From AI Sceptic to AI Sceptic Who Ships
summary
A tiger team of five prototypes with used Claude and Whisper and ships two production systems: a domain scoring workflow that goes live behind customers and a usage-based billing ledger that sustains 10,000 events per second on Postgres. The scoring service lands as ~21,000 lines of code backed by BDD suites and docs; a refactor arrives as a 4,500-line PR that still passes tests. Teams battle hallucinations and imagined prompt injections, build a harness of plugins, hooks and hinters, and after months the services run with minimal handholding.
AI-written summary · about
chapters (13)
- 0:00From skeptic to AI shipper
- 3:11Paddle’s role in global transactions
- 8:26Tiger team’s blank check experiment
- 12:18Steel thread prototype
- 15:51Prototype to production pressure
- 19:42Success after three months
- 23:55Hallucinations and beguilement
- 28:04AI harness at Paddle
- 33:27Usage-based billing dilemma
- 40:28Refining after rapid engineering
- 43:14Teams in AI acceleration
- 46:45Avoiding reverse centaur syndrome
- 48:36Q&A
related talks
01h 00mHow to survive as a developer in the exponential age of AI
35mBeyond the Hype and Hyperbole
57mOne Team’s Journey to Using AI Without the Hype
40mA Developer’s Adventures in AI Wonderland
38mFrom scratch to production in three months: the AI-first delivery model
45mThe AI Revolution in Software Development: From Code Writing to Quality Assurance
32mFrom Three Days to One Minute: Architecture Trade-offs in Production Enterprise AI
38mOne Android Dev, Ten AI Teammates: Scaling Delivery With Claude Code