Evermind: ambient awareness without surveillance

I founded Evermind, a venture-funded aging-in-place company in Nashville, on a small idea: an older person’s ordinary routines are the most honest signal of well-being there is. Inline power sensors on a coffee maker, a lamp, a TV — when they switch on, the family gets a quiet ping. No camera, no wearable, no help button.

The design started from what older people won’t do. As my family put it of my 96-year-old grandmother Lois, who had a test system in her home: she would “never wear the help button in a million years, she just wouldn’t do it.” Medicalized monitoring fails not because it doesn’t work but because it turns a home into a hospital and a person into a patient. The frame had to be dignity, not surveillance — and that frame had to hold across the hardware, the alerts, the family’s reading of them, and even the marketing. “Daily rituals,” as I said at the time, “are signs of stability and markers of good health.” I know Lois has had her coffee because I know her coffee pot comes on. You can meet her in the short film our family made about living with the system — I’m the grandson in it, explaining how the ambient monitoring works.

The question that vexed me

Underneath the warm product was a genuinely hard research problem — I eventually gave a whole talk on it, “The Design of Representation in the Study of Human Activity”: what right do we have to go from low-level sensor data to a much higher level of abstraction — from volts and amps to “grandmother is okay”? A sensor does not neutrally observe the world; it shapes what can be seen. An inline power sensor turns a continuous draw into clean on/off events you can build on; a motion sensor gives you continuous data with no good units and a temptation to over-read it. Sensors shape phenomena. The instrument is already an interpretation.

Which led to the design principle I still hold: build for seams, not smoothness. A dashboard that simply glows “OK” hides too much; evaluation obscures the seams, and deviation is not always bad — a broken routine can be the most important thing the system has to show. The sharpest statement of the discipline came, of all places, from a smart-home API guideline I quoted to anyone who would listen: “Because there is smoke, there is not necessarily a fire… does not mean the home is now safe.” Do not editorialize. Show the signal and its seams; let the humans infer.

Sensors shape phenomena. Useful sensors are the ones that allow for honest abstractions — and leave their seams visible.

Where it went

The system had two lives. Families used the remote monitoring successfully — the consumer product the coffee pot story belongs to. And respiratory DMEs, the providers who manage equipment like non-invasive ventilators in patients' homes, used the same platform to monitor that equipment remotely for patient adherence. That second life is the one that lasted: Encore Healthcare acquired the platform in 2017, and Nexus, the post-acute care platform Evermind developed, is used today to manage over a million home respiratory patients. And the arc closed: a decade after founding Evermind, I returned to Nexus as a consultant, designing NexusGPT — an assistant that synthesizes a COPD patient's prescriptions, spirometry readings, and progress notes into insights and protocol recommendations for the respiratory therapist on the call. The platform I built to watch over one grandmother now carries an AI layer I designed to help clinicians watch over a million patients. The design argument was public early, too: my talk presenting Evermind at Aging 2.0, hosted at IDEO in San Francisco in May 2013, is still on video.

The original AI-inference problem

Evermind was solving, in 2013, the exact problem every AI agent now has: the leap from raw signal to a confident human-readable claim. An agent that says it booked the trip, summarized the meeting, or noticed you seem stressed is doing what Evermind refused to do casually — collapsing low-level evidence into a high-level assertion the user is asked to trust. The Evermind answer is the one I still argue for: make the system’s inference an inspectable contract with visible seams rather than a smooth claim of certainty, and design domestic intelligence so that it preserves dignity and leaves the consequential judgment with the people who live there. The representational lesson outlasted the company — and so, it turns out, did the platform.

Case study · 2013–14, Evermind (Nashville). Aging-in-place; IP sold to Encore Healthcare, 2017. ← The Practice