Hey, I'm Deepak. Software engineer in India, building web apps by day and after hours I build my own products with AI. I constantly keep writing it all up: what works, what flops, and what I learn trying to build stuff. Follow along.
I wrote a plain-English guide on what to do when AI wrote most of your app and you can't read the code when it breaks. Four small habits and a pre-ship checklist you can copy. https://t.co/LspZsthy3U
Austin is one of the few tracks where Turn 1 is a steep uphill climb right after the start. Braking while going uphill into a hairpin with 20 cars side by side is why the order can change before anyone reaches the top.
A side project rarely stalls because the code is bad. It stalls because version one keeps getting bigger.
Pick the one thing it has to do, ship that, and let the rest wait until people ask for it.
Read the whole error message before you paste it into AI.
Half the time the file name and line number already tell you where to look, and you fix it faster than the AI could reply.
@arvidkahl Cheap tests have the same problem as cheap code. If nobody reads them, a passing test written by the same agent that wrote the code doesn't prove much.
@HaStrive Copying the same skill into three folders is how two of them end up out of date without anyone noticing. One source of truth beats three copies.
Give your agent a clear "done" check before it starts.
If it can't tell when the task is finished, it either stops too early or keeps going and breaks something else.
A test that has to pass, a page that has to load, a file that has to exist. Small thing, big difference.
@Replit@stripe@atlas Going from building the app to setting up the company in the same chat is a big jump. The paperwork is usually the step people put off the longest.
Building an app with AI is the easy part now. Getting it ready for someone to pay for is where most apps get stuck. I wrote down the 5 things almost every vibe-coded app is missing, in plain English. https://t.co/dqGoPGlMz8