As AI Adoption Expands, Enterprises Face Operational Limits — Datadog Releases 2026 State of AI Engineering Report

NEW YORK – June 2, 2026 – Datadog, Inc., a leading observability and security platform for cloud applications, today released its 2026 State of AI Engineering report. The report reveals that as AI adoption accelerates, the biggest barrier to operating AI at scale and reliably is not the intelligence of the models themselves, but operational complexity. The report is based on real-world data from thousands of organizations running AI in production, showing that complexity is increasing exponentially as AI systems scale. Currently, about 70% (69%) of companies use three or more AI models, and agent workflows are becoming increasingly complex. Additionally, about 5% of requests to AI models in production are failing, with about 60% of those failures caused by capacity limits. This leads to processing delays, errors, and degraded user experiences in AI-powered applications. Key findings include: multi-model usage is the standard, with OpenAI holding a 63% share while Google Gemini and Anthropic Claude adoption is growing. Agent framework usage has doubled year-over-year, and the volume of data sent to AI models is also increasing. Yanbing Li, Chief Product Officer at Datadog, stated, 'AI is in a situation similar to the early days of the cloud. Successful companies are not those that build better models, but those that build management systems to operate them properly.' Guillermo Rauch, CEO of Vercel, also emphasized that observability is essential to prevent agent failures.