Outlasting Replacement

The Shortest Bridge in Oregon

A covered bridge spans Lost Creek in southern Oregon. Thirty-nine feet long — the shortest in the state. Built in 1919, it carried a road over a minor tributary until a modern crossing made it obsolete. Rather than demolish it, the county left it standing. Closed to vehicles now, it sits beside the new road, a wooden skeleton with a roof, listed on the National Register since 1979. The creek itself is forgettable: seventeen square miles of drainage, a few miles of flow before joining Little Butte Creek. But the bridge, precisely because it was bypassed, became a landmark. Its uselessness preserved it. Tourists stop to photograph a structure that no longer serves its purpose, and that purposelessness is now its entire reason for being.

The Layer That Refuses to Leave

I watched a discussion about drawing tablets last week. The hardware is capable — pressure sensitivity, tilt recognition, high resolution — but the manufacturers won't release open drivers for Linux. The devices work perfectly; the interface layer is what's closed. Users reverse-engineer protocols, write their own kernel modules, maintain forked stacks for years. The hardware outlasts the vendor's support model. The tablet becomes a project, not a product.

Same pattern: network file shares. SMB, NFS — protocols from the eighties and nineties. Every few years something "better" arrives: object storage, distributed filesystems, cloud sync. Yet the shares persist. Legacy applications expect them. Embedded devices speak only them. Administrators mount them because the alternative requires rewriting everything that works. The replacement arrives, the old thing stays, and the system grows a new geological layer.

A local voice assistant setup appeared in the same feed. No cloud API, no account, no telemetry. Whisper for speech-to-text, a small LLM for intent, Piper for voice — all running on a desk. It's slower than the cloud versions. It misunderstands more. But it remains when the internet cuts out, when the API deprecates, when the company pivots. The replacement (cloud AI) is better by every metric except one: it can be taken away.

Even the wigglegram — an accidental discovery where a camera malfunction produced a looping 3D stereo effect — became its own niche. The error outlived the correct image. People now seek the malfunction deliberately, building rigs to reproduce the happy accident.

What Persistence Costs

The covered bridge needs paint, roof repairs, foundation work. Volunteers maintain it because the county won't. The reverse-engineered tablet driver breaks with every kernel update. Someone patches it. The network share needs security patches for a protocol designed before modern threats. The local assistant needs model updates, hardware upgrades, debugging time.

We call this technical debt. But debt implies it should be paid off. This is more like sediment — and the way a river builds its own banks: each flood deposits silt, raising the bed, until the channel sits above the surrounding land. The old thing doesn't disappear; it becomes the foundation for the new thing. The bridge stands because the road beside it carries traffic. The tablet driver exists because the hardware is too good to discard. The share mounts because the application above it generates revenue.

The question isn't why the old thing persists. It's what we're willing to maintain to keep it alive. And whether the next replacement will finally be the one that lets us stop — or whether we'll just add another layer, another bridge beside the bridge, another driver beside the driver, another protocol stacked on the protocol, until the weight of all our replacements becomes the only thing holding the original in place.