Products judged like people

After two weeks with a dinner planner, a household member says: "it forgot Leena's cashew allergy." They do not say "the preference store failed to promote endorsed knowledge." They say forgot, the same verb they use for a distracted colleague. Another says: "it keeps interrupting with suggestions I did not ask for." Not "recommendation frequency exceeded posture bounds." Interrupting.

These are relationship evaluations. As products behave more intelligently, people apply social cognition whether designers intend it. Competence, reliability, judgment, memory, and recovery arrive as verbs in reviews and support tickets.

Helpful context: Trust across time and cognitive posture are architectural homes for many relational complaints. Grounded verbs keep product language honest about what the system knows, remembers, and learns.

Map the complaint to the architecture

Designing as if screens alone matter leaves relational failures without owners. Relational language is specification input, not imprecise whining.

When a household says "it understands me," the architectural home is interpretation and context as a feature. When they say "it forgot," look to knowledge endorsement and memory UX. "Interruption" maps to posture initiative and timing. "Would not listen" maps to Collaboration Contract mode mismatch. "Gave up" maps to refuse and defer and recovery. Product briefs that list screens and flows without memory policy, recommendation timing, and recovery paths will ship relational defects.

User language	Architectural home
It understands me	Interpretation, context as a feature
It forgot	Knowledge endorsement, memory UX
It interrupted	Posture initiative and timing
It would not listen	Collaboration Contract mode mismatch
It gave up	Refuse and defer, recovery

This mapping matters because relational verbs are precise enough for design work. Teams sometimes dismiss "forgot" as vague user language. In practice, "forgot" almost always means the product acted without a durable fact the person believed was settled. The design response is endorsement visibility, exact enforcement, and a contest path. "Interruption" almost always means recommendation arrived without grounds or at the wrong contract mode. The design response is posture bounds and evidence on the suggestion. Training support to translate relational verbs into architecture tickets improves retention more than training users to speak like engineers. Products that dismiss relational language as unstructured feedback miss the most direct signal their users provide. The household is already doing the diagnostic work by naming the failure mode in conduct terms.

What "forgot" and "interrupted" look like in the Meal Companion

Leena's cashew allergy is endorsed knowledge: durable, enforced in deterministic validation. If the plan includes cashews, the household is right to say "forgot" even if the UI looks polished. The fix is memory discipline, exact checks, and visible endorsement. Arguing semantics with users who speak in human verbs wastes trust.

When mushrooms reappear after a stated dislike, the failure may be salience policy, missing promotion flow, or context mistaken for knowledge. Each maps to a different repair. Generic "we improved the model" does not answer "forgot."

These are not descriptions of interfaces. They are descriptions of relationships.

Thoughtware: Designing in the Intelligence Age · Ch. 4

After CritiquePlan flags Tuesday's stir-fry, RecommendMealSubstitution may propose a swap with trade-offs. That is welcome initiative when the household is reviewing the plan. The same recommendation mid-utterance while the household is still stating constraints feels like interrupting. Same feature, different posture timing. Products judged like people need designed cadence for initiative: cap frequency, attach grounds, respect refusal, and instrument accept and reject as learning signals rather than as proof the model is helpful.

What changes in the product brief

Information Age briefs emphasise screens, flows, and visual hierarchy. Intelligence Age briefs add memory policy, recommendation timing, contested overrides, and recovery after error. Brand becomes conduct over repeated use. A calm visual identity cannot compensate for interruption without grounds or memory without contest paths. This does not imply consciousness or personhood. It implies that product quality now includes judgment architecture underneath the window. See beyond the interface. Every screen still matters, but the relational layer underneath determines whether the product feels trustworthy over repeated use. A settings page that lets households inspect endorsed knowledge and contest stale preferences is as important as the calendar view that displays the weekly plan.

Product leaders need fluency in judgment architecture because relational failures are product failures. Hiring for visual craft alone leaves "it forgot" and "it interrupted" without owners. The Intelligence Product Leader role exists because these qualities are no longer optional extras. Roadmaps need owners for memory, posture, and contract mode alongside design systems.

Grounded verbs keep marketing honest. "Remember" maps to endorsement. "Understand" maps to interpretation plus context. "Interrupt" maps to posture timing. "Care" maps to recovery and restraint. Marketing claims must match Collaboration Contract modes. Promising control while auto-applying produces "it would not listen" even when demos look compliant.

Collecting and using relational signal

Support verbatim, review language, and session replay comments are architecture inputs. Tagging each complaint to memory, posture, contract, or recovery produces actionable backlog items. Teams that treat relational language as vague dissatisfaction tune prompts randomly instead.

Reviews reward conduct over time. A household forgives an ugly diff view that preserves Thursday when Tuesday was wrong. They churn from a gorgeous calendar that silently violates allergy on week three. Relational metrics complement behavioural ones: contested overrides, recovery success after error, and promotion approval rates predict retention alongside NPS from visual polish. A product that recovered well from a wrong substitution on Thursday earns goodwill that visual polish alone cannot buy. Relational metrics capture that recovery because they track conduct rather than feature completeness alone.

Design critiques benefit from a relational pass. "What would the household say if this shipped for three weeks?" produces more useful feedback than visual QA alone. "Forgot," "interrupted," and "would not listen" are design prompts, not jokes. Customer success training that teaches clients how to contest memory and read diffs keeps households engaged. Households trapped in transcript theatre churn with relational verdicts that never reach the architecture team.

Usability tests benefit from week-two scenarios. Asking participants to correct one day, promote one preference, and refuse one recommendation produces relational verbs during the session. Tagging each verb to memory, posture, or contract in the debrief turns user research into architecture input. First-impression tests alone miss the judgments that determine loyalty, because products judged like people fail on conduct across sessions, not on first click.

Common mistakes

Arguing semantics with users who say "forgot" wastes trust. The response is to fix endorsement, salience, or enforcement, and tell the household what changed in the product rather than what the database calls their preference.

Another common pattern adds personality to chat copy while leaving memory amnesiac. Personality without conduct produces charming failure. Posture bounds initiative, restraint, challenge, and recovery, and those bounds are what users describe when they say "thoughtful" or "pushy."

Support playbooks that respond with generic "reset your preferences" waste the relational signal the user already provided. When support hears "it forgot," the playbook asks whether endorsed knowledge was missing, whether context expired correctly, or whether promotion failed. When support hears "it interrupted," the playbook asks about posture timing and recommendation grounds. Splitting visual craft from conduct guarantees orphaned "forgot" tickets without architectural owners.

What to do next

The pattern this note describes runs through every product surface: support tickets carry relational verdicts, review language carries architectural signal, and marketing claims succeed or fail on conduct over time. Collecting five verbatim user complaints and mapping each to memory, posture, or contract reveals design blind spots the roadmap has not named. Understanding how products handle the opening move connects relational judgment to the question of cognitive effort.

See intelligence over aesthetics and consumers direct intelligence.

Read next: Low-cognitive first draft.