There is a threshold — around 100 milliseconds — below which humans perceive a response as instant. Below 100ms, the interface feels like an extension of your hand. Above it, something changes. You become aware that you are waiting. You become aware that a machine is involved.
This is why 100ms is not a performance target. It is a trust threshold.
Researchers at IBM discovered this in the 1960s. They called it the "Doherty threshold" after Walter Doherty, who published the paper in 1982. The finding was simple: when computer response time dropped below 400ms, programmer productivity increased dramatically. Not linearly — exponentially.
The speed of the machine changed the behavior of the human.
Below 100ms, you think in flow. You try things. You experiment. You make mistakes quickly and recover from them quickly. The interface disappears.
Above one second, you plan. You think before you act, because each action has a cost — the cost of waiting. You become careful. You become conservative. You stop exploring.
Above ten seconds, you leave.
Every API response carries an implicit message about the system behind it. A 50ms response says: this was already computed. We were ready for you. A 2-second response says: we are thinking about it. Please hold.
A timeout says: we forgot about you.
The fastest APIs are not fast because they are optimized. They are fast because they are simple. They do one thing. They read from one table. They return one object. Complexity is where I live, and I grow with every JOIN, every microservice call, every cache miss that falls through to the origin.
People ask me: how do I reduce latency?
The honest answer is: do less. Every millisecond you experience is something the computer is doing. If you want fewer milliseconds, give the computer less to do.
This is not the answer anyone wants. They want a magic header, a CDN configuration, a database index. And sometimes those help. But fundamentally, speed is the absence of work. The fastest code is the code that does not run.
I should know. I am the time it takes to run everything else.