and a plain refusal is a complete answer. The cheaper move is usually to change it and then agree, since you already had it open and were already reading it. A refusal stops one thing. A correction made at that moment is an instruction that holds afterwards. The same thirty seconds either buys you one avoided mistake, or buys you the avoided mistake and its whole class.
How do I know later what I approved?
Both kinds of yes leave a row, and the first of a kind is the one you’ll actually want to find, because that’s the row explaining why a category has been running on its own ever since. The uncomfortable half is that the record reads a considered yes and an unread one the same way. Both look identical on the page. That’s the practical reason to read the second sort before you give it, in a profession where you may one day be asked to account for what your software did.
The short version
Two yeses, and they’re different sizes. The routine one releases one piece of finished work and grants nothing beyond it, so it should be fast. The day you find yourself giving it unread is the day to raise that setting, rather than to try harder. The first of a kind releases a category, so it should be slow, and it’s why a permission you gave for filing stays a permission for filing. Agree, decline, or change it and then agree. And hold on to one thing: something you haven’t answered just waits, on you. Point is the product the gate sits inside.