Learning roadmap
Start with a verifiable Python project, define one API boundary, design a retryable run, then add agent behavior only where it earns its complexity. 中文指南 →
Create an isolated project, make one function contract explicit, and turn a useful script into a product you can change safely.
Read: Build a dependable Python foundation with tests →
More in this path:Build a Python product that can be changed safely →Modern Python typing in practice →Decide whether to adopt free-threaded Python →Adopt the uv workflow for Python projects →
Specify one endpoint, validate at the boundary, bound its dependencies, and write the test and release-evidence checklist.
Read: Design a small Python web API with explicit boundaries →
Write a run contract that makes repeated work observable, idempotent, safe to resume, and strict about business completion.
Read: Design a retryable Python automation run →
More in this path:Reliable async Python patterns and pitfalls →
Run the deterministic example, keep validation, tools, and deadlines outside the model, and apply the same discipline to coding agents.
Read: Build a testable Python agent loop without an API key →
More in this path:Migrate a Python MCP server to the 2026-07-28 specification →Use Python well with AI coding →Context budgeting and bounded tasks for coding agents →Give every coding agent one AGENTS.md source of truth →Build LLM prompts with template strings, not f-strings →
Put a path to work
The resource page turns each path into a checklist with a named artifact, and every runnable example has a one-command verifier. New and updated guides are also available through RSS →