05
Orchestrate
Sets the loop
Chat interfaces
Standalone AI assistants. No system connections.
Claude · ChatGPT · Gemini
AI with tool use enabled
Chat AI connected to actions via APIs. Human-initiated.
Claude MCP · n8n · Zapier AI
Agentic workflow platforms
Multi-step agents with human approval nodes.
Relay.app · Vertex AI Agent Builder · Agentforce
Autonomous agent platforms
Self-directed pipelines. Observability essential.
ServiceNow AI · Vertex AI Agents · AWS Bedrock Agents
Custom orchestration layer
Proprietary multi-agent mesh. Agents coordinate and self-direct.
Bespoke builds on hyperscaler foundations
Unstructured individual adoption
Self-taught, no shared practice or organisational norms.
Owner: Individual / Line manager
Centre of Excellence forms
AI CoE established. Champions embed shared habits across teams.
Owner: AI champion
Process owners + approvers
Clear accountability at every checkpoint. Structured training.
Owner: Business unit lead
AI Operations roles emerge
AIOps specialists monitor, intervene, and continuously improve agents.
Owner: COO / Transformation lead
Human-agent teaming as org design
Humans set goals and govern agent teams. Execution fully delegated.
Owner: CEO / Board
Shared awareness
Leadership names AI as strategic reality before rumour does. Clear, honest vision shared company-wide.
Owner: CEO / MD
Fear addressed directly
Managers have structured 1:1s on job security. Capability framing, not threat framing. Not a town hall.
Owner: Line managers / HR
Identity shift begins
Teams see themselves as AI-enabled, not AI-threatened. Early adopters become visible exemplars. Safety to experiment.
Owner: Team leads / People and Culture
Confidence and agency
AI use is proactive, not directed. Shared language and norms in place. Curiosity is rewarded. Fluency becomes identity.
Owner: COO
AI as cultural norm
No distinction between using AI and simply working. Embedded in onboarding and development. Leadership models it visibly.
Owner: Board / CEO
Personal habit change
Low organisational effort. Individual behaviour only.
Owner: Individual / Line manager
Team workflow adoption
New norms, shared tools, peer-led change.
Owner: Team lead / AI champion
Process redesign
Approval flows, governance, retraining. BU lead and change office.
Owner: Business unit lead
Accountability restructure
Redefine performance frameworks, KPIs, and ownership models.
Owner: COO / Transformation lead
Operating model transformation
Culture, org redesign, enterprise architecture alignment.
Owner: CEO / Board
Identity baseline
SSO enforced. Data classification in place. Access controls established.
Okta · JumpCloud · Ping Identity
API access controls
Scoped API keys, role-based permissions. No shared credentials.
Gravitee · Kong · MuleSoft API Manager
Action permissions + audit logging
Every agent action logged, reviewable, and escalatable.
Splunk · Elastic SIEM · Datadog
Autonomous failure response
Rollback capability, kill switches, anomaly detection.
Rubrik · CrowdStrike · PagerDuty
Agent identity management
Zero-trust. Each agent has a scoped identity — agents verify each other.
CyberArk · HashiCorp Vault
AI acceptable use policy
Approved tools defined. Usage guidelines published. Staff acknowledgement.
Owner: General Counsel / IT
Vendor governance in place
AI vendor DPAs signed. Model training, residency, and retention covered.
Vendor DPA · Termly · Model T&Cs reviewed
Per-use-case compliance mapping
Regulatory obligations mapped. Consent mechanisms in place. AI ethics review defined.
Privacy Act · Spam Act · APRA CPS 234 · CDR
Continuous compliance monitoring
Audit-ready documentation. PIAs completed for all high-risk use cases.
Ongoing obligation tracking · PIA register
Enterprise AI governance framework
Board-level AI governance. Model risk management. AI liability standard in all contracts.
Proactive regulatory engagement · AI ethics board
Human decides entirely
AI generates outputs only. Human interprets and acts. No decision logic.
Owner: Individual
Rule-based logic
Explicit if/then criteria defined by humans. AI applies fixed rules, no inference.
Owner: Process owner
AI recommends, human approves
AI produces recommendation with visible reasoning. Human retains final authority.
Owner: Business unit lead
Autonomous within guardrails
Confidence thresholds set. AI decides within bounds. Edge cases escalated.
Owner: COO / AI Ops
Multi-agent consensus
Agents negotiate and arbitrate outcomes. Authority hierarchy defined by humans.
Owner: Board / CEO
Session context only
You bring data to the AI in the prompt. No integration, no persistence.
Owner: Individual
Single source, read-only
AI reads from one live system via documented API. No write-back.
Snowflake · SAP · HubSpot
Multi-source reads
AI pulls from several systems per step. Consistent schemas required.
dbt · Fivetran · Airbyte
Read and write, governed
Real-time access. AI updates records. Data quality SLAs enforced.
Collibra · Alation · Hightouch · Databricks Unity Catalog
Unified semantic data layer
Agents query, update, and share context in real time. Live memory.
Databricks Lakehouse · Palantir · Pinecone / Weaviate
No integration
Standalone tools only. AI fully isolated from business systems.
Owner: Individual
Point-to-point API
Single system, stable REST endpoint, human-initiated calls.
Salesforce · ServiceNow · SAP via REST API
Managed iPaaS integration
Versioned APIs, human-gated cross-system actions.
MuleSoft Anypoint · Dell Boomi · Workato
Event-driven architecture
Async, webhook/queue-based. Full pipelines run without human initiation.
Kafka · Confluent · AWS EventBridge
Agent mesh communication
Real-time inter-agent APIs. Agents invoke and coordinate autonomously.
Emerging · AWS Multi-Agent · Google A2A protocol
