Spend leftover Claude and Codex quota on small pull requests
A founder built a bot that turns spare subscription tokens into pull requests for small tasks he never gets around to.
A web developer gave Claude Code a baseline site, ad accounts with a budget, and scraping tools, and lets it make marketing decisions every four hours.
HamadMalikKhan gave Claude Code a baseline website, access to Google, X, LinkedIn and Reddit Ads with a budget, Apify, and OpenRouter. The site is personalized per visitor on the client side using DeepSeek and many device, context and behavior signals.
So far, he says, it mostly does 'tomfoolery': spinning up lots of SEO pages and adding share buttons to try to go viral. He finds it fun to watch it decide every 4 hours.
An experiment, not a success story. Cap ad spend at the account level. Heavy visitor fingerprinting raises privacy and consent issues in many regions, so check the rules where your visitors are.
Our own starting point, written from the source. Adapt it to your setup.
You run growth for this site. Every run: read the last 24h of analytics and ad results, choose ONE change (copy, page, or ad tweak) with the best expected impact, make it, and append your reasoning and the expected result to DECISIONS.md. Never raise spend above the caps.A founder built a bot that turns spare subscription tokens into pull requests for small tasks he never gets around to.
A developer built nightmux to control Claude Code agents from Telegram and auto-resume past usage limits overnight.
A solo iOS developer who calls Claude Code 'basically my whole team' cut trial churn from 50% to 30% by only changing paywall timing.
A bootstrapped YC founder explains how his sandbox service runs personal Claude Code and OpenClaw agents for about $0.05-$0.10 per free user a month.
Five real AI agent use cases each morning, each with a recipe and a source. Plus a Sunday roundup. Free.