AI ops is the bottleneck

- Datadog's 2026 State of AI Engineering argues the main constraint for scaling AI is operational complexity — not raw model capability — calling out governance, tracing, routing and cost visibility. - The report is based on telemetry from thousands of AI-agent environments and reframes the problem as platform engineering rather than model selection. - That points platform teams toward building control planes for telemetry, cost attribution, policy and failure analysis rather than chasing model benchmarks. (datadoghq.com)