How Did Airbnb Migrate 3,500 Tests in Six Weeks Instead of 18 Months?

Airbnb’s engineering team just pulled off a migration feat that would normally take 1.5 years—and compressed it into six weeks. How is that possible? The answer lies in LLMs, prompt engineering, and automation.

Why Move from Enzyme to React Testing Library (RTL)?

Modernizing test frameworks is essential for long-term maintainability and developer productivity. Airbnb decided to retire the older Enzyme suite in favor of React Testing Library, which offers better alignment with how components actually behave in the browser. But migrating thousands of tests by hand? That’s a recipe for burnout.

What Role Did AI Play?

Airbnb engineers fed an LLM with up to 100,000 tokens of code context—covering test setup, mocks, and assertions—to generate first-draft conversions from Enzyme to RTL. These “prompt chains” allowed the model to understand complex test logic and produce accurate, executable code. Engineers then reviewed and refined the outputs, ensuring quality without rewriting each test manually.

How Did They Maintain Trust and Quality?

Even with 80–90% of code generated by AI, Airbnb kept a human-in-the-loop model:

• Automated sanity checks for syntax and imports
• Peer reviews of AI-converted tests
• Custom QA scripts to catch edge cases

This hybrid workflow meant that engineers could trust the AI to handle bulk transformations, while still exerting final control over critical logic.

Are Other Tech Leaders Following Suit?

Yes. Google reported a 50% reduction in migration time for Google Ads tests, with roughly 80% of the new code authored by AI. To bolster confidence, Google built an LLM-powered QA layer that flags inconsistencies and potential bugs before merge.

Meanwhile, Amazon Web Services (AWS) surveyed developers and found that most treat AI tools like colleagues—they welcome AI assistance, but retain oversight, preferring transparent workflows over “black-box” automation.

What’s the Bigger Takeaway?

Airbnb’s success illustrates a broader shift in software engineering:

• AI accelerates repetitive tasks
• Engineers focus on architecture, edge cases, and integration
• Trust, transparency, and control are non-negotiable

In this new paradigm, AI doesn’t replace developers—it amplifies their impact, turning multi-month projects into multi-week sprints.

What Comes Next?

Expect more teams to adopt LLM-driven migrations, building tailored prompt libraries and QA pipelines. As these practices mature, AI will handle the grunt work—letting engineers innovate faster on features that matter most.

Source: Airbnb