Skip to content
#DevEx

[2/2] This Year in Uber’s AI-Driven Developer Productivity Revolution (Filed under DevEx)

#DevEx #Management #Leadership #Software #SoftwareDevelopment #Metrics #Devops #Processes #AI #ML

[2/2] This Year in Uber’s AI-Driven Developer Productivity Revolution (Rubric #DevEx)

Continuing storyI'll describe the kids' approach to test generation and migration. n

——— Generating tests ———

  • Guys use the standard testing pyramid, where end-to-end tests are at the top, modular tests are at the bottom. Here are some ideas for using AI in testing: End-to-end testing with AI-agents. It's funny that we've been here for years. 5 Back to the crowd testing where e2e was checked by the guys at Klex, our internal tool like Toloki I. And now it is possible not to attract crowds, but AI agents. Test quality analysis (mutational testing) Predictive selection of tests to accelerate runs At the lower level, you can use AI to generate unit tests. So the guys did Autocover with a focus on regression tests, increasing code coverage and leaving developers in a cycle of interaction with autocover. (keep the developer-in-the-loop). Interestingly, the guys divide the stages of work on tests for deterministic, as well as probabilistic.

——— Java to Kotlin migration ——— Uber historically had a lot of Java code, gradually the guys got hooked on Kotlin and then decided to migrate to it. The platform team made a migration tool and provided it to product teams in a decentralized mode. If this is done gradually and manually, then this migration will end somewhere in the future. 10 years that were not expected The guys decided to automate it using AI. To do this, they decided to combine approaches based on AST analysis. (abstract syntax tree) LLM to improve the rules for working with AST trees This approach allowed them to speed up the migration process by an order of magnitude.

Overall, the guys at Uber are optimistic about the application of AI in SDLC, but note that it is important to choose the right metrics to assess the effect of improvements and their side effects. They suggest using a time-saving metric for developers. (saved developer hours). It reminded me of the metric on the hours of the video I watched on YouTube. Details in my review "Like, Comment, Subscribe" (YouTube. How did the most popular video hosting company conquer the world?).

#Management #Leadership #Software #SoftwareDevelopment #Metrics #Devops #Processes #AI #ML #DevEx