Interaction Engineering · 1 / 10

It Runs. Now What?

Last chapter trained you to spot「what's ugly」. This one trains「what's hard to use」. AI-shipped apps have every feature and respond to every click, but they feel rough everywhere. This lesson lays the chapter foundation: interaction bugs hide deeper than taste bugs. Then we take a theory checklist and catch five in the act.

Opening thesisConsiderate checklistSpot-the-flaws labSelf-check vote
A glance vs a walkthrough

Taste bugs show fast: open the page, colors clash, spacing is cramped, you frown in three seconds. Interaction bugs hide deep: the two-minute demo looks fine. They flare when a real user deletes the wrong row, drops the network, or opens an empty first screen.

That's the new pit in the Vibe Coding era. AI made「shipping it」easy. Its default interaction stops at「it runs」: blank empty states, alert popups for errors, deletes with no confirm, Save with no feedback. You click twice in review and feel fine. Users want to uninstall after a week.

DimensionTaste bugsInteraction bugs
When they showFirst three seconds after openAfter use: wrong delete, offline, first open
Who finds them firstYou can spot them at a glanceUsers often find them for you—cost is churn
How AI gets them wrongRainbow colors, fighting type sizes—looks offDemo looks fine; every edge case is a trap
How to reviewLook: a screenshot is enoughUse: walk the flow, try every bad case
Looks good or bad shows in a glance. Usable or not only shows when you use it. So interaction bugs hide deeper—and need a checklist as a safety net.
Considerate is a checklist. Flip it and it becomes a flaw list

Where does the list come from? Alan Cooper answered in About Face 4, Chapter 8,「Digital Product Etiquette」. Drawing on Nass and Reeves, he turned the research into one design principle:

「Software should be considerate, like a person.」
Alan Cooper, About Face 4, Chapter 8

Cooper's observation cuts deep: interactive products anger us less from missing features than from being inconsiderate. He listed a dozen considerate traits. We pick the six best for checking AI output. Read each in reverse and you get a flaw standard.

Considerate software…Cooper's meaningAI's default flip side
Is anticipatoryPredicts what you need next and preps itFirst open is blank; next step is pure guesswork
Informs you in timeTells you progress you care about—you shouldn't have to askYou hit Save and nothing moves; a spinner with no word
Doesn't burden you with its problemsDigests tech failures itself; no code-flavored blame dumpGoes offline and throws Error: code 500 in your face
Is confident, but keeps a way backDoes what you say without nagging; if you err, helps recoverEither asks three times before delete, or delete is permanent
Doesn't ask needless questionsRemembers choices it can; sets defaults it canAsks the same question every time you open
Helps you avoid dumb mistakesSees you about to err, quietly steadies you—no scoldingWatches you mis-delete, then pops「operation failed」

That right-hand column is the target for the spot-the-flaws lab next. Don't memorize—catch them once and they'll stick.

Quick division of labor across chapters: Taste Engineering owns looks, this chapter owns usability, Psychology owns how it feels. Each ships one acceptance checklist. Together they're the「three checklists before launch」.

One upside: AI can fix all six. It just doesn't by default. Name the bug and it can prescribe. That's why this chapter keeps drilling「say what you want clearly」.

Hands-on · Five inconsiderate spots—click each one

Below is an AI-generated todo app,「Today's List」. Feature-complete. It runs. It hides five inconsiderate spots, all on this screen. Click whatever feels wrong. Hit one and a case card slides in on the right: which checklist rule it breaks, and how to fix it.

Pick apart「Today's List」 Found 0 / 5
Five flaws sit in the top banner, toolbar, error bar, list, and archive area. Wrong clicks aren't punished
todo.example.com
Aug 11 · 3 left Save
Turn on dark mode? Turn on No thanks Asks every time you open
Error: request failed with status code 500 Confirm
Buy freeze-dried treats for the catToday 18:00 · Shopping×
Pay this month's electric billDue tomorrow · Bills×
Reply to the landlordAlready two days late×
TodayArchived
Case cards land here. Start with what stings most: if a block makes you pause for a second, that block has a problem.
All five found. Look back at the five case cards: none is a missing feature. The app still runs. The bugs are all manners. That's Cooper's point: what angers users is usually inconsiderate behavior; missing features rank lower. How to fix empty states and errors fills the next lesson; gone-for-good deletes wait for Lesson 3 on error prevention and reversible actions.
Save succeeded—do you need a fanfare?

On the flaw list, two diseases get mixed up: no feedback is a bug, and over-feedback is a bug too. Cooper called this out in Chapter 8: interactive products love unnecessary notifications that show off.「Document saved!」modals waiting to be dismissed only interrupt you. Both editors below give Save feedback. Tap the one you find more considerate.

Same Save action, two kinds of feedback A/B duel
Tap a card to vote; the verdict unlocks after
Plan A
docs.example.com/weekly
Weekly draft486 words · Last edit 21:47Save

Wrapped the order-export feature this week. Integration passed; shipped Thursday.

Next week: expand export formats to Excel. About three workdays…

Notice
Saved successfully!
Confirm
Every Save pops a celebration. You must tap「Confirm」before you can keep writing
Plan B
docs.example.com/weekly
Weekly draft512 words · Last edit 21:47Saved

Wrapped the order-export feature this week. Integration passed; shipped Thursday.

Data side filled in a 30-day retention board; metrics aligned with product.

Next week: expand export formats to Excel. About three workdays…

Auto-saved · Just now
Button becomes「Saved」in place; a timestamp sits in the corner. Hands never stop
Cast a vote · What have you been quietly enduring?

You've seen the flaw list. Think of the software you use every day—including what you had AI build. Six kinds of inconsiderate. Pick the one that tortures you most, then see what others put up with.

Six inconsiderate habits—which tortures you most? Cast a vote
Tap a row to vote once. After voting, the distribution shows
Blank empty state—next step is guesswork
Errors full of code jargon
One slip deletes it—gone forever
Tapped with no response—did it work?
Same question every single time
One crash and hours of work vanish
Review interaction—walk it three times, in this order

You have the flaw list. You still need a how-to. Taste review can use screenshots. Interaction review means walking the flow—three times, each as a different person.

PassWho you playHow you walkWhat bugs to catch
FirstPower userFollow the happy path through the main flowDaily friction: Saved-no-feedback, Asks-forever
SecondUnlucky userErr on purpose: delete a row, drop the net, mistype a format, double-submitBlame-dump errors, Gone-for-good, Crash-and-lose-all
ThirdNew userClear data and start from zero. Pause on every first screenBlank empty state, next step is guesswork

Passes two and three get skipped most often—and that's exactly where AI's bugs pile up. Demos always show pass one: smooth network, data ready, path straight. Walk the other two for users and you catch the bugs before launch.

Keep the considerate checklist nearby. On every screen ask: Is this anticipatory? Does it inform? Is it dumping blame? Is there a way back? Asking needless questions? Helping me avoid dumb mistakes? Six questions done—that screen passes.

Quiz · Why do interaction bugs hide deeper?
AI shipped an app that runs. Why do interaction issues slip through most easily? Single choice
Wrong answers get explanations. Keep going until you hit the right one
ABecause AI-written interaction code has more bugs
BBecause interaction issues only show in real use—two review clicks never reach those moments
CBecause users don't care about interaction, only features
DBecause interaction design has no standards, so you can't tell right from wrong
Key Takeaways

Interaction bugs hide deeper: Taste issues show in three seconds. Interaction issues wait for wrong deletes, offline, or first open—and users often find them for you.

「It runs」is only a pass: AI's default interaction stops at「it runs」. What angers users is usually inconsiderate behavior; missing features rank lower (Cooper, About Face 4, Chapter 8).

Flip the considerate list into a flaw list: Anticipatory, informs in time, no blame dump, confident with a way back, no needless questions, helps avoid dumb mistakes. Run all six against AI output.

Review interaction by using it, not only looking: Walk the flow. Try the bad cases: delete a row, drop the net, open a fresh account. The next four lessons unpack each handle—starting with the three states.

Source: Original to Xiaoshan Academy's「Interaction Engineering」series; some interaction principles adapted from About Face 4: The Essentials of Interaction Design.