This is what your agent ships when nobody is judging it.

Ive's lens: a placeholder is not a label, it is a label that deletes itself the moment you need it. The reader types three characters, looks away, looks back, and the field no longer says what it wants. Worse, a half filled form becomes unreadable: the answered fields have lost their questions. He does not accept a control that stops explaining itself under use. If it fails while being used, it does not work.

What changes: each input gains a persistent label above it, and the placeholder either goes or drops to a format hint that is no longer carrying the field's identity.

What he would ask: come back to this form half finished and tell me what the third box wanted. If you cannot, who did you build the placeholder for?

Risk: labels above every field make the form taller, and a taller form reads as more work. Move 4 is what pays that back.

Ive's lens: the outline was removed because it was thought ugly, and nothing was put in its place. For anyone on a keyboard that is not a tidier form, it is a form with the cursor switched off. Beauty that costs the user their position is a misunderstanding of what beauty is for. It cannot look resolved by breaking the one cue that says where you are.

What changes: a two pixel focus ring returns, with an offset so it reads clearly against the field edge, at no less than 3:1 against the surface behind it.

What he would ask: tab through this form without touching the mouse. Where are you? If you cannot answer instantly, what did removing the outline actually buy?

Risk: a ring drawn tight to the border can look like a validation state. The offset is what keeps focus and error legible as different things.

Ive's lens: ten pixel type, set in red, floated away from the field it judges. Three failures at once: too small to read, too far to connect, and colour doing work alone for anyone who cannot see it. An error is the most urgent sentence on the page and it has been given the least of everything. This is not restraint, it is an apology hidden where it will not be found.

What changes: the message moves to thirteen pixels, sits directly beneath the field it belongs to, and gains a text prefix so red is no longer the only thing saying something is wrong.

What he would ask: if this reader cannot see red at all, how do they learn the email was rejected?

Risk: an adjacent message shifts the fields below it when it appears. Reserve the line so the form does not jump under the reader's hands.

Ive's lens: three of these fields are not required and are asked anyway, because they were cheap to add and someone downstream might want them. Every one is a toll charged to a person who has not yet decided to trust you. Simplicity is not what is left when you run out of ideas, it is deciding what does not deserve to be there. Four questions, asked once, is the whole form.

What changes: company, phone and job title come out of the signup path and move behind a clearly named optional step after the account exists. The form is name, email, password, confirm.

What he would ask: which of these do you need to create the account, and which do you merely want? Ask the second set later, or admit you are charging for them now.

Risk: the optional fields still need somewhere to live, and a step nobody returns to collects nothing. Put them where the answer is actually useful.

Ive's lens: a column of money has one job, to be read down its length and compared. Left aligned proportional figures defeat that job on purpose: the decimal points wander, a 1 takes less room than an 8, and two amounts a reader wants to weigh against each other never share an edge. He does not treat that as a neutral choice. If the column will not let you compare, it does not work, and on his terms that settles what it is.

What changes: the amount column moves to the right edge and switches to tabular figures, so every decimal point sits in the same place and the digits stack into real columns. No value changes.

What he would ask: what is a column of money for, if not to be compared down its length, and why have you made that the one thing it cannot do?

Risk: right aligned numbers can drift from a left aligned header label. Right align the header cell too, so the title sits over its own column.

Ive's lens: eleven pixels packed into a twenty six pixel row looks efficient in a screenshot and fails the moment someone has to read a screenful of it. He does not read small and tight as sophistication. Text set below what the eye can take in does not work, so this is not the restrained choice it presents itself as, it is the unreadable one. Room to read is not wasted space.

What changes: the body text goes to thirteen pixels, the rows open to roughly forty four, and the header lifts off the body with its own rule, so the eye can carry a single row across the table without losing the line.

What he would ask: who actually has to read a full screen of this, and have you ever tried to, or did it only need to look tidy from a distance?

Risk: taller rows show fewer at once, so a long list means more scrolling. That is the right trade for a table meant to be read rather than glanced at, but confirm the density still suits the task.

Ive's lens: a bare icon asks the reader to guess, and what a control does is not a thing anyone should have to guess. The lesson he took from the controls that had to be relearned is blunt: a function a person cannot identify is not available to them, however clean the glyph looks. An eye and a sheet of paper are decoration until they are named.

What changes: each action keeps its glyph and gains a real name and a visible text label, so the row reads view and receipt in words instead of leaving the reader to hover and hope.

What he would ask: can a first time reader say what each of these buttons does without clicking one to find out?

Risk: labels widen the actions column and cost horizontal room. If space is genuinely tight, keep the labels and drop a less used action rather than hiding every name again.

Ive's lens: the vertical rules and the striping are scaffolding, and here the scaffolding is louder than the building. Every cell is boxed and every other row is filled, so the eye has to push through a grid to reach the one thing on the page that matters. Decoration must not outrank the content it frames. A rule earns its place only where the layout would fail without it, and most of these do not.

What changes: the vertical gridlines and the zebra fill come off, and a single hairline stays under each row. Nothing frames the data any more, it only separates one row from the next.

What he would ask: which is louder right now, the numbers or the lines drawn around them, and which did the reader come for?

Risk: strip too much and a wide row can lose its tracking across the table. The one horizontal hairline is what keeps the eye on its line, so keep that and only that.

Ive's lens: the timestamps sit at 1.79:1. That is not quiet, it is off. It reads as restraint in a screenshot and disappears entirely in daylight on a phone held at arm's length, which is the only place this screen ever runs. Muting text until it cannot be read is not taste, it is a decision to publish something and hope nobody needs it.

What changes: the meta text moves from #3d414b to #8b929e, taking it from 1.79:1 to 5.62:1 against the screen, which clears AA for body text with room to spare.

What he would ask: read the third timestamp outdoors. If you had to tilt the phone, who exactly was the low contrast serving?

Risk: lifting every secondary line can flatten the hierarchy. Raise the meta to legible and leave the primary above it, so the order still reads.

Ive's lens: a decorative gradient occupies the top third of a screen whose entire purpose is one figure. The ornament is bigger, brighter and better lit than the data, so the eye lands on the flourish and has to work its way back to the point. Decoration must not outrank the content it frames. This is the clearest case of it on the whole page.

What changes: the blob shrinks and drops in intensity, and the figure grows and moves to full white, so what the person opened the app to see is the first thing they see.

What he would ask: what did this person open the app to find out, and is that the first thing their eye reaches?

Risk: strip the warmth entirely and the screen turns clinical. The gradient can stay, it simply cannot be the loudest thing on the screen.

Ive's lens: read the times down the list: 07:12, 12:40, 18:05, 16:30, 21:10, and a sleep entry from the night before sitting at the bottom. The evening walk is filed above the afternoon ride, and the night that started the day is last. There is no rule here, which means the reader cannot skim, cannot predict where anything is, and has to read all six rows to be sure of any one of them. A list whose order means nothing has made the reader do the sorting, and on his terms a thing that hands its work to the person using it does not work.

What changes: the rows sort chronologically, so last night's sleep opens the day and the entries run forward to the evening stretch. No row is added, removed or restyled; they are only put in the order they happened.

What he would ask: without reading every line, tell me what you did most recently. If you had to check all six, what was the order doing for you?

Risk: time is the right order for a log, not for every list. If someone opens this to find their biggest effort rather than their latest, sorting by figure serves them better, and the two orders cannot both be the default.

Ive's lens: the action bar floats over the list and permanently hides the last row. The screen is telling the reader that the list ends where the bar begins, and that is not true. It is a small lie about what fits, told for the sake of a floating control. He would not accept a layout that hides its own content to look uninterrupted.

What changes: the scroll area gains bottom padding equal to the bar's height, so the final row clears the button instead of living underneath it.

What he would ask: scroll to the end of this list. Is that the last entry, or is that only where the button starts?

Risk: padding leaves an empty band under a short list. That is the correct trade against a row nobody can read.