I've been using AI to write tests and it works surprisingly well! Here's how.

Swizec's platform provides resources for developers, covering a diverse range of topics such as programming languages, web development frameworks, data visualization techniques, and software engineering best practices. Through tutorials, guides, and coding challenges, Swizec offers hands-on learning experiences and practical insights for developers at all skill levels. Readers can explore  technologies, industry trends, and software development methodologies to enhance their technical proficiency and advance their careers in the tech industry. Additionally, Swizec shares personal anecdotes, career advice, and productivity tips to inspire and motivate aspiring developers on their journey of continuous learning and professional growth.

swizec.com

Using AI, specifically Cursor, to write tests can be surprisingly effective. It assists in generating tests based on code comments and intent, identifying bugs, and iterating input combinations. This automation helps focus on creating meaningful input scenarios rather than writing and managing tests manually, making the testing process more efficient and insightful.

AI writes good tests, actually

Learned something new? Read more Software Engineering Lessons from Production

Great post! What’s usually worked well for me is writing an initial test myself — it helps set the expectation and gives the AI a clear pattern to follow. Once that’s in place, the AI tends to generate much better tests that align with my style and intent.

I share the same experience. I wrote a custom SortedMap implementation for immutable sets, optimized for memory size, and had the AI (JetBrains Junie in my case) create the tests and it directly highlighted two bugs that I would probably overseen otherwise. One was that the keySet() method must return a set that iterates the keys in sorted order - I totally overlooked that as the method is declared to return a Set, not a SortedSet.

Great post!
Yes, but we need to define some standards. In complex apps, the agent will re-mock all your dependencies — but once that’s set up, it’s insanely powerful.
You really save a lot of time, and writing good tests starts with writing a good prompt.

These simple tests of 1+1 = 2 will be because the feature tests that we do here the AI ​​does not understand even half or know how to follow the pattern we have in the project even though it is asked explicitly. Added to rules etc. etc.

Terrific post! Thought process is spot-on.