Abstract

DevOps has become an integral part of modern software development, emphasizing communication and collaboration between software developers and IT operations professionals. However, with the rise of artificial intelligence (AI), the future role and necessity of devops has been brought into question. This paper explores the prospects for devops in an increasingly AI-driven world, including devops’ potential contributions to and collaboration with AI, as well as the possibility of eventual replacement by intelligent systems. Both theoretical analysis and real-world trajectory suggest devops retaining relevance, albeit with an evolving role and skillset demands.

Introduction

Over the past decade, devops has rapidly gained prominence as a software development methodology integrating development and operations teams and practices. The devops movement emphasizes collaboration, communication, integration, automation, and measurement of cooperation between software developers and IT operations professionals (Lwakatare et al., 2019). This interdisciplinary approach leads to benefits like faster release cycles, more reliable releases, and better-aligned business objectives (Erich et al., 2017).

However, as artificial intelligence (AI) and machine learning (ML) continue advancing at a rapid pace, the future role and need for human devops engineers has been called into question (Miller, 2019). Some speculate that AI could automate much of the routine work currently performed by devops teams, especially in areas like infrastructure provisioning, configuration management, monitoring, and incident response. This raises questions regarding the prospects for devops in an AI world.

This paper analyzes the future of devops in AI environments, assessing areas where AI could replace or enhance devops work as well as future directions for the practice as a whole. The potential for devops and AI collaboration is also explored. Both theoretical implications and real-world trajectory point towards devops retaining an integral, albeit evolving, place in AI-centric software ecosystems.

The Promise and Perils of AI Automation

Much devops work focuses on automating key processes like building, testing, releasing, and monitoring software systems. As ML/AI matches or exceeds human capabilities in perceiving patterns and making analytical judgements (Agrawal et al., 2018), the question arises whether AI could replace substantial elements of devops work. Infrastructure provisioning and configuration management are prime areas where ML could conceivably automate decisions and implementation.

AI also shows promise for improving monitoring and incident response by analyzing system logs, metrics, and traces to automatically detect anomalies, diagnose issues, and suggest fixes without human intervention (Scully, 2019). Such innovations could greatly reduce manual alert triaging and troubleshooting burdens on devops teams. AI could also potentially write test cases, do code reviews, and take over software release management processes.

However, while AI will empower automation of many devops tasks, full replacement of human devops appears unlikely in the foreseeable future. Mission-critical systems require nuanced judgement calls which advanced AI has yet to achieve. Scenarios involving new conditions and contexts where historical training data doesn’t apply pose challenges as well (Bughin et al., 2017), along with interpreting and implementing business objectives. Humans thus seem likely to maintain central roles directing and supervising AI to ensure sound judgements.

The Evolving Role of DevOps

As AI proves capable of handling a growing range of conventional devops work, human devops roles seem set to evolve from operational focus towards higher-level software development and infrastructure engineering roles. Understanding business contexts and setting strategic objectives will grow in importance compared to tactically executing routine tasks. Deeper software development skills like coder training, algorithm optimization, and designing for maintainability will likely become more central for devops as well (Miller, 2019).

Infrastructure architecture and engineering skills will also become more essential as environments grow more automated. Devops will need to design increasingly dynamic and intelligent infrastructure platforms, architecting automated policy engines to manage provisioning, configuration, scaling, healing, and optimization based on business objectives. Human oversight and guidance focused on managing the managing systems, not just managing infrastructure, will play a key role (Morris, 2020).

The importance of monitoring, metrics, and observability to feed AI systems suggests data and analytics skills will also become more important within devops. Understanding what events and metrics carry meaningful signals within complex and noisy systems requires engineering sophistication around data capture, storage, and analysis. Such observability capabilities will serve as the foundation for training and continuously improving automated systems.

The Promises and Perils of Human-AI Collaboration

Rather than full automation and replacement of devops, though, the most impactful paradigm seems likely to involve effective collaboration between human devops experts and AI automation. While AI can handle huge volumes of routine work, adapt to new contexts, and detect signals across massive datasets, human judgement still outpaces AI in many regards — especially in interpreting the significance of observations in business contexts and defining strategic direction.

As machine learning researcher Andrew Ng noted, “It seems that the more we automate, the more critical it is to have AI work well with humans and empower humans” (Levy, 2021). Complementary human-AI partnerships appear well-positioned to leverage the strengths of both. The most capable future systems will feature tight human-AI loops where human input continues training automation and automation continues enhancing human insights.

Devising reward functions that accurately align observed metrics with business objectives poses one major challenge for automation (Amodei et al., 2016), as does transparency around how AI systems arrive at conclusions. Bias in algorithms and training data also requires ongoing human awareness and redress (Mehrabi et al., 2021). Ethical application of AI likewise remains an inherently human responsibility. Careful oversight and governance of intelligent systems seem essential to delivering robust, trustworthy automation.

Fortunately, promising approaches like Google’s Model Cards and IBM’s AI FactSheets aim to increase visibility into model provenance, intended uses, testing approaches, and performance limitations to support more ethical and transparent AI adoption (Mitchell et al., 2019). Ongoing progress around best practices for human-centered AI design and deployment spark optimism for synergistic human-AI partnerships where automation enhances rather than replaces human expertise.

Conclusion

In conclusion, while AI shows promise for automating a growing array of devops functions, full replacement of human devops in the foreseeable future appears unlikely. Although routine operational tasks will increasingly fall to automation, human skills like strategic judgement, creative troubleshooting, ethics, and business understanding remain unmatched by AI. The most potent paradigm will likely involve integrated human-AI teams where automation handles scaling operational burdens as human devops shift towards more strategy, design, and optimization-centric roles. Adoption of transparency and accountability measures around AI will prove critical to fostering effective collaboration. With responsible implementation, devops teams have much to gain from partnership with artificial intelligence.

References

Agrawal, A., Gans, J., & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Press.

Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., & Mané, D. (2016). Concrete Problems in AI Safety. arXiv preprint arXiv:1606.06565.

Bughin, J., Hazan, E., Ramaswamy, S., Chui, M., Allas, T., Dahlström, P., Henke, N. & Trench, M. (2017). Artificial intelligence: the next digital frontier?. McKinsey and Company, 47.

Erich, F., Amrit, C., & Daneva, M. (2017). Cooperation between software development and operations: A literature review. Proceedings of the International Conference on Software and System Processes (ICSSP 2017), 26-35.

Levy, S. (2021). In conversation with Andrew Ng. Wired. https://www.wired.com/story/andrew-ng-artificial-intelligence-graduate-school/

Lwakatare, L. E., Kilamo, T., Karvonen, T., Sauvola, T., Heikkilä, V., Itkonen, J., Kuvaja, P., Mikkonen, T., Oivo, M., & Lassenius, C. (2019). DevOps in regulated software development: Case medical devices. Information and Software Technology, 105, 1-16.

Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Computing Surveys (CSUR), 54(6), 1-35.

Miller, R. (2019). Will DevOps Survive Artificial Intelligence?. DevOps Institute.

Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model cards for model reporting. Proceedings of the conference on fairness, accountability, and transparency, 220-229.

Morris, J. (2020). The future of DevOps is convergence of humans and machines. DevOps.com.

Scully, M. (2019). The Machine Learning Feedback Loop that’s Changing DevOps. Medium.