The AXD Playbook: 12-Week Agentic Readiness Programme

Conduct an AXD Readiness Assessment using the Four Pillars diagnostic (Trust Architecture, Delegation Design, Agent Observability, Human Re-engagement) to establish your organisational baseline.

Map every existing human-agent interaction in your organisation - identify where delegation is explicit (designed) versus implicit (defaulted), and where trust is earned versus assumed.

Establish the AXD vocabulary within your team - the 64 canonical terms provide a shared language that prevents the discipline from being absorbed into generic UX or AI product management.

Assign AXD roles: Trust Architect (owns trust architecture), Delegation Designer (owns scope and authority), Observability Lead (owns agent legibility), and Re-engagement Designer (owns human return paths).

Define your organisation's trust architecture principles - what trust means in your context, how it is earned, how it is measured, and how it is recovered after failure.

Implement the Intent Architecture Framework - design the translation layer between human intent and agent action, specifying how vague human desires become precise agent mandates.

Build your Delegation Design Framework - define the grammar of authority: what can be delegated, to what scope, under what conditions, with what revocation mechanisms, and for what duration.

Design your Trust Calibration system - implement the mechanisms by which trust levels are set initially, adjusted based on performance, and recalibrated after failures or context changes.

Establish your Operational Envelope - define the boundaries within which agents may act autonomously, the thresholds that trigger human re-engagement, and the circuit breakers that halt autonomous operation.

Create your Interrupt Pattern library - design the taxonomy of interruptions (informational, advisory, blocking, emergency) and the criteria that determine which pattern applies in each situation.

Implement Agent Observability - build the systems that make autonomous agent behaviour legible to humans who were not present when the agent acted, including structured logging, behaviour summaries, and decision traces.

Design the Autonomy Gradient for each agent interaction - map the spectrum from full human control through supervised autonomy to full agent autonomy, and define the trust thresholds that govern movement along the gradient.

Build your Trust Recovery protocols - design the specific sequences that restore trust after different categories of failure (minor deviation, significant error, trust violation, catastrophic failure).

Implement the Memory and Continuity Framework - design how agents maintain relationship context across sessions, how trust state persists, and how delegation history informs future interactions.

Deploy your Orchestration Visibility system - make the coordination between multiple agents and between agents and humans visible, traceable, and interruptible at every stage.

Implement outcome measurement - shift from measuring agent outputs (actions taken, tasks completed) to measuring outcomes (human satisfaction with delegated results, trust trajectory, relationship quality over time).

Establish your AXD governance framework - define who reviews trust architecture decisions, how delegation scope changes are approved, and how trust failures are investigated and remediated.

Conduct your first Trust Debt audit - identify where trust has been borrowed (assumed rather than earned), where trust recovery has been deferred, and where delegation scope has expanded without corresponding trust verification.

Train your organisation on AXD principles - move beyond the core team to embed trust architecture, delegation design, and human-agent interaction thinking across product, engineering, design, and leadership.

Establish continuous AXD improvement cycles - monthly trust architecture reviews, quarterly delegation scope audits, and annual AXD maturity assessments using the Four Pillars diagnostic.

Trust is established through a designed onboarding sequence, not assumed from first interaction.

Trust levels are tracked as explicit state - the system knows the current trust level for every human-agent relationship.

Trust calibration mechanisms exist - the human can adjust trust levels up or down based on observed agent performance.

Trust recovery protocols are designed for at least three failure severity levels (minor, significant, catastrophic).

Trust debt is monitored - the system tracks where trust has been borrowed and flags relationships where trust recovery has been deferred.

Every delegation has explicit scope - what the agent may do, what it must not do, and under what conditions it must pause.

Delegation duration is specified - every grant of authority has a defined expiry or review trigger.

Revocation is immediate and propagating - when a human revokes authority, every agent in the delegation chain stops within defined time bounds.

The delegation contract is comprehensible - the human can understand and evaluate what authority they are granting before they grant it.

Delegation audit trails are maintained - every grant, modification, and revocation of authority is logged with timestamp, scope, and rationale.

Agent actions are logged with reasoning - not just what the agent did, but what it considered, what it chose, and why.

Behaviour summaries are generated at multiple levels of abstraction - from high-level outcome summaries to detailed action traces.

The human can reconstruct what happened during their absence without replaying every individual agent action.

Anomaly detection flags unusual agent behaviour for human review before it compounds into larger failures.

Post-mortem tooling exists - when failures occur, the full context can be reconstructed for investigation and learning.

The human return path is designed - when the human re-enters the experience, context is restored and decisions since departure are summarised.

Interrupt patterns are categorised (informational, advisory, blocking, emergency) with clear criteria for each level.

The system adapts to the human's re-engagement patterns - frequent reviewers receive different summaries than infrequent ones.

Authority reconfirmation is triggered after significant context changes, extended absences, or trust-relevant events.

The experience degrades gracefully when the human is unreachable - the agent reduces autonomy rather than proceeding with stale authority.

Success is defined in terms of outcomes (human satisfaction, relationship quality) not outputs (tasks completed, actions taken).

The human specifies desired results, constraints, and conditions - not step-by-step instructions for the agent to follow.

Outcome measurement includes temporal dimensions - immediate results, medium-term relationship trajectory, and long-term trust accumulation.

Failure is defined in terms of outcome deviation, not just action errors - an agent that completes all actions but produces the wrong outcome has failed.

Outcome feedback loops exist - the system learns from outcome assessments to improve future delegation, calibration, and autonomous operation.

Trust Trajectory: The direction and rate of trust change over time - is trust accumulating, stable, or degrading? Measured as a time-series of trust state changes.

Trust Debt Ratio: The proportion of trust that has been assumed versus earned - high trust debt indicates fragile relationships that are vulnerable to single-point failures.

Recovery Effectiveness: How quickly and completely trust is restored after failures - measured as the percentage of pre-failure trust level recovered within defined time windows.

Calibration Accuracy: How well the system's trust level matches the human's actual confidence - measured through periodic trust surveys compared against system trust state.

Delegation Scope Utilisation: What percentage of delegated authority the agent actually exercises - low utilisation suggests over-delegation, high utilisation suggests the scope may be too narrow.

Scope Violation Rate: How often agents approach or exceed delegation boundaries - a leading indicator of trust architecture stress that precedes actual failures.

Revocation Frequency: How often humans revoke delegated authority - high revocation rates indicate delegation design problems or trust calibration failures.

Intent Translation Fidelity: How accurately agent actions reflect human intent - measured by comparing agent outcomes against human-specified desired results.

Absence Duration: How long humans remain absent between re-engagement events - increasing absence duration indicates growing trust, decreasing duration indicates trust erosion.

Interrupt Appropriateness: Whether agent-initiated interruptions were warranted - measured by human assessment of whether each interrupt was necessary and correctly categorised.

Re-engagement Satisfaction: How effectively the human return path restores context and enables informed decision-making - measured through post-re-engagement surveys.

Outcome Alignment: How closely agent-achieved outcomes match human-specified desired results - the ultimate measure of agentic experience quality.

Assumed Trust: Granting agents full autonomy from first interaction without a designed trust-building sequence. The agent has authority it has not earned, and the first failure destroys a relationship that was never properly established.

Missing Recovery: Designing the happy path without designing the failure path. When the agent fails (and it will), there is no protocol for acknowledgement, explanation, compensation, or rebuilding. Trust collapses permanently.

Static Trust: Treating trust as a one-time setting rather than a dynamic state. The system never adjusts trust levels based on performance, context changes, or the passage of time. Trust becomes stale and disconnected from reality.

Trust Debt Accumulation: Borrowing trust through defaults and assumptions without ever earning it through demonstrated performance. The relationship appears stable until a single failure reveals that the foundation was never built.

Unbounded Delegation: Granting authority without explicit scope, duration, or revocation mechanisms. The agent can do anything, for any duration, and the human has no designed way to stop it. This is not trust - it is abdication.

Incomprehensible Contracts: Delegation specifications that are technically precise but humanly incomprehensible. The human 'agrees' to a delegation they cannot evaluate, making informed consent impossible.

Irrevocable Authority: Delegation designs where revocation is technically possible but practically difficult - buried in settings, delayed in effect, or incomplete in scope. Revocation must be immediate and total.

Scope Creep: Gradual expansion of agent authority without corresponding trust verification. Each individual expansion seems small, but the cumulative effect is an agent operating far beyond the trust level the human has actually granted.

Invisible Autonomy: Agents that act without leaving legible traces. The human returns to find outcomes they cannot explain, decisions they cannot trace, and a system state they cannot understand. This is the opposite of designed absence.

Information Overload: Observability systems that log everything but summarise nothing. The human is presented with raw action logs instead of meaningful behaviour summaries, making re-engagement cognitively overwhelming.

Missing Return Path: No designed experience for the human's return. The human re-enters the system with no context restoration, no decision summary, and no mechanism to review or reverse agent actions taken during their absence.

Interrupt Fatigue: Agent-initiated interruptions that are too frequent, too urgent, or too trivial. The human learns to ignore interrupts, which means they also ignore the genuinely critical ones. Interrupt design requires the same rigour as trust architecture.