An analysis of 5,677 talks from eight major industry conferences (2020–2024) reveals that a tiny cluster of technologies shapes modern software architecture, most tools serve late DevOps stages, and early design phases remain a blind spot.
Source: Table 9 in Su et al. (2025). Each technology can map to multiple DevOps phases. Plan, Code, and Release collectively account for under 11% of tool coverage.
The researchers gathered every available talk title from eight of the largest practitioner-oriented software conferences (KubeCon, AWS re:Invent, QCon, Google Cloud Next, and four others) between 2020 and 2024. From an initial pool of 16,778 titles, they filtered down to 5,677 that genuinely addressed software architecture, using inclusion criteria validated by multiple reviewers.
They then used a pipeline of large language models to extract three things from each title: the technology mentioned, the purpose of the talk, and the context in which the technology was applied. A "large reasoning model" (Marco-o1) did the initial extraction, and three separate validation models (Mistral, Qwen, Llama) independently checked its output. Human experts reviewed disagreements, achieving over 90% overall accuracy for the primary extractor across all dimensions.
After de-duplicating and cleaning, the team identified 450 distinct technologies, 232 usage contexts, and 11 purpose categories. They then classified each technology by its DevOps phase, deployment environment (cloud, on-premise, or both), and cloud provider. Finally, they built co-occurrence networks using Gephi, calculated centrality metrics, and ran community detection to find natural technology clusters.
Four technologies sit at the absolute centre of everything: Kubernetes (1,054 mentions), Cloud Native (584), Serverless (325), and Containers (203). These four connect to over 80% of all other technologies in the co-occurrence network. Kubernetes alone has 64 direct connections and the highest scores on every centrality measure the team computed, whether that is weighted degree, closeness, or betweenness.
The DevOps phase distribution tells a stark story. Technologies cluster overwhelmingly around Build (81%), Deploy (94%), Operate (96%), and Monitor (99%). The Plan phase has just 2.89% coverage. Code sits at 3.11%. Release manages 5.11%. The industry has invested enormously in the right side of the pipeline and left early architectural phases comparatively bare.
Community detection revealed five coherent technology clusters: (1) deployment automation and infrastructure-as-code, centred on AWS EKS, GitOps, and CI/CD; (2) service communication, with Microservices, Service Mesh, and Istio; (3) cloud AI and serverless computing, where Generative AI appeared as the second-most-central node; (4) observability, security, and performance, led by OpenTelemetry and eBPF; and (5) cross-cloud and cloud-edge collaboration, anchored by Multi-Cloud and Hybrid-Cloud.
AI's arrival is late but accelerating. It had zero mentions in the top-10 before 2023, then jumped to 34 mentions that year and 54 in 2024. Generative AI ranked high on betweenness centrality, acting as a bridge between previously disconnected technology domains rather than becoming a hub in its own right.
Over half of all talks (2,945 out of 5,677) fell into the "Introduction & Overview" purpose category. Practitioners are primarily introducing tools and orienting audiences, not documenting architectural vision or migration strategy. Categories like Vision & Roadmapping (130 total mentions), Architecture & Infrastructure (173), and Migration & Modernisation (101) remain marginal, though each showed a clear upward trend from 2020 to 2024.
The researchers frame this bluntly: practitioner conferences provide "a snapshot of the current use cases for technologies, more like a clue than a fundamental shake-up of the state-of-the-art." Architecture-related keywords (design, pattern, architecture) appear far less often than implementation and infrastructure terms (Docker, Kubernetes, CI/CD). Tooling may be mistaken for process maturity. The conference stage favours the shiny and deployable, not the strategic and structural.
24 technologies including AWS EKS, GitOps, CI/CD, and Argo. Focuses on infrastructure-as-code, container orchestration, and deployment pipelines.
22 technologies including Microservice, Service Mesh, Istio, and API. Centres on inter-service communication, architectural decomposition, and migration from monoliths.
22 technologies including Generative AI, MLOps, SageMaker, and AWS Lambda. Covers AI pipeline automation and event-driven computing.
17 technologies including OpenTelemetry, eBPF, Prometheus, and LLM. Focuses on monitoring, distributed tracing, and performance in cloud-native environments.
8 technologies including Multi-Cloud, Hybrid-Cloud, and Edge Computing. Reflects growing interest in vendor independence and edge-cloud coordination.
The software architecture conference circuit tells us where the industry's attention sits: deployment, operations, and monitoring. A handful of technologies (Kubernetes above all) form the connective tissue of modern practice. But the early, strategic phases of architecture, planning, coding, design, get almost no tooling support and minimal conference airtime. AI is arriving fast as a bridge technology, connecting previously isolated domains, though its role is more catalyst than core. Practitioners chase the cutting edge; researchers may be the only ones watching the foundations.
Su, R., Ahmad, N., Esposito, M., Janes, A., Taibi, D., & Lenarduzzi, V. (2025). Emerging trends in software architecture from the practitioner's perspective: A five-year review. arXiv preprint. https://arxiv.org/abs/2507.14554