TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 18—22 · POSITIONAL · HELD

The polish disarms you

A finished-looking artefact is the highest-risk kind of output, and it is checked line by line.

The dilemma

Two outputs from the same model. One is a messy paragraph of half-formed reasoning. The other is a formatted document: headings, bullets, a summary table of site constraints. Which one do you check harder? Be honest. Now ask which one deserves it.

The choices

Let the finish set how hard you check: messy gets questioned, tidy gets pasted. Or turn it round on purpose. The more finished a thing looks, the more deliberately you check it, because finish is the cheapest thing a machine produces and the strongest signal your brain accepts.

The consequence

Let the polish decide and at some point you will paste a polished error into a client document: a site constraint that was never true, a byelaw clause that does not exist, set out in a table that made it look settled. Turn the reflex round and the tidy artefact gets the treatment a tidy junior's work gets. It is read line by line, precisely because it was easy to trust.

The case

A student asks a model for the setback rules for a plot in a Karnataka town. It returns a clean table (front, side, rear, by plot width) with a confident note on FAR. It looks like a page from the byelaws. She pastes it into her site analysis. The tutor asks which edition of the zoning regulations it came from. There is no edition. The table was simply the most likely-looking table, and looking likely was all the authority it had. A messy answer would have been checked.

Our position

We stay sharp while arguing with a machine and go slack the moment it hands us an artefact. We mistake how finished a thing looks for how true it is. The Lab calls this the artefact paradox, and treats finished-looking output as the highest-risk kind of output, not the lowest. The position rests on an observed pattern and on what the Lab sees at the desk. The cause is not yet established.

Why we hold it

Anthropic's February 2026 analysis of several thousand conversations with its own chat model reported that when the model produced an artefact (a document, a block of code, a formatted list), users showed less visible checking inside the conversation — noticing missing context, checking facts, questioning the reasoning — by a few percentage points each. The study is observational, covers one company's model, and sees only what happened inside the chat. It cannot say why the checking dropped, or whether people checked elsewhere. These were general users, not designers. We hold the position anyway, because the pattern matches what the Lab sees in studios, and because the artefact is also the thing most likely to be forwarded.

The strongest objection

The evidence is correlational. "Conversations with artefacts contained less visible checking" does not show that the polish caused anyone to check less. The tasks that produce artefacts may simply be the tasks people check less — routine ones, or ones where the checking happens outside the chat, in a drawing or on a site the study could not see. Trained users may behave differently from the general users measured. Until somebody varies the polish and holds the task still, "the polish disarms you" is a hypothesis with a good story, not a mechanism.

What would make us revise it

A controlled study that varied how finished an output looked, held the task constant, and found no difference in checking — or found the difference vanished in trained users — would move this card from a standing position to a first-semester warning. The opposite result would make it a mechanism. Either way we review it against new studies every edition.

Try it

Catch one moment this week when a tidy, formatted output made you check it less than a messy one would have. Write down what you skipped. Then explain, in two sentences, why completeness and accuracy are not the same thing, and where in that output the gap was.

Take it to crit

When a student is handed a neat, finished-looking deliverable, do they check it more, not less? Ask them what they checked in it, and what made them decide to check that.

How it works

Formatting is a pattern the model learnt like any other: what finished documents look like. It costs the model nothing and says nothing about whether the contents are true. Your brain, though, reads finish as effort and effort as care. That is a reasonable shortcut among humans, and a likely trap with a machine for which finish is free. The fix is a procedure, not an attitude. Decide before you read what you will verify in any artefact, so the polish cannot lower the bar afterwards.

What this idea builds on

What this idea opens up

Sources

Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON