Positions, stated so you can disagree with them

In consumer software, a confusing interface costs a conversion. In care software, it costs accuracy in a record someone will later rely on. A care worker on hour nine of a shift, interrupted, holding a device one-handed, is the design condition — not the degraded case we handle if there is time.

This changes what counts as a serious design argument. "This adds a step" is a safety observation. "This screen asks the user to classify before recording" is a safety observation. Treating those as polish is how high-consequence software quietly gets worse.

Most teams design the happy path and then bolt on error handling. In records-based software, get it the other way round: design how someone fixes a mistake before you design how they make an entry.

If correcting a record is hard or feels shameful, people do not correct it. They work around it, or they hide it. Hidden errors are the expensive kind. A plain, non-punitive amendment path is one of the highest-leverage things you can build, and it almost never appears on a roadmap because nobody demos it.

Safety trade-offs get made whether or not anyone notices. A threshold gets picked, a warning gets sized, a step gets removed. The failure is not choosing badly — it is choosing implicitly, in a ticket, by someone who did not know they were choosing.

Design's job in a regulated domain is to surface the trade-off, cost both sides in terms the accountable person can weigh, and then build the interface that makes the chosen option survivable. Not to have the best instinct in the room.

Consistency is the least interesting reason to build one. The reason is that keyboard completeness, contrast, error prevention and the way clinical risk gets expressed should be solved once and inherited — not re-decided in every review by whoever remembers.

That reframing changes how you fund it, too. "Design system" competes with features for budget and loses. "The mechanism by which every screen inherits our accessibility floor" is a different conversation.

Describing a risky flow in a document gets you a nod. Putting a working prototype in front of a clinical stakeholder gets you an argument, which is what you actually wanted. The disagreement surfaces before anyone has built it, when changing your mind is nearly free.

This is where AI tooling has earned its place fastest in my team: not generating interfaces, but making a testable artefact quickly enough that we can afford to be wrong in front of people.

The moment design leadership becomes the last approval before shipping, it becomes a cost, and costs get routed around. Being earlier and less powerful beats being later and more powerful every time.

Bring options with costs instead of recommendations with rationale. Say out loud what evidence would change your mind. Escalate trade-offs, not disagreements. Spend credibility on almost nothing, so that the floor you will not move — accessibility, clinical safety — never has to be argued for twice.