Over $2,000 below MSRP on a 2026 Volkswagen Golf R, negotiated entirely by an AI. That's the result a developer documented after pointing Claude at 20-plus California dealerships and letting it handle the outreach.
The setup was simple: Claude's desktop app with the Chrome extension in coworking mode, which gives the AI direct browser control. The developer created a dedicated email address, wrote a prompt saying he wanted $3,000-plus off MSRP and preferred email-only communication, then let Claude loose.
What Claude Actually Did
Claude navigated dealership websites on its own, identified general managers and sales managers by name rather than sending messages to generic contact forms, then drafted personalized emails to each one. Without being told to, it added urgency tactics like "first dealer to hit my number gets the business" and included social proof language. It handled follow-up emails as responses came in.
The developer considered using OpenClaw, an open-source browser automation tool, but found it too clunky compared to Claude's built-in browser control.
Volume Was the Real Advantage
The developer's own takeaway is honest: Claude didn't write better emails than a human could. The value was doing it 20 times simultaneously. Contacting that many dealerships, finding the right person at each one, and managing parallel negotiations is hours of tedious work that most buyers skip. They walk into one or two dealers and take whatever price they get.
This is a pattern worth paying attention to. The most practical AI agent use cases aren't the ones where the AI is smarter than you. They're the ones where it handles the boring, repetitive legwork that you'd never bother doing yourself - and that legwork turns out to be worth real money.
The blog post used Claude Opus 4.6 for the entire workflow. At current API pricing, the total cost of the Claude usage was almost certainly under $20, making the return on that investment roughly 100x.