Caffeine AI; deep frustration with "writing automated tests"

Is there any way to bypass caffeine having its QA bot writing automated tests? There are literally things it is unable to test by itself, such as simulating a human player opponent, or opening a secondary browser, but it insists on doing these automated tests, going round in circles, costing a lot of credits and many, many hours, only to produce nothing at the end

Previously, when its QA automated tests were attempted, and failed, it still delivered the code with “sorry, couldn’t do the test, but try it yourself”.

This latest update is making me wonder if this money burn is intended, or a bug, or (please, lord) something I’m doing wrong which is fixable

It could do these things by launching puppeteer or something similar. If it can’t but thinks it can, its a problem with the environment the Caffeine team has set up for the AI to run in.

I share your frustration regarding Caffeine. I haven’t touched it for almost a year. It’s not for building anything serious - that’s what you want codex on your own VM for, for complete power and flexibility.

With Opencode + Deepseek, it can do all the testing and report back what working, if bug, fixes itself.

I think Caffeine has limited premissions on whatever hardware its testing, limited access.

Why cant you just try other building options, at this point you just wasting money.

You’re probably right. I hope the transition from the caffeine playpen to the real world isn’t too jarring

I’m too lazy :weary_face:
Or was too lazy. It’s time to get better

I’m open for any tutorial anyone has to share :grin:

Here you go: https://hcgk6-wqaaa-aaaab-qhmxq-cai.icp.net/posts/2

I like that AI will do all itself, build, test, deploy. I only make canisters and make It co-controller.

Thanks. You’ve mentioned this a few times before. Taken me a while to come round. Cheers :clinking_glasses: