Key Questions Answered
Which AI inferencing tools do cloud native developers consider mature and reliable enough for production deployment?
How do ML orchestration technologies compare in terms of developer satisfaction and recommendation likelihood?
What are the emerging adoption patterns for agentic AI platforms and frameworks?
Can technologies that span multiple use cases (like BentoML) achieve market leadership across all categories, or do specialized tools maintain advantages?
Click to expand
Methodology
In our research, we employed Likert scales to capture developers' opinions on the maturity and usefulness, from 1 to 5 stars, of the various multicluster application management and batch computing technologies surveyed. While these ratings are inherently subjective, reflecting individual perceptions and experiences, they provide valuable insights into the developer community's views. The nature of our research is centered on investigating developer perceptions of these aspects, making the subjective nature of the ratings not only acceptable but also valuable for our analysis. Although the subjective nature of Likert scales may influence the interpretation of results, as different respondents may have varying standards for rating, this variability enriches our understanding of the developer experience.
Respondents were initially asked about where their projects ran or were deployed, to identify their position as a ‘cloud developer.’ Following this, they were asked which technologies they were currently using that we associate with cloud native development approaches, including technologies such as Infrastructure as Code, service meshes, and serverless computing.
Respondents were recruited from third-party panels. For privacy and data minimization purposes, exclusion is based on internal consistency and survey-taking behavior metrics. As such, information on the organization the respondent works for is not carried through to any analysis. This privacy also helps encourage greater honesty from respondents, who do not have concerns that their expressed opinion will be associated with them.
About this Report
As AI and machine learning workloads increasingly converge with cloud native infrastructure, understanding which tools developers trust and recommend becomes critical for technology strategy. This Technology Radar report surveys over 300 professional developers working with cloud native technologies to assess their experiences with AI inferencing engines, ML orchestration tools, and agentic AI platforms.
The research categorises technologies into four adoption tiers — adopt, trial, assess, and hold — based on developer ratings of maturity, usefulness, and likelihood to recommend. For AI inferencing, NVIDIA Triton, DeepSpeed, TensorFlow Serving, and BentoML emerge as adoption leaders. In ML orchestration, Airflow and Metaflow achieve adopt status, while BentoML demonstrates cross-category versatility. For the rapidly evolving agentic AI space, Model Context Protocol (MCP) and Llama Stack show strong developer confidence.
The findings reveal important patterns: technologies can excel in maturity while struggling with usefulness for specific use cases, high recommendation rates don't always correlate with top maturity scores, and cross-functional tools like BentoML can succeed in multiple domains without dominating any single category. With 41% of ML/AI developers now classified as cloud native—a proportion expected to grow — this research provides actionable intelligence for organisations building AI/ML infrastructure strategies. The report also highlights the CNCF ecosystem's crucial role in cultivating technologies from experimental innovations to production-ready infrastructure.