Adaptive AI feedback loop
A human-centered AI product should collect useful feedback during interaction and use it to improve future behavior.
29 张卡片
AI risk map before build
Before building an AI feature, map the user context, affected parties, trustworthiness needs, and potential harms.
AI-assisted iteration cycle
AI-assisted iteration works best when generation is paired with feedback, scoring, and reusable learning.
Context-level accessibility testing
Accessible components still need testing in the actual page or workflow context where users encounter them.
Correction affordance for AI output
AI output should be easy to edit, refine, undo, or recover from when it is wrong.
Discovery decision gate
Discovery should end with a decision about whether the evidence justifies moving into a more expensive exploration phase.
Eval-driven AI development
An AI product should define how success will be evaluated before the team invests in deeper prompt, model, or workflow work.
Feedback share card
A feedback share card packages an idea as a short public question instead of a polished announcement.
Human-calibrated eval scoring
Automated eval scores need human calibration so the measured result still matches the product question.
Idea capture loop
An idea capture loop turns a generated idea into a small set of next actions before enthusiasm fades.
Knowledge card distillation
Knowledge card distillation turns useful product lessons into short, source-linked, reusable cards.
Live continuous improvement
A live product still needs research, testing, accessibility checks, quality assurance, and performance metrics.
Problem before solution
A discovery process should turn a proposed solution back into the underlying problem before committing to build.
Production log to eval case
Useful production logs can be converted into eval cases so real failures become repeatable tests.
Production readiness guardrail
An AI prototype should not move to production until security, privacy, testing, misuse, and operational limits are explicitly reviewed.
Real user needs principle
Product decisions should be anchored in real user needs rather than internal assumptions or polished generated concepts.
Reversible product bet
A reversible product bet is small enough to test without trapping the builder in a large commitment.
Review status ladder
Draft, review, reviewed, published, and archived states separate early extraction from trusted public knowledge.
Riskiest assumption prototype
A prototype is most useful when it focuses on the riskiest assumption instead of recreating the entire product.
Scoring gate
A scoring gate gives a generated idea a lightweight decision point before it receives more time.
Search query persistence
Search results are easier to review when the original search term remains visible and editable.
Server-mirrored validation
Client-side validation can improve usability, but server-side validation must enforce the actual rule.
Shrink or kill decision
A weak idea does not always need more work; it may need a smaller test or a clear stop.
Source-first card review
A Knowledge Atlas card should be reviewed against its source reference before it becomes a trusted browsing or search result.
Source-linked personal knowledge
Source-linked personal knowledge keeps reusable notes connected to the evidence or experience that produced them.
Task-specific eval objective
A useful AI eval starts with a task-specific objective that names what the system must do well in its real product context.
Throwaway alpha learning
Alpha work should be judged by what it teaches, not by whether the prototype survives into production.
Validation brief
A validation brief compresses an idea into target user, problem, proof needed, and the smallest useful test.
Workbench commitment log
A workbench commitment log turns selected ideas into tracked projects with visible next actions.