Auto Approve vs a Button You Press
Everybody running a coding agent all day ends up choosing between two costs. Either you keep answering the approval prompt, or you switch it off and ship whatever the model wrote. Our Slam Buttons only make the first cost smaller. They cannot make it disappear, and it is worth being clear about that before we compare anything.
What are the two costs, honestly?
The cost of approving is interruption. Not the keystroke, which is nothing, but the fact that it lands while your hands are on the mouse or mid sentence, over and over, all day. It is the kind of small tax that does not show up anywhere except in how tired you are by the end of the day.
The cost of not approving is that the last human check on the agent's actions is gone. Not usually catastrophic, sometimes free, occasionally very expensive, and the important part is that you will not know which kind of day you are having until afterwards.
What is auto approve genuinely good for?
Disposable work in a contained place. Scratch repositories, throwaway branches, sandboxes, spikes you plan to delete, dependency installs that would otherwise ask you the same question again and again. If you can recreate the whole environment quickly and nothing outside it is reachable, the review is worth less than the interruption and turning the prompts off is simply the right call.
We will say the unhelpful commercial thing plainly: if that describes your work, you do not need our hardware. You need the setting, it is free, and a pedal would just be a nice object on your floor.
What is a button genuinely good for?
The case where you have decided to keep reviewing. Real branches, shared repositories, anything with credentials or a deploy at the end of it. There the prompts are doing a job you want done, and the only thing worth optimising is where the press lands. A foot pedal such as Pedal Three at $69, or a pair of arcade buttons like Slam Buttons at $29, takes accept and reject off the keyboard, so answering stops competing with your hands.
The honest caveat is the same one we put on every page about this: making yes the easiest movement on the desk also makes waving things through easier. That is a thing to notice deliberately, not a thing to discover in a post mortem.
Is there a third option?
Yes, and it costs us the sale too. Let the agent run to completion in a contained environment with the prompts off, then review the entire diff once, at the end, the way you would review a colleague's branch. You keep a real review, you take no interruptions at all, and the only thing you give up is the chance to catch a wrong turn early. For well specified tasks that trade is often the best of the three.
How do you choose?
One question. If a wrong action could reach something you cannot recreate in a few minutes, keep the prompts and make the press cheap. If it cannot, turn them off and stop paying the interruption for nothing. Most people need both answers on the same day, in different repositories, which is why the setting is worth understanding rather than setting once and forgetting.
If you land on keeping the review, the physical side is written up in a physical button for approving what your agent just did.