Datadog State of AI Engineering 2026: 5% of LLM Calls Fail in Production, 69% of Input Tokens Are System Prompts, Framework Adoption Doubled YoY

Datadog·high signal

Datadog's report based on thousands of production AI deployments reveals that 5% of all LLM spans report errors (60% from rate limits), 69% of all input tokens in customer traces are system prompt overhead, and 69% of companies now use 3+ models. Framework adoption (LangChain, Pydantic AI, Vercel AI SDK) doubled from 9% to 18% of organizations. Anthropic Claude grew 23 percentage points while OpenAI maintains 63% provider share.