CALM: Watch Before You Build

Good design begins before the first sketch — in the room where the work actually happens.

The first interface is the work people already perform.

Real work is shaped by timing, handoffs, interruptions, artifacts, physical reach, and local shortcuts that rarely appear in requirements. Under pressure, missing these details can lead to extra clicks, hidden checks, training effort, and late rework.

Calm rule: before adding or redesigning a workflow step, watch the current task in its real setting and name the workload your design must not increase.

A team spent several years developing a transcutaneous gas measurement module for the neonatal intensive care unit (NICU). The team did not involve NICU nurses early enough to observe the workflow the module would need to support. When the project reached its first customer demonstration, the room was full of NICU nurses. They were not pleased. Certain workflow steps felt out of sequence, and the interface required multiple clicks — more than the pace of neonatal care allows. The problems were not technical failures. They were observation failures. The team had designed around the task, but not around the shift.

In the better version of that story, the team schedules a field visit in the early project phase. Two team members spend a morning in the NICU observing real workflow: which tasks get interrupted, how nurses navigate existing equipment under time pressure, and what a gloved, one-handed interaction at the bedside looks like. That input is entered into the requirements before any interface decision is locked. The demonstration still happens — and the nurses in the room recognize their workflow in the design.

The same cost appears in enterprise software. A team ships an admin console for a production workflow that no one had observed. Security teams perform every key action by keyboard shortcut. The UI never supported it. Two more sprints follow to build what a single observation session would have surfaced in an afternoon.

One thing to remember: Every unobserved assumption becomes hidden verification work. Extra clicks, missed shortcuts, late redesign, training effort, or complaints after release. The later it appears, the harder it is to change.

How teams ship this: Schedule at least one field observation session before the first interaction design review. Write a one-page observation brief: the question you are trying to answer, the tasks to watch, and the rule — do not guide or correct the user. Require that at least one insight from that session appears in the design rationale before the review opens.

Field Observation Brief · one page completed before every observation:

Contextual Design by Karen Holtzblatt and Hugh Beyer
The core method — watching real work in context and building design models from what is observed — applies whether the artifact is a NICU module, a surgical robot, or a deployment dashboard. The book remains one of the most complete guides to turning field observation into actionable design input.

Which part of your current design looks simple only because no one has watched it in use, where it will live?

The next time you are somewhere you know well — a kitchen, a commute, a waiting room, …

spend two minutes watching it as if you had never been there before.
What do people actually do, versus what the space or process assumes they will do?
What small adaptations happen that nobody designed for?

Do not try to fix anything. Just notice what the environment missed.

AI can make field observations easier to prepare and analyze, but it cannot replace being there.

A model can help draft an observation plan, turn raw notes into a first draft of a workflow map, cluster repeated friction points, compare observed steps against a task analysis, and generate follow-up questions for a formative study. That can reduce documentation effort and make hidden patterns easier to see across multiple sessions. Current UX tools also increasingly offer AI-assisted synthesis, summarization, and analysis, which can speed up the path from raw material to structured evidence.

The risk is polished distance. AI can make a late synthesis look complete even when the team never observed the incubator-side reach, the one-handed interaction, the alarm interruption, the cleaning step, or the workaround that makes the current workflow survivable.

The calm use of AI strengthens the observation loop. Let AI prepare the checklist, structure the notes, and highlight contradictions, while keeping the source of truth close to the observed work: timestamps, task context, artifacts, user role, environment, interruptions, and unresolved design questions.

The rule is simple: do not ask AI to imagine the workflow before the team has observed it. Ask AI to help preserve what was actually seen.

Best,
Andreas Walden

Share this with someone designing for clarity.