My tool stack has a half-life and I didn't notice until I started counting. Some apps I installed three months ago are already uninstalled. Some I installed three years ago are still open right now. The ones I evangelized hardest tend to be in the first bucket. (this is the embarrassing part of the audit)
In physics, half-life is probabilistic. You don't know which atom decays, only that across the pile, a predictable fraction will be gone by some date. Some materials last billions of years. Some last minutes. My tool stack has both kinds and I can't always tell which is which when I install something new.
The Layer Below
My first theory was that the big labs kill tools by absorbing their features. That's partly true. Standalone voice-to-text wrappers around AI are probably on a timer. The big labs are already shipping pieces of it natively and that whole category starts collapsing.
But the survivors in my stack aren't surviving because the labs can't replicate them. Wispr Flow stuck because it works in every text field in every app, not because it's better than what a lab could build. Warp stuck because it runs the same on every OS I touch (with every AI feature turned off, which probably says something). The pattern is deep integration. Keyboard level. OS level. The app sinks into the layer below it and gets better while you stop paying attention. That's what iOS and macOS and Android do at the platform layer, and the apps that survive on top do the same thing one layer up.
The dead ones in my stack all shared one trait. They wanted me to show up to them. Open the app. Learn the interface. Sit inside their walls. Some were automation platforms, some were coding agents, some were full dev environments, and it didn't matter how good any of them were at their job. That's not a tool, that's a destination, and destinations have short half-lives because attention is a renewable resource until it isn't.
Same Shape, Same Decay Rate
Here's what's interesting about the coding harnesses. Codex, Claude Code, Antigravity, Conductor. Open any of them today. Left side agents tab. Middle main area. Right panel for files. Bottom terminal. Same silhouette, different logo. They all look the same now and they didn't six months ago.
When every tool in a category converges on the same UI, that convergence removes the last excuse not to switch. They're all decaying at the same rate because they're all the same shape. Cursor is the one pulling ahead right now, mostly on speed and model quality. But that's a performance lead, and performance leads are the fastest-decaying advantage in this space. One model release from a competitor and it evaporates. I might be wrong about this. My track record on picking winners is literally the first paragraph of this post.
A lead built on speed has a half-life measured in months. A lead built on integration is measured in years.
Run the Audit
So go run ps aux on your dock. Which icons became background processes and which ones still demand you show up? Forget loyalty. I keep switching models every time a new version drops, from Claude to GPT and back again, and I don't think about it for more than an afternoon. The tools I keep are not the ones I love. They are the ones I stop noticing. Everything else is just waiting to decay.