Why Fleet Offers Both General-Purpose Chat and Specialized Agents

AI agent tasks broadly fall into two patterns: ad hoc tasks that grasp context and provide immediate answers, and standardized tasks that are performed repeatedly using the same instructions and tools.

Fleet has adopted a dual structure to meet these two requirements.

• General Purpose Chat: For low-cost tasks that require no setup. It uses various workspace integration tools such as Slack, GitHub, and Linear, along with a virtual computer, to grasp complex context and carry out one-off tasks.
• Specialized Agents: Provides a stable environment for recurring work. For tasks that don't need to be explained every time—such as weekly plan updates, email management, and customer research—these agents have fixed instructions and tools.

Specialized Agents maintain work consistency through the following detailed configurations.

• Instructions & Tools: Defines the role, decision rules, and output format, and assigns only the specific tools needed for the task to increase focus.
• Models & Subagents: Different models can be applied to the main agent and subagents. This optimizes cost and performance, and allows subagents to manage separate conversation histories, preventing context contamination of the main agent.
• Skills & Triggers: Company-wide common Skills can be shared, or agents can be set to run automatically based on specific events (Slack, Gmail, etc.) or a schedule.