Building with the Claude API
Start from client.messages.create(). Leave with working Python for tool use, hybrid RAG, MCP servers, and agent loops. The parts the docs don't cover.
Taught by Edwin Ashdown. Founder of AIFirstPartner, 12 production AI systems shipped.
Software engineers comfortable with Python and HTTP who have not yet shipped a feature on Claude. No prior experience with LLMs or Claude required.
ML researchers who want the math, product managers who want the strategy, or people looking for ChatGPT tutorials. We skip every line of that.
Edwin Ashdown founded AIFirstPartner and has shipped 12 production AI systems for clients in government, legal, and enterprise. He is fractional CTO at Speculo, a cyber risk software company, and manages cloud and data platform engineering for a large Australian government department supporting 20+ agencies across the portfolio. He publishes his build process at 150days.aifirstpartner.com.
This course is the playbook he uses when onboarding a team onto the Claude API.
What this course covers, who it's for, and why the free docs stop where the real work starts.
1 lessons
Haiku, Sonnet, Opus: which one, when, and what each one costs to run.
1 lessons
Thirteen lessons from your first request to streaming clean JSON out. Every concept lands in code you run.
13 lessons
Build an eval pipeline with model-based and code-based graders so you know a prompt works before users find out it doesn't.
8 lessons
Four techniques that measurably improve prompt quality, with the scores to prove each one.
7 lessons
Give Claude access to real-time data through custom tools, from schema design to multi-turn tool loops.
13 lessons
One retriever class. Two index types. No hallucinated "your document says..." moments.
7 lessons
Extended thinking, image and PDF support, citations, prompt caching, code execution: when each one is worth the cost.
9 lessons
Tools, resources, and prompts: the MCP primitives built by hand with FastMCP.
12 lessons
Effective development workflows with Claude Code, and extending it with your own MCP servers.
4 lessons
Four workflow patterns you'll actually ship, plus the decision rule for reaching for a full agent.
8 lessons
Prove you can build any of this from scratch.
1 lessons
Skills summary and concrete projects to build with what you just learned.
1 lessons
The docs tell you what every endpoint does. They don't tell you which three endpoints you'll actually use, what the failure modes look like in production, or how tool use, RAG, and MCP compose into a working feature. This course is the second half of that story.
The API surface this course teaches (Messages, tool use, streaming, caching, MCP) is stable and versioned. Model names shift, the patterns don't. Lifetime access includes updates when they matter.
Six working programs in your own repo: a multi-turn chat app, an eval pipeline, a reminder-tool assistant, a hybrid RAG retriever, an MCP server and client pair, and a workflow-vs-agent reference design.
No. The course starts from your first API call and builds up. You should be comfortable writing Python and calling REST APIs. No AI or ML background is required.
56 lessons across 13 modules. Most learners complete it in 6-8 hours of focused study. You can work through it at your own pace.
Yes. The entire course is free and open. Start with Module 1 to get a feel for the voice and pace, then keep going through all 13 modules.
Yes. You'll need an Anthropic API key for the hands-on exercises. The course walks you through obtaining one. Some modules also use VoyageAI for embeddings (free tier available).
No. Building with the Claude API is independent. This course is not an Anthropic product and has not been reviewed or endorsed by them.