Airbnb AI Migration: 18-Month Code Move Completed in 6 Weeks

Caption:Airbnb logo with AI technology migrating swiftly.

The Highlights:

• Airbnb successfully completed a large-scale code migration using AI, updating 3,500 React component test files from Enzyme to React Testing Library in just 6 weeks.
• The company utilized LLMs (Large Language Models) to automate the migration process, breaking it down into validation and refactoring steps for efficient conversion of hundreds of files at once.
• Google and Amazon have also shared their experiences with LLM-driven code migrations, citing significant acceleration in the process and the need for human oversight to verify AI-generated changes due to inaccuracies or unnecessary modifications.
• AWS conducted research on human-AI partnership in code migrations, highlighting developers’ desire for control over the process and the importance of revealing limitations of AI systems to align expectations and ensure meticulous verification.

“We’d originally estimated this would take 1.5 years of engineering time to do by hand, but — using a combination of frontier models and robust automation — we finished the entire migration in just 6 weeks,” said Charles Covey-Brandt, a software engineer at Airbnb

Trending :Airbnb AI Migration ,AI workweek ,AI voice assistant ,Windows 11 preview

Airbnb AI Migration: How the 18-Month Code Move Was Completed in Just 6 Weeks

Airbnb recently shared insights on how they are increasingly using AI to manage and migrate codebases. The company successfully completed a large-scale code migration by leveraging LLM technology, updating 3,500 React component test files from Enzyme to React Testing Library in just 6 weeks. This process was estimated to take 1.5 years if done manually.

Charles Covey-Brandt, a software engineer at Airbnb, highlighted that the decision to move away from Enzyme was due to its outdated practices not aligning with modern React testing standards. By mid-2023, Airbnb had validated the use of LLMs for converting enzyme files to RTL within days and developed a scalable pipeline for an “LLM-driven migration.”

The company’s approach involved automated validation and refactoring steps, where files moved through stages of validation with the help of LLMs when checks failed. Through experimentation and iterative improvements, Airbnb managed to migrate 75% of target files in just four hours.

Despite initial challenges with some files failing validation criteria, Airbnb built tools for reruns which significantly improved their success rate. The company’s experience is not unique as other tech giants like Google and Amazon have also shared similar successes with AI-driven code migrations.

Google detailed its experiences using LLMs for accelerating code migrations by up to 50%, citing an example where manual efforts would have required hundreds of engineering years but were streamlined through AI assistance.

Similarly, AWS conducted research on human-AI partnerships in code migrations revealing that developers view AI as a collaborative teammate.

While AI plays a crucial role in expediting code migrations, human oversight remains essential for verification and review purposes. Developers prefer guiding AI based on their expertise while serving as reviewers to ensure accuracy throughout the migration process.

In conclusion, while advancements in AI technology like LLMs have revolutionized code migration processes by significantly reducing time and effort requirements, human involvement remains critical for ensuring accuracy and quality control throughout the transition phase.

Also Read:cybersecurity awareness ,cybersecurity awareness ,Submarine cable dominance ,Collaborative creativity

Conclusion:

• Using a combination of frontier models and robust automation, Airbnb successfully completed a large-scale code migration from Enzyme to React Testing Library (RTL) in just 6 weeks, which was originally estimated to take 1.5 years of manual engineering time. This demonstrates how teams are increasingly turning to AI for managing and migrating codebases, as seen in the case of Airbnb’s ‘Airbnb AI Migration’ process.
• Airbnb’s approach involved breaking down the migration into automated validation and refactoring steps using LLMs. By implementing a system that included prompt engineering and retry loops, they were able to migrate 75% of target files in just four hours. Despite some files failing validation criteria, tools were developed for target re-runs until only around 100 files remained that required manual handling.
• Google and Amazon have also shared their experiences with LLM-driven code migrations. Google reported an acceleration by 50% in their migration processes using LLMs while AWS conducted research on human-AI partnership systems for code migrations. The studies emphasized the importance of human oversight, review, and verification during such migrations despite the advancements made possible by AI technologies like LLMs.

Resources:

Airbnb, Google Cloud, Amazon Web Services (AWS)

Topics : Google,Chromebook, AI, ChatGPT