Overview
Turning AI prompting into a reusable UX production system
The AI Pattern Library organizes reusable prompt patterns around recurring UX production needs: project continuity, evidence control, artifact truth, stage governance, design-state modeling, contextual action placement, and novice-safe technical guidance.
Rather than treating prompts as one-off commands, the library treats them as a repeatable design system. Each pattern defines the problem it solves, when to use it, what risk it controls, and how it connects to real UX artifacts.
Example Callout
Structured Session Handoff
End-of-session request
We need to pause here. Generate a concise continuation prompt that will allow the next session to resume from this exact point without a broad recap.
Generated continuation prompt
Title this session: “[New Session Title]”.
This is a direct continuation of “[Previous Session Title]”.
Resume from the exact handoff point where we paused: [active workstream, stopping point, and next action].
Carry forward these locked decisions: [Decision 1], [Decision 2], [Decision 3].
Do not recap the entire project unless asked. Begin directly with the active task.
Example resumed response
The previous session ended after the volunteer scheduling rules were locked for the community garden portal. I’ll resume with the unresolved conflict-state logic for overlapping shifts and waitlist invitations.
The Challenge
AI can accelerate UX work, but unstructured prompting creates production risk
AI tools can produce useful drafts, summaries, and interface ideas quickly. The problem is that speed alone does not make the work trustworthy. Without structure, AI-assisted UX work can drift across sessions, blur evidence with inference, imply artifacts that do not exist, advance without approval, or create visually plausible screens before the interaction logic is understood.
The challenge was to create a reusable method that makes AI-supported UX work more reliable, reviewable, and portable across projects.
Method
Designing patterns around risks, not just tasks
I developed the library by identifying recurring failure points inside long-running AI-assisted UX work, especially while developing DPMA. Each recurring issue became a candidate pattern: a reusable structure for guiding AI behavior, preserving human authority, and producing clearer UX artifacts.
This shifted the work from “prompt writing” to pattern-system design. The goal was not to make AI sound better. The goal was to make AI-supported UX production behave with more discipline.
Design principle: The library is organized around UX production risks: context drift, overclaiming, missing artifacts, premature advancement, role confusion, uncontrolled mid-project changes, incomplete states, and unclear technical guidance.
Pattern System
Five categories for structured AI-assisted UX production
The current library is organized into five working categories. Together, they describe a production model that is evidence-aware, artifact-centered, human-controlled, and practical enough to support real UX workflows.
Continuity
Uses structured session handoffs to keep long-running work from restarting, drifting, or reopening locked decisions between sessions.
Evidence + Trust
Separates observed evidence, inference, assumption, unknowns, and confirmed artifacts.
Governance
Preserves human authority, role boundaries, approval gates, and controlled input changes.
UX Production
Structures screen analysis, component states, artifact schemas, and contextual action placement.
Technical Enablement
Turns technical guidance into novice-safe steps with expected results and recovery paths.
Example Callout
Evidence Control in Product Review
Prompt excerpt
Analyze the available material and separate all findings into four categories: Observed Evidence, Reasonable Inference, Assumption, and Unknown / Needs Validation. Do not present inferred or assumed material as fact.
Partial response excerpt
- Observed Evidence: The kiosk screen shows three primary tasks: reserve equipment, scan a return, and report a damaged item.
- Reasonable Inference: The task grouping appears intended to support quick self-service in a shared facilities environment.
- Assumption: The screen may be used by both staff and visitors, but the user roles have not been confirmed.
- Unknown: Whether the layout reduced service-desk interruptions cannot be confirmed without usage or support data.
Featured Patterns
A curated public set of reusable prompt-pattern cards
The public-facing version focuses on a curated set of high-priority patterns. These cards tell a coherent story: AI-assisted UX work can be structured, governed, evidence-aware, artifact-centered, and human-controlled.
Structured Session Handoff
Creates a concise end-of-session continuation prompt that can be pasted into the next chat to resume from the exact handoff point.
Observed / Inferred / Assumed Evidence Control
Separates known evidence, reasonable inference, assumptions, and unknowns so AI analysis stays defensible.
Artifact Truth Enforcement
Prevents the system from implying that files, visuals, metrics, or deliverables exist unless they are actually present or confirmed.
Approve → Confirm Gate
Separates approval of an artifact from confirmation to advance, reducing accidental progression and downstream rework.
Orchestrator vs. Operator Separation
Separates coordination, operational execution, domain production, and human authority in multi-agent workflows.
Any-Stage Input Injection
Allows new project inputs to be classified and incorporated without breaking governance or traceability.
Component × State Matrix
Breaks an interface into components, purposes, states, behaviors, restrictions, and dependencies.
Step → Expected Result → Recovery Path
Turns technical instructions into exact steps with success checks and failure recovery guidance.
Example Callout
Component × State Matrix
Prompt excerpt
Analyze this screen or wireframe as a component/state system. Create a table listing each component, its purpose, visible states, allowed behaviors, disabled/restricted behaviors, and dependencies.
Partial response excerpt
- Room Schedule Panel: empty day, partially booked, fully booked, selected booking, and maintenance-blocked states.
- Booking Action Row: available, pending approval, confirmed, canceled, and conflict-warning states.
- Dependency: the confirm action depends on room availability, user permission, and conflict resolution status.
Origin
Extracted from real DPMA production work
The library grew out of practical work on DPMA, where repeated AI/UX workflow problems had to be named, stabilized, and reused. The same issues kept appearing: how to continue across long sessions, how to separate evidence from assumptions, how to prevent false artifact claims, how to preserve human authority, and how to make technical instructions usable for a non-developer working near code.
By converting those recurring moments into reusable patterns, the library became a portable method system rather than a private set of notes.
Outcome
A working library and public-facing method model
The result is a structured pattern library with ten documented prompt patterns, five categories, metadata, risk controls, example artifacts, and a curated public-first card set. It is still a working library, but it is organized enough to support both practical reuse and portfolio presentation.
For the portfolio, the value is not the number of prompts. The value is the method: turning AI-assisted UX work into a governed, evidence-aware, artifact-centered production system.
Why It Matters
Prompting is stronger when it behaves like UX architecture
The AI Pattern Library shows how senior UX judgment can shape AI workflows before they become products. It reframes prompting as a design discipline: define the user problem, control the risk, structure the artifact, preserve authority, and keep the output reviewable.
This work extends my AI/UX practice beyond DPMA into reusable production methods that can support research, architecture, design, technical enablement, and portfolio reconstruction.