The Friction Worth Keeping
On seamlessness, safety, and sovereignty
Try to get a copy of your own medical records sometime. Depending on where you live, you might fill out a form, wait a few weeks, and pay a fee for the privilege of reading information about your own body. It is one of the small, familiar frictions of healthcare, and the most exciting companies in healthtech are built to remove friction.
That instinct is usually right. Across digital health, AI, and brain-computer interfaces, the mood of the moment is to integrate everything, strip out every step, and make it seamless, just as we do with the tools we build for ourselves, stitching our calendars, inboxes, and AI workflows into one frictionless surface. Much of the time, less friction is simply better.
But not all friction is akin to the records-request fee. Some of it is there precisely to reduce risk. Anyone who has been in hospital knows the version that feels almost absurd, your name and date of birth confirmed every single time, even for a Panadol. It is repetitive, mildly annoying, and one of the quiet reasons people don’t get the wrong drug. Healthcare has always carried a heavier regulatory burden and a lower appetite for risk than most industries, for the simple reason that the cost of being wrong is measured in people. A device can spend years in trials before the FDA, or the TGA here in Australia, will clear it, and that slowness is itself a designed friction. It is fashionable to treat that as an obstacle to disrupt. Sometimes it is. Sometimes it is the system doing exactly what we built it to do.
So the question is not whether we want more friction or less. It is which friction, where, and decided by whom. I want to look at a few companies doing genuinely impressive work, several of them Australian, and make a case for some friction.
Heidi
Heidi is an Australian-founded ambient AI scribe. It listens to a consultation, generates a structured note for the clinician, and can produce patient-facing material too, a plain-language summary, a referral letter, after-visit instructions, in well over a hundred languages.
It is genuinely excellent. Documentation consumes an enormous share of clinicians’ lives and is a real driver of burnout, and Heidi has reportedly handed back millions of hours. What it does for patients is quietly radical, too. That records-access problem I opened with? Heidi dissolves it. A patient can leave a consultation with a clear, immediate, readable account of what was discussed, sometimes in their first language rather than the clinician’s. That is a real move away from a paternalistic model toward a partnership where the patient walks out informed.
The design question its success raises is not really about Heidi, which keeps the clinician in the loop by design and is among the more thoughtful tools in its category. It is about dependency. A clinician who comes to trust any scribe completely stops taking their own notes, and the old friction of the parallel record quietly disappears, engineered away by how well the thing usually works. The friction worth keeping is light here, just enough independent practice that if the tool ever fails mid-consultation, the moment is an inconvenience rather than a void. Seamlessness saves effort, but it can also erode the backups we don’t notice we rely on until they are gone.
Everlab
Everlab works at the preventative end of medicine, comprehensive testing and the aggregation of personal health data to catch risk early.
The promise is real, and speaks to the future of care I believe in. So much of medicine is reactive, waiting for something to break. A model that is preventative, data-driven, and individualised is simply better medicine. Post-COVID, patients increasingly want autonomy and partnership over the old paternalism, yet most have no trustworthy central place to see their own health trends, and no fifteen-minute appointment can hold those conversations. Done well, a service like this hands people genuine ownership of their health, and could even narrow inequities, giving those who can’t reach niche, personalised care some version of it in their pocket.
Two design questions sit inside that promise. The first is structural. Frictionless aggregation of deeply personal biomarker and genomic data into a single profile is valuable to the patient, and to others. What can be pooled can be repurposed, so where does it live, who can reach it, and what happens when it crosses a border or the company is sold? The second is subtler. A subscription model is, by structure, incentivised to make people feel something always needs monitoring. Much testing returns unremarkable results, and a model that depends on sustained engagement has a quiet incentive to keep the patient worried. The promise and the risk live in the same feature, which is exactly why the design choice matters. The friction worth keeping is structural restraint, boundaries on how data aggregates and travels, and a business model honest enough not to manufacture the worry it then charges to soothe.
Synchron
Synchron is one of the most impressive companies to come out of Australia. The signals from its Stentrode device reaches the motor cortex through a blood vessel rather than open-skull surgery, and is heading toward a pivotal trial, an important step on the long road toward regulatory approval. For people with severe paralysis, whether from ALS, stroke, or spinal injury, it offers something close to miraculous, a way to communicate through thought alone. Having watched this field since my own research years, I can say the achievement is immense.
What I admire most is that the implant pathway is a model of good friction, regulated, staged, deliberately slow, relentlessly safety-first. Nobody is rushing a device that sits inside the skull to market, and rightly so.
Which is why one element is worth thinking about, raised with the field rather than at this company. Synchron has integrated a large language model to help patients communicate, which is remarkable and genuinely expands what is possible for someone locked in. But it places an AI inference layer between neural intent and the words that emerge. The implant is governed by deep, deliberate friction, so does that layer inherit the same rigour? When a model “helps” by predicting what someone meant to say, where is the line between assistance and authorship of another person’s speech? These are not reasons to slow the work, but reasons to extend implant-grade discipline, auditability, human verification, deliberate limits, to the part now doing the interpreting. It is worth noting where the money comes from, too. Synchron’s most recent round included In-Q-Tel, the venture arm of the US intelligence community. National-security capital is already inside the field, which brings me to the largest point.
Neuracle, and the wider frontier
Neuracle Medical Technology, in Shanghai, has its own implantable wireless brain-computer interface, NEO, in a multi-centre trial for tetraplegia. Brain-computer interfaces, a direct line to the signals of the brain and the most intimate data that has ever existed, are now being pioneered in parallel by multiple state-backed ecosystems at once. The US, Australia, and China are all advancing simultaneously. Neural data is, whether we like the framing or not, a sovereign question.
It is worth being concrete about why, because the stakes are easy to wave past. Neural data is not a leaked password or even a stolen medical history. It is a readout, however partial, of how a particular brain works, its responses, its patterns, in some cases its intentions before they become action. Picture an adversary, a hostile state or a coercive employer, able to access the neural data of a population, or of one high-value individual, such as a leader, dissident, or soldier. The ability to interfere with a device wired into their nervous system is a category of power we have never had to govern. A device that can read from or write to the brain can, in principle, be turned against the person it was meant to help.
None of this is hypothetical, and none of it is new. The same technologies that restore function to a patient are, by their nature, dual-use. DARPA was funding high-bandwidth neural interfaces a decade ago through its Neural Engineering System Design program, and its later Next-Generation Nonsurgical Neurotechnology program pursued non-surgical, bidirectional links between an able-bodied soldier’s brain and machines, for tasks like controlling drones or cyber defences by thought. When human data and/or augmentation are this entangled with national capability, friction is what protects us all. And as the field accelerates, that risk scales with it, more companies, more devices, more data, and more speed mean more surface for error, breach, and the corner-cutting of a crowded field racing to ship.
What this comes to
The striking thing, across all of these, is how much good is being done. A scribe that returns millions of hours to exhausted clinicians and hands patients a record they can finally read. Preventative medicine that catches what reactive care misses. An Australian implant giving speech back to those who had lost it. A global frontier turning science fiction into ordinary care. None of it is the enemy, and because it is so good, it is worth protecting properly.
Two ideas are worth holding onto. The first is that friction is a tool, not a failure. Some of it protects nothing and should go, the records fee, the access barrier, the half-day in the waiting room. But some of it is load-bearing, the clinician who reviews the summary, the boundary that stops data being silently pooled, the limit on the inference layer, the deliberate non-interoperability that means a system cannot easily be turned against the people it serves. From the outside, the two kinds look identical, which is how a culture optimising for speed and a smoother user experience ends up removing both at once. Telling them apart, and choosing on purpose which to keep, is the whole task.
The second is that the stakes run past the individual, and past the dramatic end of the spectrum. It is tempting to think the real risk lives only with brain implants and battlefield neurotech, while the friendly scribe and the preventative-health service are harmless by comparison. In most software, removing friction carelessly costs you a lost file. In medicine and neurotechnology, the same carelessness can cost far more. That is the reason to be deliberate now, while these technologies are still being shaped. We will never build perfect systems, but if we keep the friction that protects what we care about, we may at least solve our problems at the velocity we create them. That is not an argument for going slower. It is an argument for going fast without leaving the important things behind.