Module 9 of 13 in the AI Track, a free, self-directed curriculum for building AI that moves revenue.
The idea
This is the jump from a model that talks to a system that does work. Function calling lets the model reach your APIs: the CRM, enrichment, search. Agents add planning and multi-step loops. The judgment that matters is knowing when an agent is the right call and when a plain, boring pipeline is more reliable and a lot cheaper.
The unglamorous half is what makes it hold up in production. Orchestration, human checkpoints where they belong, retries, validation, and a verification pass before anything leaves the system. Plus keeping cost and latency in check when there are several steps in a row.
Why it matters
Real revenue systems are multi-step. Enrich, score, draft, verify, hand off. This is the module where you learn to build the thing you’d actually sell, and build it well enough to run on its own without babysitting. Most AI demos die here, at the gap between “it worked once” and “it runs unattended.”
Build this
A working three-step agent. Enrich a lead through an API, score it, draft outreach that’s been verified. Document it start to finish, including what you chose not to make agentic.
How to work through it
- Anthropic’s tool-use docs and Building Effective Agents
- The OpenAI function-calling guide
- DeepLearning.AI’s agent courses
- Build the three-step agent with a verification checkpoint