hi X
i'm pragun, 13, from india. i build AI that calls businesses' leads (leadflow) and an AI system that edits my instagram reels
gonna post what i build, what breaks, and the tech news i actually care about
experiment idea: same reel, edited by my instagram OS twice, once with opus 5.5 and once with sonnet 5.5
same prompt, same footage. i post both edits side by side and show the real time each one took
which one do you think wins?
question for people building with AI agents
what's the one thing you would never let an agent touch without a sandbox?
mine is api keys. curious what yours is
what are you building this week?
drop it in one line below. i'll reply to as many as i can with the first thing i'd fix or try
13yo, one macbook, 350+ cold calls deep. i've seen a lot of ideas get ignored
anthropic had claude prove 239 entry points across 123 open source libraries can't crash, for any input up to 64 MB
180 million lines of rocq proofs. it found hundreds of bugs fuzzers would almost never hit
writing code is easy now. proving it works is the new flex
350+ cold calls
most common reply: "baad me call karna"
nobody ever called back. not them, and honestly not most businesses with their own leads either
so i built an AI that calls every lead in minutes. it's called leadflow
duckdb 2.0 is out and the cli now has an "agent mode"
compact tables for input, errors as json. in their tpc-h test it cut the output the model has to read by 59%
dev tools are starting to be built for AI agents first, humans second. weird but i like it