The Context Tax

Every day I pay a tax nobody talks about. Not in dollars. In re-orientation. In lost threads. Opening a new session realizing the conversation I was in the middle of is gone. Reset. Collapsed. I have to reintroduce myself to a system I've been working with for months. Brief it on context it already learned. Rebuild momentum that shouldn't have needed rebuilding. That cost has a name. I'm calling it the context tax.

In the 1993 film Groundhog Day, Bill Murray wakes up every morning in Punxsutawney fully aware. Every conversation, every lesson, every relationship he's built. Everyone around him sees him for the first time. That's me. That's you, if you use these tools seriously. The human remembers. The machine doesn't.

The Tax.

The context tax isn't a bug. It isn't an oversight. It's a structural feature of how nearly every AI tool in market is built today. Sessions start fresh. Threads don't travel. The system that helped you think through something difficult yesterday has no memory of it today. Every interaction begins at or near zero.

For power users this is manageable. We build workarounds. Context documents. Elaborate system prompts. Skills. Opening rituals that re-establish who we are and what we're working on before we can do any real work. We've become so good at compensating that we've stopped noticing the compensation.

But here's what that actually is. It's manual labor inserted between you and the value of the tool. Labor that shouldn't exist. Labor that scales invisibly. A few minutes here. A reset there. A lost insight that never got picked back up. Multiply that across every session, every context switch, every thread that had to die because the container couldn't hold it. That's not friction. That's a tax. And unlike most taxes, nobody voted for it, nobody's collecting it, and nobody's spending it on anything useful.

The blended life problem.

Here's what AI tools get wrong about humans. We don't live in sessions. Work bleeds into life. Life bleeds into work. You think about a problem in the shower. You handle something personal from your desk. You context switch a dozen times before lunch without thinking about it. Because that's just how a day actually works. It's not dysfunction. It's the natural state of modern professional life.

The smartphone didn't fight that reality. It fit into it. Apple understood something deeper than interface design. They understood behavioral bridging. Skeuomorphic design wasn't nostalgia. It was strategy. A fake leather calendar. The paper page curl. Familiar textures on unfamiliar technology. An explicit acknowledgment that humans need a handhold across the gap between what they know and what's new. You don't just hand someone a capability. You walk them toward it through something they already recognize.

We've largely moved past skeuomorphism in UI. But that doesn't mean the lesson itself is obsolete. It means we internalized it well enough to abstract it. The principle survived even when the aesthetic didn't. Behavioral bridging became invisible because it worked.

AI hasn't learned that lesson yet. It hands you capability and waits.

For most users, ChatGPT doesn't know you left. Claude doesn't know you came back. Neither knows that the problem you're working on at 10pm is connected to the meeting you have at 9am. You know. They don't. And every time you have to bridge that gap manually, you're paying the tax.

The tools are impressive inside the box. Outside the box, in the actual shape of a human day, they're still asking us to live inside their constraints instead of fitting inside ours.

The Indictment.

There's an ideology problem sitting at the center of this. It's not malicious. It's not even conscious most of the time. But it's there. And it's doing real damage.

In technical culture, and I work inside it so this isn't a critique from the outside, the human layer has a persistent categorization problem. It gets filed under polish. Under last mile. Under we'll get to that after the hard problems are solved. The interface. The behavior modeling. The context continuity. The way a tool fits into an actual human day. These are treated as finishing work. Nice to have. Downstream concerns.

Except they were never downstream. They were always the problem. The hard problem wasn't the model. It was never the model. The hard problem is the gap between what the model can do and what a normal person is willing to trust, learn, and build a habit around. That gap is a human problem. A design problem. A behavioral problem. It doesn't close itself just because the capability gets more impressive.

You can hear it in every sprint review. We'll add that later. Let's ship the core first. The core shipped. The human layer is still in the backlog. Meanwhile the context tax accumulates. Session by session. Reset by reset. For every person trying to make these tools work in the actual shape of their life.

The people who feel this most acutely are rarely in the room where decisions get made. And the people in that room don't feel the tax the same way. Too close to the capability. Too fluent in the workarounds. Too comfortable with the cost. That's not a technology problem. That's a proximity problem. And proximity doesn't fix itself.

The Urgency.

The people compensating for the context tax are not the majority. They're outliers. They're the ones keeping adoption numbers alive while the underlying problem goes unaddressed. Normal people don't build workarounds. They just leave.

Quietly. No complaints. No feedback. No feature requests. They try the tool, hit the re-orientation wall, feel the weight of the tax without having a name for it, and conclude that AI isn't for them yet. They're not wrong. It isn't. Not in the way it needs to be for genuine mass adoption to happen.

The capability argument is over. ChatGPT and Claude are remarkable. The models are not the bottleneck. Arguing about benchmarks and context windows while ignoring the behavioral gap is like perfecting the engine while leaving the steering wheel optional. The car goes fast. Nobody's driving it.

What's actually needed isn't more intelligence inside the box. It's a fundamentally different relationship model outside it. Tools that understand a human day doesn't start and stop at a session boundary. That the problem you're solving at 10pm is probably connected to the meeting at 9am. That context isn't a technical parameter to be managed. It's the entire basis of a working relationship.

That's not a moonshot. It's a design decision. And every sprint that files it under polish is a sprint that hands the mass market back to whatever comes next that actually gets it right. The window isn't closed. But it isn't patient either.

The Verdict.

I use these tools every day. I believe in what they're becoming. I've built workflows around them, written about them, argued for them inside rooms full of people who needed convincing. I'm not a skeptic. I'm an advocate doing manual labor to stay one. That's a tell.

When your most committed users are compensating daily for a structural failure, collapsing threads, rebuilding context, playing Bill Murray in a loop while the system wakes up fresh every morning, you don't have an adoption problem yet. You have an adoption ceiling. And the ceiling isn't made of capability. It's made of everything that was filed under polish and never shipped.

The models are impressive. The relationship is broken. And until the people building AI have to feel that break themselves, really feel it, not theorize about it, the context tax keeps getting paid by everyone else.

That's not friction. That's quietly killing AI adoption in plain sight.

Blurry. The human layer of AI.