The U.S. Congress recently banned its staff from using Microsoft’s AI Copilot — an integrated large language model chatbot that enables automation for Microsoft’s products such as Word, Excel, PowerPoint, Outlook and Teams — citing security concerns. And they are not alone in this sentiment, as many professionals teeter between excitement and dread when it comes to AI. In the meantime, Microsoft has confidently touted Copilot, promising that the power of its AI will reduce the daily grind of running a business. The company has segued from merely talking about AI to infusing it across every layer of its tech stack. Its recent introduction of Copilot Runtime even allows developers to use AI in their own programs, so it’s no surprise that most chief information officers are experimenting with Copilot. Its productivity promises and the resulting scope expansion and cost efficiencies for potential reinvestment are too enticing to ignore. Microsoft is certainly leading by example, and the only question is how others should follow. Over the last decade, agile organizations have outperformed others in making and managing decisions more quickly. Agile adoption in IT has introduced changes in the way infrastructure, applications, data and skills are produced, consumed, or both. The foundational elements of agile, such as collaboration, automation and continuous improvements, are the key sources of innovation pertinent to application, development and deployment.
Recent events have disrupted agile progression in IT. The COVID-led era of flexibility in how, when, and where we work is challenging the status quo in the ways we collaborate. In addition, Generative AI and the associated complexities to IT governance have also shaken up agile progression. Consequently, an incremental economy has been created that forces every company to take the opportunities and the challenges more seriously.
Distributed Agile has already been in practice for quite some time. The remote and hybrid work realities are just extensions of what we have already seen in distributed teams. However, the productivity promises of tools such as Copilot are new, so assuming that current Agile practices will work with GenAI practices is a mistake. So what can we do?
Here are some thoughts on how to integrate agile methodology amidst the Copilot adoption.
Extend DevOps to Include Representation From DataOps and MLOps
Given the importance of data as well as AI and machine learning models, DevOps teams must include representation from the DataOps and MLOps teams (ModelOps is a subset of MLOps). Only then can the objective of bringing the “production” and “operations” closer be achieved. At first glance, replacement may seem to be AI’s biggest threat, but its first act will rather likely be to reveal and deepen cracks in collaboration.
Software Intelligence is More Important Than Ever
Not understanding the application systems holistically before auto-producing the code in production will be disastrous. Corporate IT is a mixture of AI and non-AI applications and IT assets. Additionally, the “explainability” of the code can be achieved only when the software intelligence on the code that GenAI produced is attained. Ultimately, it is not the functional correctness, but the architectural fitment that matters most for unlocking productivity improvements. AI moves fast, but letting speed take too high a precedence sets the stage for failure.
Continuous Compliance and Continuous Security are Equally Important
One of the main concerns of GenAI tools is the vulnerabilities that auto-generated code can introduce to corporate IT, which is the main reason the U.S. Congress has banned the use of Copilot. It is important to make proper adjustments at the pod level for compliance and security, so that they are designed and delivered continuously, rather than audited and assured periodically. Regulation is famously behind innovation, and companies must intentionally think ahead of the curve to avoid future pain.
Augmenting the Quality Gates in Your CI/CD Pipe for AI Assistants
The founding principles of open source — transparency, inspection, and adaptation — can be extended to GenAI products l. “Inspection” should not only cover the quality, performance, security and UX aspects of the code that the tools provide, but also the architectural fitment in the enterprise IT.
Measure Success, and Be Transparent With Shortcomings
AI’s impact can be foggy, but some results must be measurable to justify its adoption. Crafting AI-specific KPIs can help solidify the Copilot’s role on the team. Finding the right metrics to measure is of the utmost importance, and is a large challenge in itself.
In the detailed review of AI’s impact, it’s also important to openly accept shortcomings. The system is, of course, imperfect by nature, and these imperfections need to be tracked and addressed. Because AI is evolving so rapidly, many issues will likely be solved in the short term. Note the flaws, and be intentional with revisiting them at regular intervals.
A Mismatch of Skills Will Impact Productivity Promises
Tools are only as good as those who wield them. An experienced developer can demonstrate an above-average productivity level with an AI assistant, but an inexperienced developer can quickly create more problems than solutions. Training the development and quality assurance communities to master the tools and guidelines for the code and test governance requires adjusting the Agile operating model.
Remember that DevOps teams are more than developers, DataOps teams are more than data engineers, and ModelOps teams are more than data scientists. The cross-disciplinary skills of the DevOps teams will significantly change when AI is part of the conversation. As the boundaries between these disciplines melt away, those willing to adapt will rise to the top.
Adjusting the methodology for the perceived challenges should not curtail the potential benefits that GenAI can produce. If used properly, GenAI can help in hyper-automation of development and quality assurance tasks, evaluating design options through rapid prototyping, simplifying the documentation process, monitoring the production environment to predict performance bottlenecks, and so forth.
If we don’t adjust our agile methods to match these new realities and unlock value at a faster pace, “time to market” and the associated cost benefits will be perceived poorly. Changes to agile methodology are inevitable because GenAI and Agile are providing real competitive advantages. Don’t take it easy.
