Claude

Get comfortable using Claude as an assistant before writing a single line of code.

• Create an account at claude.ai and have a real conversation — ask it to explain something, draft something, or debug a small problem.
• Learn the basics of a good prompt: be specific about the task, give relevant context, and say what “done” looks like.
• Try the same question on a faster, cheaper model and a more capable one, and notice the difference in depth versus speed.
• Learn what a context window is — the amount of conversation and documents Claude can consider at once, measured in tokens, not characters.
• Upload a document or image — Claude can read PDFs directly and reason over screenshots or photos.

Move from chatting with Claude to calling it from your own code.

• Create an API key in the Anthropic Console and install the official SDK for your language — Python or TypeScript are the easiest starting points.
• Make your first request to the Messages API: a model name, a max_tokens limit, and a messages array is all it takes.
• Learn the difference between a system prompt (how Claude should behave) and user messages (the actual conversation).
• Stream responses so your app shows output as it's generated, instead of waiting for the full reply.
• Install Claude Code and use it on a real project — it's the fastest way to see an agentic coding workflow firsthand.
• Learn the basics of prompt caching: reusing a stable prefix, like a system prompt, across requests cuts both cost and latency.

This is where Claude stops just answering and starts acting.

• Define a custom tool — a function with a name, description, and JSON schema — and let Claude decide when to call it.
• Build a basic agentic loop: call the API, execute any tool Claude requests, send the result back, and repeat until the task is done.
• Use Anthropic's server-side tools — web search, web fetch, and code execution — so Claude can look things up or run code without you hosting anything.
• Turn on extended thinking for genuinely hard problems, and learn to tune the effort level as a cost-versus-quality dial.
• Try structured outputs to force a response into a JSON schema you define — useful for extraction and classification.
• Use vision for screenshots, charts, or scanned documents instead of plain text.

Production agents need more than a single tool call — they need architecture.

• Decide between a hand-rolled agent loop (full control) and a managed option where Anthropic runs the loop and hosts the sandbox, based on how much infrastructure you want to own.
• Connect Claude to real systems with MCP servers instead of writing a one-off integration for every tool.
• Design for long-running sessions: compaction and context editing keep a conversation usable well beyond a single context window.
• Coordinate multiple agents — a coordinator that delegates to specialized sub-agents for review, testing, or research.
• Get deliberate about caching strategy: stable content first, volatile content last, and know exactly what invalidates a cache prefix.
• Choose a deployment path that fits your constraints — the first-party API, or a cloud-hosted option — since they run the same models with different infrastructure trade-offs.
• Read the migration notes whenever a new model ships — defaults and behavior shift between versions, and treating an upgrade as “just changing a string” causes regressions.