What they did
Gave Claude an API’s swagger file to write code that pulled public data, cleaned it, queried a second API, combined the sources, extracted data from PDFs with tesseract, and built a dashboard for their users. Ran call recordings through local Whisper and had Claude extract action points, and turned an expert interview into an article outline with key quotes. Also had Claude write scrapers, and searched podcast transcripts instead of listening.
How it went
All of it ran alongside calls and meetings, with them just checking progress and writing new prompts. They estimate they got about three times as much done as they would have a year earlier, and finished work at 3pm.