So other than gardening and fence painting (it seems to take forever!), I’ve been doing a bit of the Anthropic Academy training on using Claude, whilst continuing to play with Claude Code in the background. What the training has helped me with is understanding how to use Claude more efficiently, which has been interesting – although I’ve gone down a few rabbit-holes too.
What I’ve done today is add some new skills to my Claude Code project that allows me to run testing on my web application – looking for bugs, testing accessibility and auditing security. I’ve given Claude some basic testing for each of these, and it suggested some other use cases too, and I can request it runs the tests and it goes away and does it.
I can absolutely see the use of AI for testing – how often have you deployed code to find an issue, perhaps that didn’t appear on test or that you didn’t test adequately (some web applications have huge numbers of use cases, and a human being can’t test them all!). I set up Claude to use Playwright for my bug testing, connecting them together. The testing can happen behind the scenes or you can run it in a browser so you can watch the testing, which is fun.
The accessibility testing that Claude did for me uses axe-core (an open source accessibility testing engine) and checks the web application using rules for different versions of WCAG.
“With axe-core, you can find on average 57% of WCAG issues automatically. Additionally, axe-core will return elements as “incomplete” where axe-core could not be certain, and manual review is needed.” Axe-Core Git Hub read me
Once that has run, Claude also runs some testing with Playwright to check some things that axe-core doesn’t cover (ie ensuring dark mode works, checking keyboard navigation etc). This testing won’t catch everything but it’s a start and can be done quickly (which isn’t the case for manual testing – my experience of that is that it takes a-g-e-s to do it and so there’s a limit to how regularly you can do it).
My next job is to make sure the test cases cover all the different scenarios I want to test for each test.