Workflow vs Agents -
- Workflows are systems where LLMs and tools are orchestrated through predefined code paths.
- Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.
Tradeoffs -
- Agentic systems often trade latency and cost for better task performance, and you should consider when this tradeoff makes sense.
- Workflows - Predictability and consistency v/s Agents - Flexibility and model-driven decision making is required.
Building Block is an LLM. You then inject workflows, add features etc to it.
MCP gives access to the llm for many other tool calls. So part 1 is having an llm with the necessary tools to do this., which has a read/write memory and retrieval. Normal LLM’s like Claude supports them and additional tools can be injected through access via a MCP server.
Workflows
- Prompt Chaining -
Its just basic chaining and failsafes, intuitive.
When to use - When task can be decomposed into subtasks clear and isolated.
Issue - trading off latency for higher accuracy.