0
Data-Driven Automation
Rule-based automation and predictive analytics. Systems execute instructions written by humans and surface insights humans interpret.
What the system does
- Executes scripted sequences across applications (RPA)
- Applies machine learning models to forecast or classify
- Produces dashboards and reports
How to recognise it
- Automations break when an interface changes
- Analytics exist but decisions are still made in meetings
- Every exception returns to a person
Where it stops
Automation is brittle because it encodes steps, not intent. Anything outside the script becomes manual work.
1
Conversational Copilots
Natural-language assistants embedded in enterprise workflows. They draft, summarise and answer; humans decide and execute.
What the system does
- Drafts documents, emails and code
- Summarises meetings, contracts and reports
- Answers questions about internal documents
How to recognise it
- Adoption is measured in seats and usage, not in process outcomes
- Value varies enormously between individuals
- No business metric has moved as a result
Where it stops
Productivity gains are individual, not organisational. The process runs exactly as it did before.
2
Cognitive Reasoners
Generative systems that draft decisions and explain their reasoning. Humans still execute every action.
What the system does
- Analyses a case and proposes a decision with justification
- Compares options against criteria
- Flags anomalies and inconsistencies
How to recognise it
- The AI produces a recommendation a person copies into another system
- Reasoning is visible but not connected to systems of record
- Quality is evaluated by reading, not by measuring
Where it stops
The last mile is still manual. The system thinks but does not act, so throughput is capped by human transcription.
3
Autonomous Agents
Systems that plan and execute multi-step processes through APIs and tools, inside the company's own systems, under defined policies.
What the system does
- Completes a business process end to end
- Writes to systems of record
- Escalates exceptions to a human by policy
- Produces an audit trail of every decision
How to recognise it
- A process has a state: running, awaiting approval, escalated
- Permissions and thresholds are configured, not improvised
- The process owner can answer why an action was taken
Where it stops
Each agent operates within its own domain. Work crossing business units still requires a human to coordinate.
4
Swarm Intelligence
Cooperating agents that self-orchestrate across domains, hand off work and reach consensus without human coordination.
What the system does
- Routes work between agents across functions
- Resolves conflicting objectives against shared policy
- Reallocates capacity dynamically
How to recognise it
- Cross-functional processes complete without a coordinator
- Governance operates at the level of the system, not the single agent
Where it stops
Governance complexity grows faster than capability. Very few production deployments exist today, and accountability models are still immature.
5
Autonomous Organisations
Enterprises where the operating structure is largely executed by AI, with virtual org charts and a synthetic workforce. Humans set objectives, policy and boundaries.
What the system does
- Runs entire functions without a human operating layer
- Adjusts strategy execution against outcomes
How to recognise it
- No verified production example exists today.
Where it stops
Level 5 is not achievable with current technology, and the legal and accountability frameworks it would require do not yet exist. It is a direction of travel, not a roadmap item.
