AI Coding Assistants in 2026: Honest Productivity Numbers
Cutting through the demo theater to look at what AI tooling actually does to engineering velocity.
After three years of breathless productivity claims, the field is finally producing serious longitudinal data. The honest summary: AI coding assistants are a real, measurable lift for greenfield work and well-documented domains; a marginal lift for mature codebases with strong internal idioms; and a negative lift for some categories of debugging where the model's confident hallucinations create new investigation work.
The productivity story is also strongly bimodal. Senior engineers use these tools as a fast typing surface for code they already understand. Less experienced engineers use them as authoritative answers, with predictable consequences. This is not a tooling problem — it is a training and review-culture problem.