Self-initiated case study · July 2026

Designing human accountability into agentic work.

A concept exploring how governed AI agents could support regulated enterprise work through visible plans, traceable evidence, safe failure and human approval.

Challenge

Make powerful agentic work useful without hiding decisions or removing human accountability.

Role

Concept lead, product designer, systems designer and hands-on prototyper.

Prototype

Six responsive routes, eight deterministic scenarios and twenty governed AI patterns.

Simulation

Local event transitions, fictional enterprise data and no live LLM or production integration.

Linda’s AI design framework

Five principles for governed enterprise agents

The framework turns responsible-AI intentions into observable product behaviour.

01

Agency must be visible

People should be able to see, edit, pause and stop an agent’s plan. Required controls remain visibly protected.

02

Consequential actions require human control

Sending, exporting and changing regulated wording always require an explicit, qualified human decision.

03

Claims must be traceable to evidence

Every material sentence connects to an approved source, owner, version and region using qualitative evidence strength.

04

Failure must preserve progress

A conflict, tool outage or privacy boundary stops only unsafe work and keeps completed steps available.

05

AI patterns must be measurable and revisable

Evaluation data reveals where patterns should be revised, tested or retired rather than treated as finished UI.

Three-minute walkthrough

How to present the concept

“How do we allow an AI agent to complete useful work without obscuring what it is doing or removing human accountability?”
  1. 01
    Task and agent plan

    Show that the plan is editable and stoppable while mandatory compliance checks remain protected.

  2. 02
    Draft and evidence mapping

    Select a claim and connect it to the exact source, owner, version and region.

  3. 03
    Conflict or privacy boundary

    Demonstrate that the agent stops affected work, preserves progress and escalates the decision.

  4. 04
    Pattern Library

    Explain that these behaviours are reusable patterns with content, accessibility and implementation rules.

  5. 05
    Evaluations

    End with how patterns are measured, revised or retired using real performance signals.

What this demonstrates

Strategy made tangible through an interactive system.

The prototype demonstrates product thinking, responsible-AI literacy, system design, accessibility, hands-on execution, technical awareness and executive storytelling. It does not claim to be a validated clinical system or a deployed global AI programme.