Hi PH ๐ Shridhar here, building with Claude Code, open to work
byโข
I built the production-grade system behind Logiance, a startup where I was an individual contributor, including the distributed-systems side. These days I build AI-first: Claude Code lets me go from idea to a working prototype fast, while I own the design decisions, reviews and testing.
I'm here to try new AI and dev tools, give makers honest technical feedback, and connect with founders who need help shipping.
Question for you: what's one thing AI coding tools still get wrong in your real projects?
5 views
Replies
Welcome, Shridhar. One thing, since you asked: the tools still get lifecycle wrong in a framework they otherwise know well. In my Flutter app the agent kept reading the locale or the theme from the context inside the object that is constructed in initState, which throws at runtime, compiles fine, and looks perfectly idiomatic. It happened often enough that the rule is now written in the project notes with the exact error message, and the agent reads it before every screen. The pattern behind it is the real answer: the tool knows the API and not the order in which the framework calls it, so anything that depends on timing, lifecycle, async gaps after an await, a trigger that drops events in its first minute after deploy, is where I read the diff twice. The second thing, smaller, is translations: given 43 locales it will quietly shorten a string in the languages it is less sure of, or add a line break nobody asked for, so the rule there is complete and natural, never abbreviated, and no line breaks unless I ask.