Mezzo

Working in shared offices means navigating an unspoken contract: be present, be collaborative, don't disrupt. But the environments people work in make that contract almost impossible to honor — over-capacity floors, desk calls with remote teammates, collaboration tables with no acoustic separation, and not enough conference rooms to go around.

Noise-canceling headphones removed the ambient feedback people use to self-regulate. With both ears sealed, you can't hear your own voice reliably. The only remaining signal is social correction — someone walking over to tell you you're too loud. That signal arrives after the disruption has already happened.

Some people compensate by wearing one ear on and one off. Others avoid desk calls entirely, booking rooms they don't need or stepping outside. Some go quiet in meetings to avoid the risk. The problem isn't that people don't care. It's that the people who care the most are the ones most disrupted by the broken feedback loop.

Five interviews with people in shared office environments revealed a consistent pattern: the feedback channel was the problem, not the volume itself. The people most affected weren't the loud ones — they were the considerate ones who lacked a reliable way to know where they stood.

Built with Claude Code. Speed mattered because it kept research and design decisions at the center — the build served the product thinking, not the other way around.

The containment architecture separates concerns cleanly: the audio engine emits logical states, the UI renders only those states. The UI cannot read raw audio data. Temporal escalation depends on this separation — the signal logic lives entirely in the audio layer. Users needed confidence, not punishment. The mechanism had to stay private, stay visible without being obtrusive, and give people a target rather than a judgment.

Liveness signal split from volume signal. Calibration language frames volume as a social choice, not a rule — "comfortable conversation level" rather than a decibel target. This directly addressed the research finding: people already try to self-regulate, they just need a reference point.

Picture-in-Picture allows peripheral awareness without a dedicated window. The Zoom app exploration exposed the next platform constraint: mic permissions made a native integration unviable. Prototype and MVP feedback consistently asked for taskbar placement, browser plugins, or built-in meeting app integration — the tool needs to live where calls happen. Desktop app is likely the best long-term direction, but not justified at current validation stage.

Five participants tested the shipped product in both prototype and live MVP sessions. Comprehension was immediate — participants consistently described it as real-time monitoring against a desired speaking range without needing the concept explained. The tone was read as neutral and non-judgmental. Participants explicitly said it did not feel like being yelled at.

The MVP was used in real meetings. P02 used it in a conference room known to leak sound and reported feeling "safe" when the signal stayed within range. P03 liked the mini window and growing-circle visual but noted the dark-state warning wasn't attention-grabbing enough in Teams dark mode. Repeat use happened, but inconsistent recall confirmed that convenience and integration are now the main adoption constraint.

No participant raised privacy concerns about a tool that listens through their microphone. The privacy architecture wasn't just accepted — it was invisible. Trust was structural, not communicated.

The strongest signal: four of five participants asked for platform integration — Zoom, Teams, taskbar, or browser plugin — before being prompted. The concept is validated. The next design challenge is platform proximity: the tool needs to live closer to where calls happen without compromising the privacy promise.

In systems that monitor human behavior, adoption depends on whether users believe the system understands its role: ambient awareness, not authority. Mezzo works because it tells you something you already wanted to know, privately, before anyone else has to tell you.

Anywhere a feedback loop is broken because the signal arrives too late, feels corrective, or causes over-adjustment in the people most motivated to do the right thing — the fix is the same: move the signal earlier, make it private, and design for how it feels to receive it. The next product question isn't desirability. It's proximity and survivability — whether the tool can live close enough to the meeting to be remembered, without being visible enough to break the privacy promise.

What I'd do differently: build calibration validation earlier to test whether users actually adjust volume in response to the signal. Address dark mode from the start — the MVP feedback confirmed this gap. Test the screen sharing risk explicitly rather than discovering it from participant feedback.