UX Design Lab
Discuss a project
  1. SignalsRecognize the problem.01 of 08: Signals. Recognize the problem.
  2. OutcomesWhat is different when the work is done.02 of 08: Outcomes. What is different when the work is done.
  3. ApproachAssess, align, implement, iterate.03 of 08: Approach. Assess, align, implement, iterate.
  4. ProofGoverned knowledge for AI.04 of 08: Proof. Governed knowledge for AI.
  5. ServicesFocused ways in.05 of 08: Services. Focused ways in.
  6. EngagementScoped to the problem, not a package.06 of 08: Engagement. Scoped to the problem, not a package.
  7. PavelWho you work with.07 of 08: Pavel. Who you work with.
  8. ContactStart a conversation.08 of 08: Contact. Start a conversation.

AI Product Design

Build governable AI

Design AI around useful work, visible evidence, deliberate authority, human control, and recovery.

Discuss your AI product

Recognize the problem.

If two or more of these sound familiar, start here.

  1. S/01

    The AI feature exists before its role in the workflow is clear.

  2. S/02

    Users cannot tell what the system has done.

  3. S/03

    An agent can take consequential action without an explicit authorization boundary.

  4. S/04

    Model output is fluent, but the evidence behind it is hard to inspect.

  5. S/05

    Failure states depend on prompt instructions rather than product controls.

What changes.

What is different when the work is done — things you can point to, not adjectives.

  1. Change 01

    AI responsibilities are separated from human responsibilities.

  2. Change 02

    Consequential actions have visible authorization points.

  3. Change 03

    Users can see relevant evidence, system state, and recovery paths.

  4. Change 04

    Known failure scenarios have product-level controls.

How we work.

Assess. Align. Implement. Iterate. Applied to the work with the most at stake.

  1. Assess

    Map users, workflows, evidence, model behavior, authority, risk, and existing failure modes.

  2. Align

    Define the AI's role, human responsibility, trust requirements, authorization, and recovery.

  3. Implement

    Design workflows, interaction states, evidence presentation, controls, and prototypes.

  4. Iterate

    Evaluate behavior, corrections, failures, stale information, and changing user needs.

Proof.

Comparable work, with its provenance stated.

Proof · AI PRODUCT DESIGN · AI Product Design

Governed knowledge for AI.

Context

A governed evidence layer connected approved organizational sources to multiple AI workflows while preserving permissions, ownership, freshness, provenance, and human approval boundaries. A limited pilot validated the core model.

Provenance

UX Design Lab client engagement.Read the case: Governed knowledge for AI. →
Architecture diagram: approved company sources feed a governed knowledge layer (ownership, permissions, freshness, provenance) exposed through MCP to multiple AI workflows, with a human approval boundary.
FIG. 03.1Governed knowledge layer between approved sources and AI experiences: ownership, permissions, provenance, MCP access.

Business pattern

AI becomes a governance problem when answers move faster than evidence, ownership, and accountability.

Pavel Bukengolts smiling and gesturing by a window overlooking a city skyline

Senior work stays senior.

You work directly with Pavel Bukengolts, UXDL founder and principal, on assessment, major decisions, and direction. Specialist collaborators join when the work requires additional expertise or capacity.

Recognition

  • ADPList 100

    Recognized among the top mentors on ADPList.

  • 2026 GNTC Tech Educator of the Year

    Finalist, Greater Nashville Technology Council.

  • UX Magazine contributor

    Published on product, accessibility and AI practice.

Define the authority before the product gives it away.

Bring the product, workflow, delivery problem, or decision. No polished brief required.

Prefer email? [email protected]

Project inquiry

All fields are required except timeline.