I cut my active product count down to three. Two others surface from time to time.
That was how I phrased it, anyway.
“Surface from time to time” makes them sound as though they spend the rest of the week somewhere else, politely taking up no space. They don’t. They go quiet.
I still have to remember where each one stands, what was promised, why we chose the ugly solution instead of the elegant one, and which temporary fix has quietly become part of the building.
A dormant project is still a project you know too much about.
So the honest number is five.
Five product contexts, before everything else. They have started crowding out books, games, long conversations, idle curiosity, and any subject that asks me to load another world into my head before I can enjoy it.
This is why work-life balance feels like the wrong unit of measurement. It divides the day into two clean categories that do not exist. Work is not one context. Life is certainly not one context.
I can spend eight hours on one difficult problem and finish the day more or less intact. Or I can spend ninety minutes answering messages from seven projects, talk to a doctor, check a school portal, approve an invoice, review a contract, and remember a call I promised to make. After the second kind of morning, I can look at an empty afternoon and have no idea what it is for.
The reloads are what wreck me.
Here comes the embarrassing part: an engineer who has spent too much time around language models is going to explain how the human mind works.
This is how engineers become unbearable at dinner. We find one abstraction that behaves nicely and start seeing it everywhere. Programmers discover APIs in marriage. Economists discover incentives in grief. I discovered the context window.
I don’t think a brain is an LLM. I do think the metaphor catches one irritating fact: knowing something is not the same as having it loaded.
A model may contain an enormous amount of knowledge, but it can only work with so much at once. The analogy I care about is narrower than a theory of consciousness. People also know far more than they can keep live at any given moment.
You may know the history of a company, the architecture of a product, the personalities involved, the market, the mistakes already made, and the reasons behind a dozen decisions. But only some of that can remain active while you are making the next one.
The trouble is that a product is not one item in memory. It is a world.
It has its own vocabulary, old feuds, people who own things only on paper, decisions nobody would make today but nobody can reverse, and workarounds everyone has forgotten are workarounds. Returning to it after a week is less like opening a document than reconstructing the state of a small country.
The more of those countries you keep alive, the less room there is for a new one.
I noticed this most clearly with games.
I used to love games that made it obvious I would need a notebook. Give me rival empires, dead gods, a civil war four centuries ago, a crafting system, seven currencies, and three kinds of damage, one of which is apparently spiritual.
Perfect.
I would read wikis, ruin a character build, start again, and spend a week learning a world that had no interest in being legible in its first twenty minutes.
Complexity felt like a promise.
Now it often feels like paperwork.
I recently started a game that introduced a fallen dynasty, two competing orders, elemental resonance, and a man whose name I forgot while he was still saying it.
Ten minutes later I had crafting.
Then a skill tree.
The game had not yet given me a reason to care, but it was already asking for office space in my head.
I closed it and opened an older game whose systems are objectively no easier. The interface is worse. The balance is occasionally absurd. Years ago, I learned parts of it from forum posts written by men who treated damage calculation as a private religion.
But I had already loaded that world.
I knew the map, the items, the economy, the old war, the stupid name of the northern king. The whole thing expanded from storage with almost no effort.
We do not return to old games only because they were better. We return because we have already paid the cost of loading their world.
This would be a good place to stop reading and go for a walk.
Seriously. Close the tab. Go around the block. Come back later, or forget the essay.
It has already taken up a fair amount of your context with an unnecessary theory about context, which is the kind of self-importance essays usually disguise as depth.
Come back if the thought is still there. If it isn’t, that is probably the correct review.
I spent most of my life calling the results of this kind of compression “conclusions.”
A great deal happens. There are motives, accidents, exceptions, contradictions, details, and people who behave differently on Tuesday. Eventually the whole mess gets reduced to something small enough to carry:
Never work with this person again.
A project without an owner will become yours.
A client who cannot describe the problem will still know you solved the wrong one.
I used to think these were truths extracted from experience. Now I suspect many of them were simply ways of shrinking the case file.
A conclusion lets you keep the output without rerunning the entire experience. Most of the original context is discarded. What remains is a compact rule that can be loaded quickly when something similar happens again.
Sometimes the rule is good. Sometimes the detail we threw away was the only one that mattered.
We compress anyway, because the alternative is to relive every relevant episode before making breakfast.
I suspect stereotypes come from the same machinery.
A full person arrives with a biography, motives, loyalties, contradictions, fears, habits, and an internal history you do not have time to reconstruct. So the mind loads a cheaper category instead: bureaucrat, salesperson, engineer, investor, therapist, founder.
Now you roughly know what to expect and at what point the person will propose a strategy session.
That makes stereotypes dangerous, not mysterious. The category is cheaper than the biography. It is a compressed file, and a lossy one. The trouble starts when we forget what was discarded.
When there is room in the active window, new context feels inviting. You want to learn the names, understand the mechanic, read another book, talk to someone whose view of the world takes work to assemble.
When the window is full, the same thing feels intrusive.
Modern games are bloated. New films are stupid. Music ended twenty years ago. Interfaces have been ruined. The new tool is hype. The book is overrated. The person is boring.
To be fair, some games are bloated, many interfaces have been ruined, and the music may simply have learned to hide.
But it is convenient how often our taste spares us the cost of learning anything new.
A full system rarely says, “I do not have room for this.” It says the thing is not worth understanding. That may be part of what we call becoming set in our ways. Not just belief, but memory management.
I should be careful here. The context-window metaphor is not a theory of the brain, and age is not a software bug. But the practical experience is hard to miss: the system becomes less forgiving while the baseline load grows.
At twenty, health mostly runs in the background. Money can be managed by not having much of it. There may be school, an early job, a few relationships, and a great deal of empty space that looks, at the time, like boredom.
Later there are children, aging parents, employees, companies, property, taxes, doctors, insurance, lawyers, investments, obligations, and passwords for services no one can close because no one remembers why they were opened.
Bad sleep matters more. Illness leaves residue. A minor injury can become a management problem. The machine is older, and more things launch at startup.
So yes, perhaps the window narrows. More importantly, less of it is empty.
We also fill it by choice.
We do not close projects. We place them in the ceremonial state known as “I may come back to this.”
We delegate the task but keep the full state in our own heads. We build systems nobody else understands, then become annoyed that nobody else can take them over. We agree to become the escalation point for other people’s helplessness.
Competence makes this worse.
The more you can solve, the more context gets routed toward you. In the same meeting, one person hears what is being said. Another is also holding the technical history, the commercial constraints, the personalities, the previous promises, and three ways the plan can fail.
Both spent an hour in the room.
Only one loses the evening.
This is also why I hate test context in conversation, especially in interviews.
Someone spends half an hour constructing a synthetic universe. There is a company, a product, a market, a client, a metric that rose, another that fell, a team that cannot execute, and executives who want growth, lower costs, and no unpleasant news.
Then they ask a question and watch to see how well the model in your head performs.
A lot of case interviews do not test judgment. They test how quickly you can load someone else’s toy world, infer its hidden rules, and produce the answer its creator already considers intelligent.
The actual problem may be simple. Most of the effort goes into reverse-engineering the house religion.
“This is too dumb for Fable.”
Ask an LLM.
It will accept every premise, produce four hypotheses, rank them by impact and effort, draw a matrix, and conclude that the real answer is cross-functional alignment.
A person with operating experience tends to ask the questions the exercise has excluded.
Why did the company get here? Who made the earlier decisions? Which numbers are real? What incentives are people responding to? Why does the process exist? Who benefits from it remaining broken?
That can look like resistance.
It is resistance.
I do not enjoy spending scarce context on a fake world built to test how politely I can inhabit someone else’s assumptions.
Context also has weight.
Five stable projects may be easier than one small project run by erratic people. An unresolved conversation can occupy more space than a launch. A person who may send a consequential message at any moment is present even while silent.
Calendars do not see any of this. They record events, not maintained state.
A personal context does not observe business hours. A sick child, a parent in trouble, a legal problem, an unstable relationship—none of these needs a meeting invitation to take over the day.
You can do no visible work and still get no rest.
Free time without free context is not rest.
It is time during which you are not acting but are still maintaining every open system: remembering who expects what, what may change without you, and what breaks if you get something wrong.
I have started to think of unclaimed mental space as a form of wealth.
A weekend is not much use if every open loop comes with you. What matters is having some portion of attention with no owner, deadline, KPI, or likely emergency attached to it. Enough room for a book, a game, an idea, or another person to enter without evicting something important.
That is what I mean by context balance.
Work-life balance counts hours. Context balance counts the worlds you are responsible for keeping alive.
It cares less about whether you stopped working at six than whether five different worlds can change while you are away. Whether they depend on information held only in your head. Whether any of them can survive your absence. Whether some should have been shut down months ago but remain open because closing them would require one final burst of effort.
The goal is not to carry nothing.
Children, work worth doing, friendship, love, and ambition all take space. Some of the things that consume us are the things we chose and would choose again.
The problem is everything else sitting beside them because we never made it leave.
Naturally, I now want to build a system for this.
Count the contexts. Assign weights. Track switching costs. Mark volatility. Run a weekly review. Maybe build a product that helps people manage the problem.
My answer to too many systems is apparently another system that has to be maintained.
That would make six.
So I will leave it alone and open an old game.
I paid that loading cost years ago, when I still had room.
With low effort.
Work-life balance no longer works. Maybe it never did.