Advanced · 5h 49m · 22 lessons
Build with the Claude API
Go from a blank file to a working AI feature, one Claude API call at a time.
What you'll learn
- ✦Explain how the Claude API differs from a chat app, and what you gain by calling it directly
- ✦Construct a well-formed messages array with system, user, and assistant roles
- ✦Tune parameters like max_tokens and temperature to control a reply's length and randomness
- ✦Build a multi-turn conversation by resending the full message history on every call
- ✦Stream a response, call a tool, and request structured JSON output
- ✦Keep an API key secret, and handle errors, rate limits, and cost like a production app should
- ✦Use prompt caching and the Batch API, and add image and long-document input
- ✦Build a small agent loop, evaluate an AI feature's quality, and ship it with a launch checklist
Part of these career paths
- AI-Assisted Developer path · course 5 of 5
Certification projects
Build 4 projects to earn your certificate
Hands-on work that proves you can apply what you learned — part of the certificate requirements.
Syllabus
Module 1 · 1h 3m
Meet the Claude API Free
What you'll be able to do
- Can explain why the API has no built-in memory, unlike Claude.ai's chat window.
- Can pick a model for a task and store its API key safely.
- Can send a first Messages API request and read the model, text, and usage back from the response.
- Can install and call the official Python or TypeScript SDK instead of raw curl.
Module 2 · 1h 8m
Requests & responses Pro
What you'll be able to do
- Can build a valid messages array that alternates user and assistant turns correctly.
- Can write a system prompt that sets a persona, rules, and context once for the whole call.
- Can choose max_tokens and temperature to fit a task, from deterministic extraction to creative writing.
- Can read stop_reason, content blocks, and usage from a response to know what happened and what it cost.
Module 3 · 1h 16m
Building real features Pro
What you'll be able to do
- Can maintain a multi-turn conversation by resending the growing history on every call.
- Can turn on streaming so replies render word by word instead of all at once.
- Can define a tool schema and handle a tool_use request by running the real function.
- Can get reliable JSON back using prompt-based or forced tool-call structured output.
Module 4 · 1h 2m
Production concerns Pro
What you'll be able to do
- Can classify an HTTP error and decide whether to retry it with backoff or fix the request.
- Can plan around requests-per-minute and tokens-per-minute limits before hitting them.
- Can track input and output tokens to keep an API bill predictable.
- Can keep an API key server-side only, and knows what to do if one leaks.
Module 5 · 40m
Going further Pro
What you'll be able to do
- Can decide when a fixed prompt chunk is worth marking as cacheable.
- Can choose the Batch API over a real-time call for large, non-urgent jobs.
- Can send an image or a long document and explain why extra context still costs tokens.
Module 6 · 40m
Ship an AI feature Pro
What you'll be able to do
- Can build a bounded reason-act-observe agent loop around tool use.
- Can assemble an eval set to measure an AI feature's quality before users do.
- Can check a feature against a launch checklist before it goes live.
More AI courses
✦ Pro
AI for Developers
See how LLMs really work, then use AI in your daily dev work without trusting it blindly.
✦ ProChatGPT & Claude for Coding
Use ChatGPT and Claude to write, debug and understand code, and check every word they say.
✦ ProPrompt Engineering Basics
Learn to talk to AI models so they actually do what you want.