The gap between teams like this and everyone else is widening fast. AI coding agents have fundamentally changed the rate at which software gets written, but not the rate at which it reaches customers. Commits are surging, and CI/CD pipelines are busier than ever. Yet, features shipped to production have not kept the same pace. The bottleneck is not the agent’s ability to generate output. It is the agent’s access to the knowledge it needs to make good decisions, and the team’s willingness to restructure work around that reality.

We call the teams that have figured this out “frontier teams.” They are not confined to elite labs. They exist across industries and company sizes, and they share a common discipline: they treat AI adoption as an engineering investment, not a tool rollout. Any engineering team can become a frontier team; we can show you how to get there.

AI-native software development treats AI as the foundation of how software is built, with increasingly capable agents directed by human experts. How teams direct those agents determines outcomes. At Amazon, the primary drivers for AI in development were to reduce the time developers spent on non-coding tasks such as documentation, coordination, and operations, retire technical debt, and minimize coding inconsistencies across thousands of small “two-pizza” teams of developers. We have been experimenting across hundreds of engineering teams and have identified at least three paths: a pathfinder initiative with experts tackling a challenge, a structured sprint to execute on a well-defined plan, and an in-situ experiment splitting teams in half between existing approaches and AI-adapted workflows. The paths differ in structure but converge on the same insight.

The pathfinder initiative was a controlled experiment. Six senior engineers received a single mandate: rebuild the Amazon Bedrock inference engine, a project originally estimated at 30 developers working 12 to 18 months. Rather than adding headcount, the team spent its first weeks redesigning workflows around AI, shifting from discrete tasks to goal-driven outcomes, running multiple agents in parallel, and setting up systems for AI to work independently during off-hours. The project was delivered in 76 days. Individual developer productivity increased approximately 20x as measured by normalized commit velocity (the number of commits per developer per week, adjusted for repository complexity and team size). Commits went from 2 per week to 40. The team shipped more high-quality code in five months than it did on projects over the previous ten years, as measured by lines deployed to production.