How the world's leading engineering teams are tackling legacy code, eliminating technical debt, and modernizing their systems — with measurable results.
Gergely Orosz surveys a wave of AI-driven framework migrations at Asana, Airbnb, and Uber, and asks the harder question: why are companies suddenly willing to touch code they'd avoided for years, and what it means that analyst rankings still put legacy vendors above the AI-native tools actually doing this work.
GitHub's own guidance for teams adopting Copilot agents inside deterministic workflows: let AI take the ambiguous first pass, let CI checks produce the repeatable signal, and keep human judgment on anything high-risk. A useful framework for any team deciding where that line belongs.
Asana's own account of using AI agents to finish a frontend test migration that had sat deprioritized since 2022 — a five-sentence prompt run in parallel across the codebase delivered an estimated $6M of engineering value for about $12K in model costs. A textbook case of AI agents clearing backlog work that never gets prioritized on its own.
Meta's new coding agent is built specifically for planning and validating changes across large, sprawling repositories, not just greenfield features. Another major engineering org betting that the hardest, least glamorous code is exactly where agentic tooling pays off first.