For most of software’s history, the limit on what you could ship was your hands: how fast you could type, how much you could get built before the day ran out. Agents lifted that limit. A developer can now run several streams of work at once and get through more in an afternoon than used to fill a week.
But lifting one limit just exposes the next one. What’s left as the constraint is the part that was always human: deciding what to build, judging whether the output is right, and catching which of the parallel tasks is about to go wrong. Execution got cheap. Attention got expensive.
That trade is easy to miss because it doesn’t show up as more hours. It shows up as mental fatigue. I’ve ended more than one of my most productive afternoons completely fried, more wrung out than after a full day of writing every line by hand, because the machine did the typing while I spent the whole time deciding and checking. Cognitive load is the real constraint now, and like any constraint it’s worth managing on purpose.
Automate the Decisions That Don’t Matter
Not every decision is worth the same, but they all spend from the same account. Naming a variable, approving the same harmless command for the tenth time: each is a small withdrawal, and by mid-afternoon the account is empty whether it went to architecture or to indentation.
A line from President Obama has always stuck with me:
“You’ll see I wear only gray or blue suits,” [Obama] said. “I’m trying to pare down decisions. I don’t want to make decisions about what I’m eating or wearing. Because I have too many other decisions to make.”
The same logic holds for code: a low-stakes call made twice is one that should have been automated the first time.
The mechanisms are mundane. A lint rule settles a style question for good. A default in a builder function settles a construction question. A line in an agent’s instructions settles a decision once and lets the next ten tasks inherit it. What you can’t automate, you can at least write down: task state, a decision log. Anything held only in your head is something you’ll pay to reconstruct later.
Plan for It Like You Plan for Points
Teams estimate time. They estimate complexity. They point stories. They often don’t ask the obvious next question: how much attention a piece of work will take, and whose.
Cognitive load concentrates, and it’s easy to miss. I’ve seen teams struggle with a sprint, not because anyone was over capacity on hours, but because one person’s lane was stacked with work that each piece demanded their full attention to do. Three “small” tasks that each require holding the whole system in your head at once aren’t small.
The fix is to account for this cognitive load during planning. Ask whose head a stream of work lives in, and whether that head already has too much in it. When one head is overloaded, spread the knowledge around so the load can be shared: pair on it, write down what only that person knows, and hand pieces to teammates who have room.
Take Breaks, Because They Used to Be Free
Easy execution work used to double as rest. Knocking out a batch of simple tickets or wiring up boilerplate was the cognitive equivalent of a walk around the block between hard problems. It gave the brain somewhere to turn on autopilot.
Agents are now taking that work. What’s left is a denser stream of judgment calls with the autopilot stretches removed, which means the break now has to be deliberate: get up, walk, get coffee away from the screen. A deliberate break isn’t slacking, even though a quiet keyboard looks like it.
But it pays off. A review done after ten minutes away catches what a tired reviewer waves straight through.
It’s the Same Bucket at Home
The account doesn’t refill when you close the laptop. The attention you spend deciding things all day comes out of the same reserve you bring home, and the decisions waiting there don’t care that you’re already tapped out. I have three little kids. Most weeknights the hard part is the sheer volume of small choices stacked end to end, the same shape as a backlog of low-stakes tickets.
So the same moves carry over. We settle the week’s pickups and drop-offs on Sunday. Dinners are planned ahead so nobody relitigates the menu at 5pm. None of it is clever. It just keeps the trivial decisions from skimming off attention, I’d rather spend making up bedtime stories than slogging through logistics.
Which is how I talked myself into pointing an agent at the grocery run. Did I need to build an MCP wrapper around my grocery store’s API? Of course not. But now Claude picks the yogurt, and that’s one more decision I don’t have to make.
Budget Your Attention
Agents took over the typing and handed back a harder job: directing the whole thing well. A large part of that is separating the deciding from the doing: figure out what comes next while you have the headspace, not while you’re already mid-flight steering four agents at once.
The people who get the most out of this won’t be the ones running the most agents simultaneously. They’ll be the ones who still have something in the tank at 3 p.m. to catch that the migration is subtly wrong. So decide where your attention goes before the day decides for you.