Anthropic's building-effective-agents guidance separates workflows (predetermined code paths) from agents (the model directs its own process). Build all three workflow patterns against the same task set so the tradeoff is empirical rather than theoretical.
What to build
- Chaining: summarise a document, then extract entities from the summary, then classify the entities. Each step's output is the next step's input.
- Routing: classify an inbound support message, then dispatch to one of three specialised prompts. The classifier must be a separate, cheap call.
- Parallelisation: score one document against four independent rubrics concurrently, then aggregate.
- Add a validation gate between chained steps that halts the chain on malformed output.
- Time all three and record cost per run.
Done when
- Chaining fails fast at the gate rather than propagating garbage downstream.
- Routing uses a small model for classification and a stronger one only where needed.
- Parallelisation's wall-clock is near the slowest branch, not the sum.
- You can state, for each pattern, the property that makes it the right choice.
If you want to go further
- Add a fourth path where the model chooses the pattern, and note what you gave up in predictability.
The trap this exercise teaches
Reaching for an autonomous agent when a fixed chain would do. Workflows are more predictable and cheaper; agents earn their cost only when the steps cannot be known in advance.
Before you start
The theory behind this build is covered in Workflow patterns: chaining, routing, parallelisation. If any step below is unfamiliar, read that first — the exercise assumes it. Primary source: Anthropic — Building effective agents.
Check yourself against the exam
This exercise sits in Domain 1, which is 27% of the CCAR‑F exam. Once you have built it, run a domain drill in the study app and see whether the questions read differently.