Agent Operational Models & Trade-offs

Agent versions and their explicit trade-offs in cost, speed, and testing reliability.

The Agent's operational model defines its performance profile—the fundamental trade-offs it makes between execution speed, accuracy, and resource cost. Configuring the operational model is the first step in optimizing your Agent for its specific business purpose, whether that is achieving lightning-fast responses or maximum accuracy for regulatory compliance.

The two critical configuration settings that determine the Agent’s operational model are Agent Mode and Reasoning Strategy.

Agent Mode (Speed vs. Accuracy vs. Cost)

Agent Mode dictates the quality of the underlying Large Language Model (LLM) and the associated compute resources used per step. This directly impacts both cost and the integrity of the Agent’s output.

Agent Mode	Description	Trade-offs	Ideal Use Case
Advanced	Utilizes premium reasoning and the most accurate LLMs.	Maximum accuracy; Higher cost and slower latency.	Auditable workflows, complex financial analysis, or client-facing applications where factual integrity is non-negotiable.
Balanced	Provides a blend of performance and cost efficiency.	Cost-effective and delivers steady, reliable accuracy.	General automation, back-office tasks, or internal tools where speed is important but errors are manageable.
Flash	Optimized for minimum latency and instantaneous response.	Lightning speed; Lowest cost per execution, but accuracy may vary.	Conversational interfaces, rapid data retrieval (using Flash Knowledge), or real-time voice agents.

Reasoning Strategy (Reliability vs. Speed)

The Reasoning Strategy dictates the Agent's planning depth, which directly correlates to the reliability of its execution plan (analogous to internal testing).

Strategy	Planning Depth / Execution Model	Trade-offs	Impact on Reliability
Intelligent Planner	The Agent creates a complete, multi-step plan before executing any action.	Higher latency at the start; more resource-intensive.	Provides high reliability; ensures the Agent checks prerequisites and anticipates errors before starting.
Next Best Step	The Agent determines only the immediate next action based on the current state.	Faster latency per action; lower resource usage per step.	Lower reliability for complex chains; susceptible to runtime errors if conditions change mid-workflow unexpectedly.

Autonomous Level (Human Interaction vs. Oversight)

The Autonomous Level defines how often the Agent will pause its execution and ask for human confirmation or guidance. This setting is crucial for workflows involving sensitive decisions or high-risk financial transactions.

Level	Description
High	Requires minimal human intervention; the Agent proceeds with its plan autonomously.
Medium	Requests intervention at key decision points or when facing an ambiguous situation.
Low	Requires frequent supervision and asks for permission before executing most high-value actions (recommended during initial testing).

Selecting the Right Model

Always align the Agent's operational model with the application's risk profile.

Use Advanced Mode with an Intelligent Planner for underwriting or compliance workflows.

Use Flash Mode with a Next Best Step strategy for high-volume, low-stakes customer support tasks.