Kaushik Joshi, Hemant Patil, Harsh Patel, Pravin Wavare

DOI: https://doi.org/10.5281/zenodo.24129040

In today’s cloud Amazon Web Services (AWS), Microsoft Azure and Google Cloud Platform (GCP) environments, traditional incident response becomes cumbersome, and vulnerable to human error. While Artificial Intelligence for IT Operations (AIOps) helped with machine learning in the area of anomaly detection, it remained advisory. The recent advances in Large Language Models (LLMs) and Agentic AI enable the construction of cloud architectures that can self-repair and function autonomously.With the recent advancements in Large Language Models (LLMs) and Agentic AI, it becomes possible to develop cloud architectures capable of self-healing and self-operationalizing. While there are many works that introduce a theoretical approach, or review a single vendor’s built-in AI incident-response agent, no empirical study directly compares the built-in AI incident-response agents of the major hyperscalers in a similar workload and using the same fault injections. In AWS, Azure SRE Agent and AWS DevOps Agent, and in Azure and GCP, Azure SRE Agent and GCP Security AI, a standardized application (cloud_incident_tester) is deployed to allow for a fair comparison of the two agents. Detection Latency, Investigation Wall Clocks Time, Root Cause Analysis (RCA) accuracy, Tool execution error recovery, and Governance guardrails are what we measure. Azure SRE Agent reduces autonomous resolution time (3m 40s- 5m00s), but needs to have a binding response plan, and will incur continuous billing costs of $0.40/hr. AWS DevOps Agent provides a deep investigative capability for a multi-region infrastructure with parallel subagents (9m 30s to 15m 23s), and will identify misconfigurations to the infrastructure without relying on a governance model– it just proposes them. The difference between GCP and other autonomous agents is that GCP does not provide any individual agent for users, but only enterprise organization security features with AI. The findings provide useful reference points for technology leaders and reveal configuration issues at a system level that could affect the reliability of a cloud.

AIAgents, AIOps, Autonomous Systems, Cloud Computing, Incident Response , Native AI Agents

In this empirical research, we discover that this increase is not theoretical; function ranges differently across various cloud service providers supplying self-governing cloud AI incident reaction solutions. The trade-offs of Azure SRE Agent are that it has a quick investigation time (less than 5 minutes), is focused enough to follow up fast alert investigations ($0.40/hr cost of ongoing investigation), and is very quick to investigate scenarios from the onset. AWS DevOps Agent provides the most of the details of the investigation, cross-region visibility, and categorization of 2nd order defects within 9-15 minutes of the proposed system’s pure governance model.Systemic Configuration Pitfalls are operational risks due to the failure of AWS regex filter, GCP aggregation default, Azure response plan prerequisite, etc. that are hard to monitor. It is believed that watch-less pipelines can blow unexpectedly.When implementing an engineering cloud AI agent: Ensure all the write gates are only for the creation of proposals; set alarm persistence rules with stealthy alarm writing syntax until it is fully understood; cross-validate writing of the proposal alarm syntax with the native engineering tools; and, last but not least, do not take for granted that AI will always provide the right answers when an incident occurs

1. Kephart, J. O., & Chess, D. M. (2003). The vision of autonomic computing.
2. IEEE Computer. Zhao, L., et al. (2019). Real-world challenges in AIOps: A systematic study. ICSE.
3. Du, M., et al. (2017). DeepLog: Anomaly detection and diagnosis through deep learning. CCS.
4. Ghosh, R., et al. (2023). AIOps for cloud reliability. IEEE Trans. Cloud Comput.
5. Chen, Y., et al. (2025). IRCopilot: LLM-assisted incident response. arXiv:2505.20945.
6. Kumar, A., et al. (2026). AIDR-Cloud: AI-driven incident response. IEEE SoutheastCon.
7. Patel, S., et al. (2026). AI Cloud Sentinel: Autonomous cloud monitoring and self-healing. ICICV.
8. Liu, X., et al. (2025). Multi-agent LLM orchestration for deterministic incident response. arXiv:2511.15755.
9. Wang, M., et al. (2026). Autonomous multi-agent SRE architectures using CrewAI. IEEE Access
10. Zhang, H., et al. (2025). IFSHM: Integrated LLM and deep reinforcement learning arXiv:2506.07411. for self-healing microservices.
11. Smith, K., et al. (2025). Beyond AIOps: Systematic review of autonomous cloud operations. ACM Comput. Surv.
12. Tan, J., et al. (2026). LLMs and Agentic AI for cloud reliability: A systematic literature review. IEEE Trans. Serv. Comput.
13. Davis, E., et al. (2026). SecRespond: Evaluating LLM agents in security incident triage. arXiv:2607.26791.
14. Wilson, B., et al. (2026). SIR-Bench: Benchmarking security incident response agents. arXiv:2604.12040.
15. Martinez, C., et al. (2026). OpenSec: Open benchmarks for cloud security automation. arXiv:2601.21083.
16. Lee, D., et al. (2026). ORCA-bench: Benchmarking AI agents for cloud operational reasoning. arXiv:2607.28545.
17. Gupta, F., et al. (2026). SIABENCH: Comprehensive evaluation of AI agents in network and memory forensics. arXiv:2603.06422.
18. Taylor, R., et al. (2026). Self-healing cloud systems: A critical survey and evaluation framework. ICIICE.
19. Anderson, T., et al. (2026). LLM multi-agent systems survey: Architectures, challenges, and cloud applications. Springer Comput. Sci. Rev.
20. AmazonWebServices, (2026). AWS DevOps Agent User Guide and Technical Reference. AWS Documentation.

Kaushik Joshi, Hemant Patil, Harsh Patel, Pravin Wavare (2026). A Comparative Literature Review of Built-in AI-Powered Incident Response Agents Across AWS, Microsoft Azure and Google Cloud Platform. International Journal of Computer Techniques, 13(6), 647–650. ISSN: 2394-2231. DOI: https://doi.org/10.5281/zenodo.24129040