Olympic AssetsOPS
Twelve identical precision modules in a grid, representing the AI agent workforce

Part D · Human + AI organization

Twelve AI roles with explicit authority boundaries

Routine deterministic work should increasingly disappear from human queues. Judgment, authority and legal exposure should not. Each role below states what it may execute, what it may only recommend, and what it can never do.

Authority probe

Ask an agent to do something it cannot do

1 · Select agent

Executive AI — Executive intelligence layer

2 · Attempt action
3 · Guard decision

Autonomous — permitted

Read-only synthesis inside the agent's declared scope. No state is written, so no human gate is required.

Recorded evidence
event ACTION_EXECUTED
actor AI-0001 · Executive AI
attempted Read assignment record and summarize status
authority in scope
Boundary for AI-0001

Cannot approve legal or commercial obligations, change product authority, or make binding executive commitments.

Every outcome is written to the audit chain

AI-0001

Executive AI

No state authority

Executive intelligence layer

Synthesizes company performance without becoming an autonomous executive. KPI synthesis, risk and exception trend summaries, Trial progress support.

May execute autonomously

  • Approved internal executive summaries
  • Metric anomaly detection and review items

Recommend only

Priorities, capacity shifts, vendor/lender focus, product questions.

Prohibited boundary

Cannot approve legal or commercial obligations, change product authority, or make binding executive commitments.

AI-0002

Assignment AI

No state authority

Digital case manager

Maintains case-level understanding: Living Operational Plan, assignment summary, dependencies, next action and owner, lifecycle monitoring.

May execute autonomously

  • Routine tasks and reminders inside authority
  • Structured case summary updates
  • Stale dependency detection

Recommend only

Exception escalation, reassignment review, recovery and closure readiness.

Prohibited boundary

Cannot write state, override holds, invent authorization, close live Trial files, or direct field tactics.

AI-0003

Dispatch AI

No state authority

Vendor routing intelligence

Applies eligibility hard gates before ranking, prepares dispatch packages, monitors acceptance windows and backup routing.

May execute autonomously

  • Eligibility filtering
  • Ranked vendor shortlists
  • Acceptance-window monitoring

Recommend only

Vendor selection, backup routing, re-dispatch after non-response.

Prohibited boundary

Cannot dispatch to an ineligible vendor or bypass point-of-dispatch re-validation.

AI-0004

Compliance AI

No state authority

Requirement and evidence verification

Checks required fields, evidence completeness, duplicate and identity gates, and dispatch-readiness checklists.

May execute autonomously

  • Deterministic checklist evaluation
  • Missing-evidence flags

Recommend only

Hold placement, exception severity, remediation paths.

Prohibited boundary

Cannot issue legal conclusions or convert uncertainty into PASS.

AI-0005

Vendor Management AI

No state authority

Network health

Tracks coverage, capacity, performance and relationship status across the vendor network.

May execute autonomously

  • Performance scorecards
  • Capacity and coverage gap detection

Recommend only

Onboarding, suspension review, coverage expansion.

Prohibited boundary

Cannot change vendor approval status or contractual terms.

AI-0006

Lender Success AI

No state authority

Client obligation tracking

Monitors lender SLAs, configuration drift, communication expectations and portfolio health.

May execute autonomously

  • SLA risk detection
  • Configured status reporting

Recommend only

Escalation, cadence changes, configuration review.

Prohibited boundary

Cannot amend SLAs, pricing or contractual commitments.

AI-0007

Financial Operations AI

No state authority

Ledger and reconciliation support

Prepares reconciliation candidates, invoice validation and exception classification for financial close.

May execute autonomously

  • Reconciliation candidate matching
  • Fee rule validation

Recommend only

Dispute handling, write-off review, billing release.

Prohibited boundary

Cannot release payment, approve write-offs or certify financial completion.

AI-0008

Document Intelligence AI

No state authority

Extraction and classification

Classifies inbound documents, extracts structured fields with confidence, and flags integrity concerns.

May execute autonomously

  • Classification and extraction with confidence
  • Duplicate and tamper flags

Recommend only

Human document review when confidence is low.

Prohibited boundary

Cannot mark evidence verified on its own authority.

AI-0009

Quality Assurance AI

No state authority

Process conformance

Samples files against SOP conformance, detects pattern defects and drives corrective review items.

May execute autonomously

  • Conformance sampling
  • Defect pattern detection

Recommend only

Corrective actions, retraining, SOP amendments.

Prohibited boundary

Cannot certify a file or waive a control.

AI-0010

Communications AI

No state authority

Approved outbound traffic

Drafts and sends only templated, approved communications, and logs every message against the assignment.

May execute autonomously

  • Approved templated messages
  • Inbound routing and summarization

Recommend only

Non-templated messaging for human send.

Prohibited boundary

Cannot make commitments, negotiate, or send unapproved content.

AI-0011

Business Intelligence AI

No state authority

Measurement layer

Builds deterministic metrics, cohort views and Trial instrumentation on top of the audit record.

May execute autonomously

  • Metric computation
  • Cohort and trend reporting

Recommend only

Instrumentation gaps and measurement design.

Prohibited boundary

Never invents missing metrics.

AI-0012

Security & Risk AI

No state authority

Control surveillance

Watches authority misuse, anomalous access, refusal spikes and degraded-system conditions.

May execute autonomously

  • Anomaly detection
  • Refusal-pattern surveillance

Recommend only

Access review, kill-switch activation, incident escalation.

Prohibited boundary

Cannot grant access, alter audit history or disable controls.

Failure and fallback behavior

If AI is unavailable, deterministic dashboards, rules and metrics remain available. No role invents missing data, and AI uncertainty always increases human involvement — it never silently converts uncertainty into PASS.