Agents and tools

**Agents vs. Workflows**

• Workflows: Best used when the precise series of steps required to complete a task is known.
• Agents: Best used when the exact steps required to complete a task are unknown or unclear.
• Agent Function: An agent is given a task and a set of tools, and it relies on its internal reasoning to create a plan to complete the task using those tools.
• Agent Advantage: Flexibility; agents can solve a wide variety of tasks by combining tools in unexpected ways.

Agent Tool Usage and Flexibility

• Tool Combination: A key strength of agents is their ability to combine simple, disparate tools to achieve complex goals.
• Dynamic Planning: Agents can perform multi-step reasoning, such as:
• Using a tool to find current information (e. g. , Get Current Date/Time).
• Modifying that information (e. g. , Add Duration to Date/Time).
• Executing a final action (e. g. , Set Reminder).
• Information Gathering: Agents can recognize when they lack necessary information and proactively ask the user for clarification before proceeding with a tool call.

Designing Effective Agents (The Abstract Tool Principle)

• Core Principle: The set of tools provided to an agent must be reasonably abstract.
• Definition of Abstract Tools: Tools should be generic, general, or vague in purpose; they should not be hyper-specialized.
• Goal of Abstraction: The agent must be able to figure out how to piece together these generic tools to achieve the desired goal.
• Example (Claude Code): Instead of providing a specific "refactor tool," the agent is given abstract tools like "Bash" and must use its reasoning to determine the necessary steps to perform a refactor.
• Practical Application: When building an agent (e. g. , a video creation agent), provide general tools (Bash, Generate Image, Text-to-Speech, Post Media) that allow the agent to create dynamic workflows, such as generating a sample cover image and awaiting user approval before proceeding with the main task.

Takeaways

• Agents are best used for tasks where the exact steps are unknown, unlike workflows which require a precise, known sequence.
• Agents utilize internal reasoning to dynamically combine simple, disparate tools to achieve complex, multi-step goals.
• A core principle of effective agent design is providing abstract (generic) tools, forcing the agent to determine the necessary workflow.
• Agents can proactively identify missing information and request clarification from the user before proceeding with a tool call.