"We measure AI by benchmarks. We should measure it by how long we stay in flow. The best AI is invisible AI."
The AI industry is obsessed with benchmarks: MMLU scores, HumanEval pass rates, perplexity metrics. These numbers are useful for comparing models in isolation, but they tell us nothing about how AI actually performs in the context of human work. They measure correctness, not experience.
We propose a new measurement: the Flow State Metric (FSM). It measures the percentage of time a user maintains deep cognitive focus while using an AI system. Every interruption—latency spike, unexpected behavior, context switch—drops the FSM. The goal is 100%: AI that never breaks flow.
Through user research, we've identified the primary flow-breakers in AI interaction:
Latency is the dominant factor. Nearly half of all flow breaks come from the AI simply taking too long to respond. This isn't about actual response time—it's about variance. A consistent 500ms response is less disruptive than an inconsistent 100-2000ms range.
Flow state requires low cognitive load. When the AI is predictable, fast, and consistent, users can offload mental effort and focus on their actual work. When the AI is unpredictable, users maintain conscious awareness of the tool, consuming cognitive resources that could be spent on the task.
Infe targets 94%+ FSM across all interactions.
Through consistent sub-100ms latency and predictable behavior, we enable users to stay in flow. When AI becomes invisible, you can focus on what matters.
The challenge with FSM is measurement. Flow is subjective. But we can measure its proxies: time between interactions, error correction frequency, session duration, and behavioral patterns. Combined, these give us a quantitative view of qualitative experience.
Stop measuring how smart your AI is. Start measuring how invisible it is. The best AI is the one you forget you're using.