CNCF finds strong bond between cloud native dev & AI tools

Artificial intelligence tools are gaining wider adoption as developers and enterprises in cloud-native environments seek more scalable solutions for AI workloads.

According to an October report by the Cloud Native Computing Foundation (CNCF) and SlashData, the industry's preference for AI inference engines, machine learning orchestration platforms, and agentic AI systems is reflecting evolving preferences as more organisations embed AI into their core operations.

The data demonstrates a significant alignment between AI/ML system design and cloud-native principles, including containerization, orchestration, scalability, and reliability. 41 per cent of developers working on AI and machine learning now identify as cloud-native, with expectations that this share will continue to rise as organisations pursue scalable, flexible infrastructure for advanced workloads.

Inference tools

NVIDIA Triton, DeepSpeed, TensorFlow Serving, and BentoML were the most widely adopted AI inference tools among developers working in cloud-native environments. Triton, in particular, stood out with the highest maturity and usefulness ratings: 50 per cent of developers gave it a five-star maturity rating and 41 per cent a five-star usefulness score.

Alternative inference tools attract loyalty in niche roles. Adlik, an end-to-end framework optimiser for deep learning models, was less used overall but garnered a 92 per cent recommendation rate from its user base.

Orchestration systems

The orchestration category saw Airflow and Metaflow designated as the principal 'adopt' solutions by developers. Metaflow led for maturity, with 84 per cent awarding it four or five stars, while Airflow topped usefulness and recommendation scores. Notably, respondents gave Airflow no 'low usefulness ratings, suggesting high developer satisfaction with the tool's capabilities.

BentoML, which straddles both inference and orchestration, achieved an 'adopt' rating for inference tasks but only a 'trial' rating for orchestration, highlighting the challenges such multi-purpose tools face in achieving consistent performance across different application areas. CNCF-affiliated tools Argo Workflows and Kubeflow earned a 'trial' classification in orchestration. Meanwhile, platforms like Flyte and Seldon Core recorded middling approval.

"These findings show just how diverse the AI/ML toolchain has become. Metaflow and Airflow are excellent examples of how developer trust is earned through stability and fit-for-purpose design," said Liam Bollmann-Dodd, Senior Market Research Consultant, SlashData. "But even newer projects like Flyte and Seldon Core are showing traction that signals opportunity for differentiation."

Bob Killen, CNCF's Senior Technical Program Manager, told TechDay that it's interesting that many projects built for CI/CD integration, such as Argo, are being used more for AI/ML workloads.

"A lot of groups have really gone to them, because it's also what they've used historically," he said. "Argo was built for CI/CD...but Kubeflow uses Argo under the hood."

Agentic AI platforms

Within agentic AI, Model Context Protocol (MCP) and Llama Stack were the only tools to achieve 'adopt' status. MCP recorded the highest combined four- and five-star usefulness ratings at 80 per cent, along with a wide developer base, indicating strong enterprise traction for agent-based approaches, such as AI-driven customer support.

Killen added that the rate of acceleration for MCP stood out among the data released in this report. 

"I would not have seen [MCP] that far along, considering it was released a year ago. That is an incredibly fast rate to go from an early adopter to trialing it to using it everywhere," he said.

Agent2Agent (A2A), though less mature and newer in the segment, achieved the highest recommendation rate. 94 per cent of users said they would recommend A2A. The report suggests that this outcome indicates developer optimism and the perceived trajectory of a newer tool can sometimes outpace its established capabilities.

Image courtesy of CNCF.