Overview

Uber has transformed from using AI for pair programming assistance to full peer programming with autonomous agents. Their shift to agentic AI has enabled engineers to focus on creative work while delegating routine tasks like upgrades, migrations, and bug fixes to AI systems that run asynchronously.

Watch the Video

Key Takeaways

  • Focus AI on eliminating toil work first - 70% of initial agent workloads were routine tasks like upgrades and migrations because they have clear start/end states and higher accuracy rates
  • Build abstraction layers for technology flexibility - The AI landscape changes rapidly, so architect systems that can swap underlying models and technologies without rebuilding entire platforms
  • Engineer adoption requires peer influence, not mandates - Top-down directives had limited impact, but sharing wins between engineers created viral adoption as developers trust other developers over management
  • Activity metrics don’t equal business value - High developer satisfaction and code generation volumes are positive signs, but connecting AI productivity gains to actual revenue impact remains an unsolved challenge
  • Cost management requires intelligent model selection - AI expenses can grow 6x quickly, necessitating systems that automatically choose appropriate models for planning vs execution tasks to optimize both performance and costs

Topics Covered