We over-trust AI code — more shipping, more breaking, and what I built about it

Transcript

So last year we started coding with AI and as usual a lot of things started happening. Number one people started making a lot more mistakes in their book request not by choice but it's more like they are like okay the code might be good. Too much over trust on AI. Now AI has improved a lot in last one year but still the trustability of it is something which makes people want to trust it because they just want to save time. It's much more easier to say the AI is good than to spend one more hour to test it out to test out all the things if everything is working or not and because of that what is happening is while companies are launching more and more things there are more things that are breaking than ever before. So what's the solution to it? Well we build one it's called BrowsingB basically can create test so it's an AI agent which can kind of browse the web your application just how a user will and it will use those features. So for example we have a feature to create presentation from documents. It will go to the website sign in login whatever we ask it to do it will sign in and then create the presentation from docs not by API call not by something else it's actually going to click one thing enter the input exactly how a user would and in the end only if we get a successful response then only it will say the test runs successful. So now what we do is whenever we have a new push changes to the website we run that and the second time we have it scheduled to run on a daily basis to test all the features we are running around 86 tests every single day in at 9 a.m that's just the automated test which we have set up for all the different features we have so that we can make sure all of them are working fine. We get a single summarized email of saying you know three tests failed out of 83 tests passed and inside that email you'll have a link to specific tasks of the tests so you can kind of see screenshots of step-by-step execution of how things got executed and if something has got break then you can see where at which step it's breaking it's much easier than having to hire a manual intern who can test these things out or someone even experienced because this ai will work at 1 a.m if you'll ask it but your intern won't this ai will work for just merely 100 dollar which is like 7 000 rupees per month anytime all the time but your ai intern won't so if you're interested to test it out visit browsingby.com or dmb somehow i'll let's arrange a call and let's get you set up with a better way of testing to create a much more reliable product than what you have right now