i was going through some anthropic docs and came across this blog where they’ve described how to build effective agents.

it breaks down agentic systems into two broad categories:

1. workflows - systems where llms and tools are orchestrated through predefined code paths.
2. agents - systems where llms dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.

basic building block of agentic systems is an llm enhanced with augmentations such as retrieval, tools and memory.

different ways to build an agentic system:

1. workflow with prompt chaining: decompose a task into a sequence of steps and then an llm(prompt) call processes the output of the previous one with programmatic checks at each step.

2. workflow with routing: a classifier llm(prompt) sits at the front routing the query to the specialized llm(prompt)

3. workflow with parallelization: multiple llm(prompt) calls work on a pre divided task and then an aggregator llm(prompt) generates the final output

4. workflow with orchestrator and workers: kinda mix of routing + parallelization

5. workflow with evaluator and optimizer: one llm(prompt) generates an output while the other generates feedback in a loop

6. autonomous agents: llms(system prompt + user query) with a lot of tools

while building workflows humans remain in control of the process, but an agent just works autonomously using its own intelligence and data plus the tools you supply.