Is Arco saying the Steward's surface should be ugly?

No. It is saying the surface should be stripped, and that what remains should receive more design care than the customer product, not less. Hierarchy, contrast, spacing and type are the disciplines that make state legible at a glance; they stay. Decoration — anything that carries no state and accepts no command — goes. A stripped surface done with care is not ugly. It is plain, and plain is what a verification surface is supposed to be.

Why not reuse the customer product's design system for the Steward console?

Because the two surfaces are optimised for opposite things. A customer product is optimised for a person who is exploring and can leave; its design system carries brand, delight and persuasion. The Steward's surface is optimised for a person who is checking and cannot leave; anything that persuades is noise. Shared components are fine where they carry state legibly. A shared palette that spends red on brand is not.

How does this relate to the UI Tax if the Steward's surface is built in-house?

The UI Tax entry describes a premium paid to vendors. The memo's use is by extension: the same cost — interface development, training, attention spent on the interface rather than the work — is paid internally on the one interface the business keeps. Decoration is the part of that cost that buys nothing, and the audit removes it.

Does removing decoration risk making the surface harder to learn?

A Fixed-Function Surface is learned once and then glanced at. Learnability comes from consistency — one meaning per signal, one place per state — not from ornament. Ornament gives the eye more to learn, not less: every decorative element is one more thing on the surface whose irrelevance the Steward has to confirm before moving past it.

Decoration Is Overhead

The corpus has a term for what human-facing software costs. The UI Tax is the cost premium embedded in software products designed for human cognition — including graphical interface development, permissions infrastructure, user training, and the subscription markup charged for enterprise UX — which is entirely redundant in an agentic stack where agents require no interface to operate. De-SaaS-ing is the discipline of refusing to pay it: integrate at the data and logic layer, and let the interface that was built for a human seat go unbought.

One interface survives that discipline, because one human remains. The Lexicon entry names a premium paid to vendors for human-facing software; this memo extends it, deliberately, to the same cost paid internally on the one human-facing surface an autonomous business keeps. The Steward's surface is the last place where the UI Tax is still paid, and it is paid in the currency the Stewardship Model has the least of: a single person's attention, at a tempo set by the system rather than by them. That changes what the tax buys. On a customer product, the tax buys retention, because a customer who finds the product pleasant stays. On the Steward's surface, there is no retention to buy. The Steward is not choosing whether to use the surface. The only thing the tax can buy there is verification speed, and every element that does not buy it is overhead.

Decoration and discipline are not the same thing

The claim has to be made carefully, because a careless version of it licenses ugly software and the corpus has already refused that. The UI That Runs the Business held Steward Experience to the same rigour as customer-facing product design, on the grounds that its quality is a direct input to the business's measured autonomy scores. That commitment stands. What this memo does is say what the rigour is for.

Design applied to a surface divides into two kinds of work. The first kind makes the surface legible: hierarchy, so that the most consequential state is the largest and the first thing the eye lands on; contrast, so that an abnormal condition is distinguishable from a normal one at a glance and under poor light; spacing and alignment, so that related states read as a group and unrelated states do not; typography, so that a number can be read at the size it is rendered. Every one of these is a discipline usually filed under aesthetics, and every one of them survives on the Steward's surface, because each one shortens the time between the Steward's eye arriving and the Steward knowing the state. They are instruments of the Fixed-Function Surface, not exceptions to it.

The second kind makes the surface pleasant: a palette chosen for brand, a transition that eases a panel in, a header carrying a logo the Steward has seen every day for a year. None of this shortens the time between the glance and the knowing. Some of it lengthens it: a brand colour that happens to be red has taken red away from the one meaning the surface needed it for, so every red on the screen now costs the Steward a second look; an animation that eases state in has delayed the state by the length of the ease, on a route whose whole function is to be faster than the system. This is decoration, and on a verification surface decoration is Coordination Tax in pixels — overhead that exists to make the interaction feel considered, at the cost of the interaction's one function.

The line between the two is not taste. It is a test that can be run on every element: name the state it carries or the command it accepts. If it does neither, it is decoration. If it does one, it stays, and it is held to the legibility disciplines as strictly as the surface can manage.

Why decoration accumulates by default

The Steward Experience entry named the mechanism. Every autonomous business applies design discipline to its customer product because a customer will leave if the product is bad, and no external party applies equivalent pressure to the Steward's interface. That absence explains neglect. It also explains its opposite, which is what this memo is about. With no churn signal, there is nothing to stop a team from treating the Steward's surface as a product to be made attractive, borrowing the customer product's design system wholesale, and shipping a governance console that looks like a marketing dashboard. It does not fail. The Steward uses it, because they have no alternative. The cost shows up two memos later as the reading the corpus already knows: a Steward who verifies less because each verification costs more, and a metric that reports the drop as autonomy.

Overhead Is a Design Choice made the general point that coordination cost is not inherited, it is chosen, one decision at a time, by people who did not think of each decision as adding overhead. Decoration on the Steward's surface accumulates the same way. Nobody decides to slow the Steward down. Somebody decides the empty state needs an illustration.

What the chip principle says about it

A chip has no decoration because there is nowhere for decoration to go. Every component has a function; anything without one is a defect in the layout. The Fixed-Function Surface adopts the standard in full. The surface carries state and accepts commands, each element does one of those, and the legibility disciplines are applied to the elements that remain so that each is read in the least time its content allows. What the surface looks like afterwards is not a question the memo answers, because it is not a question the verification route asks.

The control-room practice the previous memo cited reaches the same place from the other side. Normal conditions in grey; colour only where attention is needed; no motion unless something moved. Colour on such a display is the signal, and a screen with none is a system with nothing to report, which is the fastest state a Steward can verify: one glance, no exception, move on. The surface did not lose its design. It lost everything on it that was not shortening that glance.

The Operator's Verdict

An operator can audit a Steward surface for decoration in an afternoon, and the audit is the same as the one for Coordination Tax. Walk the surface element by element. For each one, write down the state it carries or the command it accepts. Remove everything for which the line is blank. Then apply hierarchy, contrast, spacing and type to what is left, with more care than the customer product gets, because the customer is browsing and the Steward is checking.

The result will be plainer than the operator's other software and it will be the most designed thing they own, because design on the Steward's surface is measured in one unit — the time between the glance and the knowing — and nothing else on the screen is allowed to add to it. Every addition is a slower check, every slower check is a check the Steward is more likely to skip, and every skipped check is a step toward a metric that reports the skipping as autonomy.

Technology changes what is possible. What carries state determines what stays on the surface.

KEY TAKEAWAY

Why does Arco argue that decoration on the Steward's surface is overhead, and what does it keep?

Arco's UI Tax names the cost premium embedded in software designed for human cognition, redundant in an agentic stack where agents need no interface. One interface survives De-SaaS-ing because one human remains: the Steward's surface is the last place an autonomous business pays the UI Tax, in a single person's attention at a tempo set by the system. On a customer product the tax buys retention; on the Steward's surface there is no retention to buy, so the only thing it can buy is verification speed, and every element that does not is overhead. The memo separates two kinds of design work. Legibility disciplines — hierarchy, contrast, spacing, typography — shorten the time between the Steward's glance and knowing the state, and survive as instruments of the Fixed-Function Surface. Decoration — brand palettes, transitions, logos — does not shorten it and sometimes lengthens it, by taking a colour away from its one meaning or delaying state by the length of an animation; it is Coordination Tax in pixels. The test is per element: name the state it carries or the command it accepts, and remove anything that does neither. Decoration accumulates because no churn signal disciplines the Steward's surface, so teams borrow the customer product's design system wholesale; the cost surfaces later as a Steward who verifies less and a metric that reports it as autonomy. The extension of UI Tax from vendor premium to internal attention cost is marked explicitly in the body. The result is plainer than the operator's other software and the most designed thing they own.