Building effective agents

1. Building block: The augmented LLM

   Agentic systems are built upon LLMs enhanced with retrieval, tools, and memory. These models can independently search, select tools, and retain information.

2. Workflow: Prompt chaining

   Prompt chaining decomposes a task into a sequence of steps, where each LLM call processes the output of the previous one

3. Workflow: Routing

   Routing classifies an input and directs it to a specialized followup task.

4. Workflow: Parallelization

   LLMs can work simultaneously on a task and have their outputs aggregated programmatically

   • Sectioning: Breaking a task into independent subtasks run in parallel.
   • Voting: Running the same task multiple times to get diverse outputs.

5. Workflow: Orchestrator-workers

   A central LLM dynamically breaks down tasks, delegates them to worker LLMs, and synthesizes their results.

6. Workflow: Evaluator-optimizer

   one LLM call generates a response while another provides evaluation and feedback in a loop.

Agents

LLM agents autonomously execute tasks, leveraging advanced capabilities like reasoning and tool use. They operate independently, gathering real-world data and seeking human input when necessary. Agents can be used for open-ended problems where it’s difficult or impossible to predict the required number of steps, and where you can’t hardcode a fixed path.

You should consider adding complexity only when it demonstrably improves outcomes.

Reference

https://www.anthropic.com/research/building-effective-agents