Airbnb migrated 3,500 test files with LLMs in 6 weeks instead of 1.5 years
By Airbnb (Charles Covey-Brandt)Built and shared by Airbnb (Charles Covey-Brandt), not by me. Summarised from the public sources below. Check them before you rely on a number.
- Who
- Airbnb (Charles Covey-Brandt)
- Category
- Software development
- Source
- Web
- Dates
- Published Mar 2025 · Found 17 Oct 2026
Airbnb used LLMs to move nearly 3,500 React component test files from Enzyme to React Testing Library. The work had been estimated at 1.5 years of manual engineering time and was finished in six weeks.
What they built
An automated migration pipeline that treated every file as a state machine with validation steps (Jest, ESLint, TypeScript). Failing files went through retry loops with dynamic prompts built from the actual errors. Prompts were filled with up to about 100,000 tokens of context: the component's source, sibling tests, imports and good RTL examples. Engineers then improved the prompts step by step from samples of the files that failed.
The result
75% of the files were migrated in four hours and 97% after four days. The whole project took 6 weeks against an estimated 1.5 years by hand. The last 3% were fixed manually, and test intent and code coverage were preserved (self-reported).
Sources
OpenAI team shipped a product with zero lines of hand-written code
By Ryan Lopopolo (OpenAI)
A small OpenAI team spent five months building and shipping an internal beta of a software product where Codex wrote every line of code. Engineer Ryan Lopopolo describes how the engineers' job became designing the environment, the rules and the feedback loops the agents work inside, which he calls harness engineering.
Stripe's 'Minions' agents write more than 1,000 merged pull requests a week
By Stripe (Alistair Gray)
Stripe built its own coding agents, called Minions, that take a task from a Slack thread and deliver a pull request that passes CI and is ready for review. Each week more than a thousand merged PRs are written entirely by Minions; engineers review them but write none of the code.
Non-technical Meta PM builds and ships apps without writing code himself
By Zevi Arnovitz (Meta)
On Lenny's Podcast, Meta product manager Zevi Arnovitz shows his AI workflow, which lets a non-technical person build real products. He builds a new feature in his study app StudyMate live in the episode.
Want this in
your business?
Tell me which case caught your eye. I will tell you straight whether it fits, and what it would take.
Start a conversation →