Getting coding agents to write good tests isn't just about better prompts—it's about creating the right environment and examples for them to learn from.

• Leverage Python's rich test ecosystem: Python excels for AI-generated tests because pytest patterns are abundant in training data, allowing specific prompts like "use pytest-httpx to mock the endpoints" that agents understand immediately.
• Combat test duplication proactively: The most common anti-pattern is duplicated setup code—push back with specific refactoring commands like "use pytest.mark.parametrize" or "extract common setup into a pytest fixture."
• Seed projects with quality examples: The best strategy is ensuring your project already has clean test patterns—agents automatically mimic existing code quality without extra prompting, similar to how human developers learn from team codebases.
• Use reference repositories as training material: Clone well-tested projects (like datasette/datasette-enrichments) and explicitly tell agents to "imitate the testing patterns it uses"—showing beats telling for AI code generation.