Datadog Deleted All Its AI Context. It Worked.

Datadog’s transition to agentic development centers on balancing rapid tool adoption with rigorous, data-driven governance. Managing 4,000 engineers requires a shift from manual oversight to automated evaluation platforms that replay historical incidents to assess AI performance. A critical insight emerged when deleting stale context files from repositories actually improved agent performance, highlighting the risk of context rot and the necessity of prioritizing problem-solving over pre-emptive training. Beyond technical implementation, this shift redefines hiring and career progression, moving away from traditional LeetCode-style interviews toward real-world scenarios that test engineering judgment and agent-assisted problem-solving. By using AI to facilitate rapid prototyping, teams can now validate assertions through concrete proofs-of-concept, ultimately increasing organizational ambition rather than merely optimizing for incremental productivity gains.