Overview
- An evidence review finds AI shifts value from first‑draft code generation toward verification, security, maintainability and business alignment.
- OECD syntheses report context‑dependent productivity gains of roughly 5% to over 25%, underscoring how task type and setting determine outcomes.
- The METR randomized trial found developers using early‑2025 AI tools were about 19% slower on real repository tasks despite believing they were faster.
- A January 2026 survey indicates many developers do not consistently verify AI‑generated code before committing it, creating growing verification debt risks.
- Dallas Fed research shows employment declines for younger workers in high‑AI‑exposure occupations, while demand rises for engineers who can design, operate and govern systems as tooling moves toward agentic workflows.