Perspective
The next era of care is continuous
Why the most important healthcare experience is no longer a visit, but the connected days between visits.
AI should make care easier, not busier
Imagine a daughter getting a notification that her father has been less active this week. Now she has another job. Is he unwell? Did he leave his device at home? Should she call him, call his doctor, or wait? The software has delivered information. It has not yet delivered help. I think this gap explains much of what we get wrong about AI in care.
A founder perspective on what AI in care should become. The examples are hypothetical, not patient accounts or claims about current CAN capabilities or demonstrated clinical outcomes.
An answer is not the same as help
It is easy to be impressed by a machine that answers a difficult question. But caring for someone is rarely a sequence of well-formed questions. It is a sequence of incomplete situations. Something has changed. Nobody is sure what it means. Several people know part of the story, and each assumes someone else knows the rest.
A model might summarize that story beautifully. The summary can be useful. But someone still has to decide whether to act, find the right person, explain the context, and make sure the action happened. If all of that work remains with the daughter, we have improved her paperwork more than her life.
My starting point is simple: the important unit is not the answer. It is the work a person no longer has to do, without losing control over the decisions that matter.
The work nobody designed
Consider the ordinary task of arranging a follow-up appointment. There may be a recommendation in one system, a phone number in another, a family calendar somewhere else, and a transport problem nobody has recorded. None of these is necessarily a hard reasoning problem. Together they can become a hard day.
This is the opportunity I find most interesting. Not a machine pretending to be a doctor, but software that helps carry an intention through the steps required to make it real. With permission, it could help assemble the relevant information, prepare the request, identify what is missing, and keep track of whether someone has responded.
The last step matters. A message sent is not an appointment made. An alert acknowledged is not a person helped. Software should preserve those distinctions instead of treating activity inside the software as success outside it.
Start with a smaller promise
There is a temptation to begin with an enormous promise: an AI that looks after your health. I would rather begin with a small promise that can be kept. Help this person complete this task. Make the boundaries visible. Learn where the process breaks.
For a founder, this can feel insufficiently ambitious. I think the opposite is true. A small task forces you to encounter the real constraints. Who has permission to see the information? What happens when it is wrong? Who takes over when the software cannot proceed? You cannot answer these questions with a better demo.
If a system cannot reliably help close one follow-up loop, giving it responsibility for an entire life is not vision. It is avoiding the details. The larger ambition should be built out of smaller responsibilities we have learned how to handle.
Silence is part of the product
Return to the daughter and the activity notification. A decline in recorded movement is an observation. It is not a diagnosis. The device might be unused. The weather might have changed. Her father might have spent the week reading. A system that turns every variation into a warning makes the family responsible for interpreting its uncertainty.
That does not mean the system should stay silent about everything. It means deciding when to interrupt is a serious design problem. Thresholds, missing data, escalation rules, and the availability of a real responder belong in the design from the beginning. Clinical uses require appropriate validation and oversight, not just plausible explanations.
We should ask a question that technology companies do not always like asking: how much attention did we consume to produce this benefit? A product can be used more because it is useful. It can also be used more because it has made people anxious.
Independence is not a data problem
A family may want reassurance. The person receiving care may want privacy. Both wishes are reasonable. Better prediction does not resolve the disagreement. It can make the disagreement more consequential.
This is why I do not think the goal should be to make every part of a person’s life visible. The goal should be to help them live the life they choose. That requires understandable permissions, limits on collection and sharing, and ways to correct or refuse the system. A person should not have to surrender every ordinary freedom to receive useful support.
The same principle applies to human judgment. Software can prepare, organize, and suggest. Authority should follow the stakes, with qualified people responsible for clinical decisions. When the system is unsure, saying so is a capability, not an embarrassment.
What I want us to build
At CAN, the ambition I want us to hold ourselves to is not that care should contain more AI. It is that caring should become more possible. For a family, that might mean fewer unresolved tasks at the end of the day. For a care team, it might mean less time reconstructing context. For the person receiving care, it might mean needing less help to do something they value.
These are goals to test, not outcomes to assume. We should measure completed follow-ups, unnecessary interruptions, time spent coordinating, and whether people feel more in control. If we claim a health benefit, we need evidence appropriate to that claim. A convincing conversation with a model is not that evidence.
The future I want is not one in which a daughter spends her evening managing an intelligent dashboard. It is one in which the practical work is handled well enough that she can call her father because she wants to talk to him. AI will have done something important when it gives them less administration and more of their relationship back.