Insights

Being your AI's boss shouldn't be this tiring

Roseau team · Oct 1

You handed the task over a while ago, but you never walked away. You’re still at the screen, watching it unfold step by step: which file it’s reading, which approach it’s weighing, which line it’ll change next. You’re like a supervisor who doesn’t quite trust the new hire. But getting up feels wrong too, as if leaving something this capable sitting idle is a waste. By the end of the day, a lot of work got done. You’re more tired than you used to be.

A lot of people have felt this this year. It isn’t because the AI isn’t smart enough. It’s because we haven’t worked out who we’re supposed to be when we work with it.

You’re already the boss

The moment you ask an AI to carry out anything, you’re treating it as an employee. That gives you a new role: the boss of your own AI. And a boss isn’t there to supervise. A boss is there to review. You make the task clear, you step away, and you check the result when it comes back. Direction and decisions belong to the boss; execution and the legwork belong to the employee. Draw that line, and you’re actually running things.

It sounds simple. In practice, people fall into one of two opposite traps. One is underestimating the AI: watching every step, ready to take over at any moment, until you’ve become its supervisor and repair crew. The other is overestimating it: letting it run before you know where you want to go, and not knowing what went wrong when it ends up somewhere else. A real boss sits between the two. Let go of the execution. Keep hold of the direction.

The tiredness is about decisions

There’s another kind of tiredness, one layer deeper than watching the screen. It used to take a lot to turn an idea into something real. Research, finding the right people, writing, pushing others along: one thing could take days. That was hard work, but the thinking happened quietly inside it. You figured things out while you were doing them.

Now AI takes the execution away, and what’s left is decisions. Do it or not. Which way. What’s next. A judgment you used to make once a week now comes around every fifteen minutes. Without execution to cushion it, the thinking is exposed.

So the fatigue isn’t from doing more. It’s that the most demanding part, deciding whether to keep pushing forward, is now everywhere. That part can’t be handed off. But it doesn’t have to arrive this often, in this many pieces.

So there are two kinds of tiredness: one from having to watch it, one from decisions coming too fast. A good personal AI should take both down.

A good assistant lets you walk away

Start with the first. A good human assistant doesn’t make you watch them work. When they’re done, they come find you. When you ask how they got there, they can tell you. They already know the background, so you don’t explain it every time. Many agents today hand exactly these things back to you.

To actually walk away, you first need to be told when it’s done. Waiting for results shouldn’t mean refreshing a page to check progress. It should mean the work comes to you. What you’ll want to look back on waits for you; only what needs your response interrupts you.

You also don’t need to watch it think. A boss doesn’t need to know which files an employee opened or how many drafts they tried. If you want to know how it reached a conclusion, you ask when you review. And it has to be able to answer: which email, which note, what you said last month. If it can’t, it should say it doesn’t know. Being able to ask is what lets you stop watching.

And the background shouldn’t fall to you either. Who this client is, what you promised last time, the principle you wrote down: it should already know. If you’re looking at an email or a brief, you should be able to ask right there, without copying it somewhere else and explaining what you were looking at.

Decisions should come in batches, at the right time

Now the second. Once execution and management are off your plate, the decisions that remain shouldn’t arrive every fifteen minutes either. The small things should follow rules you’ve already given. How the daily brief should read, which emails don’t need a heads-up: you say it once, and it follows it, without asking again each time. When it’s genuinely unsure, it can ask you a quick question. It shouldn’t turn every step into something that needs your nod.

The things that really need your call should be gathered into a brief and put in front of you when it’s time to decide: what came before, what’s fact and what’s a guess, what’s still missing, where the conflict is. Then it stops and waits. Say that on the same afternoon a client pushes back on price, a supplier moves a date, and a meeting shifts. You shouldn’t get three interruptions. You should get one brief before your next free stretch: what each one changes, what it suggests, and what it can’t judge. Any email or calendar change it has prepared for you waits for your go-ahead before it goes out. A kind of task you’ve handed over in advance, it can do as you instructed. What it won’t do is decide on its own that you would have agreed.

That way your day isn’t dozens of small “should I?” moments. It’s a handful of calls that are actually yours.

The part you can’t hand over

AI will keep getting more capable. But there’s one thing it can’t give you: wanting. It doesn’t come up with an idea on its own. It doesn’t care whether the result is good. It won’t change course because a direction matters more. Trace any step it takes far enough back, and the source is a person. What you haven’t thought through, it can’t help with, or it helps in the wrong direction.

So what’s left on a boss’s desk should be small: whether this is worth doing, which way to go, and who answers for the result. Those are hard, and they shouldn’t be optimized away. A good assistant doesn’t do them for you. It takes everything else, so you have the energy left to do them.

That’s what we’re building Roseau to do: take everything else. When work is done, it comes to you, and only what matters interrupts you. You can ask it why at any point, and it can tell you where that came from. You can ask from wherever you are, without restating the background. And what needs your call arrives as a brief, and then it waits.

Then you can actually get up from that screen. Execution can be handed over. The person who decides where things go is still you.