System capacity
The functions the system can perform and the intermediate choices it can make.
- Goal interpretation and task decomposition
- Planning and reasoning
- Memory and state retention
- Tool and action execution
- Learning or adaptation
- Orchestration and multi-agent coordination
Diagnostic questions
- What can the system interpret, plan, retain, coordinate and execute?
- How independently can it select intermediate steps?
Detailed diagnostic guide
Interpretation and task representation
- What information can the system process and combine?
- How are the task, current state and constraints represented?
- Can it identify missing information, conflict or uncertainty?
Planning and sequencing
- Can it construct alternatives, generate intermediate objectives and revise a plan after action?
- Can it test whether a step succeeded and recover when it did not?
State and memory
- What persists within a trajectory, and what crosses sessions, cases or users?
- Can retained information be traced, corrected and deleted?
Execution and coordination
- Can it select, order and correctly call tools, then interpret their responses?
- How are roles, task transfers, communications and disagreements handled across agents?
Change
- Can models, prompts, skills, memory mechanisms or agent structures change after deployment?
- How are resulting changes in practical capacity identified and documented?
Why it matters: indicates the tasks and choices that may be delegated, and how behaviour may develop during runtime.
Environmental opportunity
The resources, systems and people the configured arrangement can access or affect.
- Data and information access
- Tools and APIs
- Connected systems
- Read and write permissions
- Communication channels
- Infrastructure
- Other agents
- Organisational and supplier-controlled components
Diagnostic questions
- What can the system access, affect or change?
- Which systems, data, people or organisations can it reach?
- What permissions constrain that reach?
Detailed diagnostic guide
Information and connected systems
- Which records, databases, document stores and external services are accessible?
- Can the system encounter untrusted or externally controlled content?
Permissions and operational access
- Can it read, create, amend, transmit, approve, execute or delete?
- Are limits technically enforced or expressed only as behavioural instructions?
- Can permissions be expanded, reduced or revoked during operation?
Human and workflow opportunity
- Which actions require approval, and is an appropriate human realistically available?
- When does an output enter an official record, communication, decision or transaction?
Resources, dependencies and reach
- What time, compute, cost, transaction or volume limits constrain activity?
- Which suppliers, services and downstream processes extend or restrict its practical reach?
Change
- Who can change tools, connections, credentials, limits and approval pathways?
- Which changes require renewed assessment or authorisation?
Why it matters: determines whether activity remains informational or can produce consequences in administrative systems and external environments.
Activation and direction
The conditions that start, steer, prolong, interrupt and terminate activity.
- Initial triggers
- Assigned objectives
- Instructions and constraints
- Event- or schedule-based activation
- Continuation conditions
- Human checkpoints
- Escalation and termination conditions
Diagnostic questions
- What starts the system acting?
- What directs its activity?
- How long can it continue without further instruction?
- What causes it to stop, escalate or return control?
Detailed diagnostic guide
Activation
- Can activity begin through instruction, schedule, API call, incoming record, external event or another agent?
- Does each trajectory require a new human act, or can it operate under a standing trigger?
Mandate and objective structure
- Who defines the mandate, where is it expressed, and how much of the route remains open?
- How are generated subgoals checked against the initial mandate?
Selection and trade-offs
- Which priorities, thresholds and performance criteria direct the next action?
- How are conflicting objectives, constraints or affected interests resolved?
Feedback and continuation
- What feedback can revise the current plan, priorities or completion assessment?
- What causes retry, redirection or continued activity, and what limits its duration?
Escalation and termination
- What pauses, transfers or stops activity even when the objective remains unmet?
- What happens to pending actions and delegated subtasks after termination?
Why it matters: helps locate when action begins, how much latitude exists after activation, and where human intervention remains possible.