Building an AI-native engineering org

FurtherAI’s experience over nine months suggests that AI-native engineering requires redesigning the whole delivery process, not simply giving developers coding agents. Since January, each engineer has merged about five times as many code changes daily while keeping reverts below 0.5%. The team progressively automated coding, test creation, review, CI, bug intake, and enforcement of engineering rules: 91% of changes now include tests, AI reviews 99%, and CI runs 20 times as many tests in the same 17 minutes. Automated checks and AI review let up to a third of low-risk changes merge without human involvement. Slack-based agents also help teams collaborate, investigate production issues, and let non-engineers contribute code. Building this infrastructure took substantial effort; customer-facing changes per engineer rose from 17-30 monthly to 45 in September, with October projected at 48.

The faster coding exposed a new constraint: deciding what to build and reviewing outcomes. In the example given, making implementation 3.5 times faster cuts a ten-day feature cycle to only 1.75 times faster overall, because planning and review remain slow; together, they now consume 70% of the work. The proposed response is to give engineers greater end-to-end ownership rather than add management layers. Seniority increasingly means being able to plan, make tradeoffs, measure results, and stay close to customers. Managers, in turn, should understand the product firsthand, make informed product decisions, and develop engineers’ judgment. The remaining challenge is scaling organizational decision-making to match the speed of automated execution.