The Source Artificial Intelligence
Article

Overcome AI sprawl with a Value Stream Management approach

Learn how an AI strategy based on Value Stream Management can stop AI sprawl and supply chain constraints and drive ROI.

December 12, 2024 7 min read
Stephen Walters
Stephen Walters Field CTO, GitLab

With the move to platform engineering, organizations can consolidate complex toolchains and deliver higher-quality software faster and with greater security. Now, as teams look to AI for further improvements, we’re seeing a concerning trend: the implementation of disconnected AI point solutions, creating “AI sprawl.”

This AI sprawl amplifies all of the toolchain sprawl problems listed above. Implementing AI as a point solution for a small group of individuals at one stage in the value stream is just as likely to negatively impact your business outcomes as it is to improve them. An 8x boost in development speed means little if it leads to an 8-fold increase in integration overhead, maintenance time, and data reconciliation.

Having a strategy to address AI sprawl is just as important as having a strategy to minimize toolchains — but what does that strategy look like? It’s more than simply the consolidation of AI tools. It’s also about joining up lean processes and establishing clear responsibilities and lines of communication among teams. Value Stream Management holds the answer.

What is Value Stream Management?

Value Stream Management is a methodology that aims to identify areas for improvement in a process to help drive business value more efficiently. When applied to software development, Value Stream Management involves identifying and mapping out all the steps involved in delivering value to customers, analyzing and measuring the flow of work through these steps, and continuously improving the process through automation. In this case, value means any outcome that benefits the customer — whether that's new features, better performance, increased reliability, or enhanced security.

Implementing a Value Stream Management strategy brings numerous benefits to an organization, such as increased transparency, improved communication, reduced waste and bottlenecks, quicker feedback loops, and ultimately faster time to market. By gaining visibility into the entire software delivery process, organizations can identify areas for improvement and make data-driven decisions to continuously optimize their value stream.

Building a value stream and AI strategy

Reducing AI sprawl starts with identifying bottlenecks in your value stream: your process is only as fast as its slowest step. In DevSecOps, our goal is a business outcome achieved by an IT system that is delivered via a fast, safe, and secure supply chain. AI’s purpose should be to limit or remove any potential constraints in the supply chain. This supply chain is a value stream, which is why Value Stream Management is fundamental for building an AI strategy.

One example of a constraint in a software development process is time wasted waiting for security vulnerability information, a security representative, or vulnerability details. AI-powered security scanning and resolution can eliminate this bottleneck.

If we look at the Implementation Roadmap for Value Stream Management from the Value Stream Management Consortium, we can see several important steps in building the strategy before any technology decisions are made.

Value Stream Management implementation roadmap

Diagram courtesy of Value Stream Management Consortium

Here, I’ll focus on three steps. After assessing our current position and determining a vision of what we want to achieve, we must:

  • Identify our value streams
  • Organize roles and responsibilities for each of them
  • Map our people, processes, and technology to understand how all of this connects

Let’s take a closer look at each stage.

1. Identify value streams that deliver business outcomes

Understanding your value streams is critical for AI adoption because it reveals where and how AI can actually improve delivery. The first step in building an AI strategy is to ask yourself: What are your organization’s main processes and workflows that drive business value? And which of those value streams have limitations or bottlenecks — points of constraint — that can be addressed with AI? The answers to these questions will tell you where AI can deliver the best results and what the end goal of using AI should be.

When identifying your value streams, connect them directly to the business goals your IT systems are established to achieve. For AI to enable value delivery of the goal, you will need to consider the entire value stream, end-to-end, from idea to realization, with the required business goals in mind. For example, certain value streams may have regulatory and compliance requirements that other value streams do not. These should be documented in the business goals of the value stream and used to define the potential AI requirements for each value stream.

2. Organize people, processes, and tools around specific roles and responsibilities

Over the last few decades, IT has moved from a Project mindset to a Product mindset. A project mindset focuses on delivering a specific result to fulfill a set of requirements, while a product mindset focuses on the bigger picture, including long-term success for users and customers. However, this shift has often just replaced activity-based silos with artifact-based silos — in other words, we've moved from teams organized by what they do (coding, testing, security) to teams organized by what they make (apps, APIs, platforms). Methods such as Scrum of Scrums try to address this, but typically swap under-collaboration with over-collaboration, where teams spend more time discussing than doing.

With Value Stream Management comes the concept of moving from Product to Flow. Instead of focusing on individual products, teams focus on how value moves — or flows — through the organization. Flow Engineering is then about designing team structures and handoffs around making that journey as smooth as possible.

This shift changes how we implement AI. AI solutions can’t focus on either a singular role or artifact. To remove handoffs and ensure alignment, AI must understand the scope and parameters of its operation, which teams it works with, and when it must be used.

In other words, AI must have a clearly defined role and responsibility, just as humans do. AI should understand its part in enabling flow along the value stream, working in an interactive manner with other people and AI tools. Team Topologies and a Value Stream Reference Architecture are invaluable at this stage because they provide a framework for designing and documenting a team structure that will help your team create value faster.

3. Map the value stream to ensure everyone’s on the same page

AI must be implemented precisely at the point of the constraint to provide the required benefit. In other words, adding AI to the wrong place in your process can actually make things worse. Let’s say security reviews are the slowest step in your workflow because developers need to spend time going back and forth with the security team to understand and address vulnerabilities. In this case, implementing AI only before the security stage to help developers write more code faster is just going to make the bottleneck worse, because there will be even more code (and more potential vulnerabilities) for the team to sort through.

But how do we pinpoint where AI should operate? By mapping the existing workflow or golden path for a value stream. This detailed mapping allows us to identify the precise point of constraint and determine whether AI will provide the required benefit to remove or reduce the impact.

A value stream reference architecture lets us define team actions and map out an ideal future state, showing activities in their most efficient sequence and where AI fits into the bigger picture.

A simplified example of a value stream map for developing a new software feature might look something like this: It starts with a feature request from a customer. Then, developers build the code, which is followed by testing and security scanning before the code is finally deployed to production. The value stream map should include each of these steps as a distinct section.

Continuing with the example above, you’ve identified a bottleneck at the security stage. This point of constraint might have a couple of different potential solutions. Adopting an end-to-end DevSecOps platform will allow you to shift security closer to the developers’ workflow to reduce the cognitive burden. You might also identify an AI solution to help developers understand and resolve vulnerabilities faster. In the previous stage, you would have defined the role and responsibility of AI in enabling flow along the value stream. Now, the value stream map captures this whole picture — which AI tools work where, what they're meant to achieve, and how they help value flow faster through your system.

Addressing AI sprawl: A value stream-based approach

Value stream mapping helps prevent AI sprawl in several ways:

  • Identifying important value streams illustrates how different value streams can rely upon a single AI-powered platform to provide consistency and standards, particularly for regulatory needs.
  • Organizing people, processes, and tools around specific roles and responsibilities allows you to see how that platform should work holistically with context across the length of each value stream.
  • Mapping the value stream allows you to see where and how AI operates at different stages in the value stream so you can identify where duplicate efforts might be creating waste. This will enable flow in the value stream, remove handoffs, improve team alignment, and ensure that AI tools deliver on the organization’s goals.

By repeating the process for different value streams and AI solutions with different goals and roles, you’ll have a framework for your AI strategy.

Conclusion

Organizations can remove waste, improve accuracy, and ensure security by using technology to enhance and automate manual tasks across the software development lifecycle. Historically, this has been achieved through toolchain automation, but organizations can now leverage emerging technologies such as generative AI.

By identifying, organizing, and mapping AI to value streams, we can strategically implement AI to enable flow and remove waste. AI isn’t a standalone solution, but should rather be integrated into a holistic strategy with clear roles and responsibilities. By viewing AI through the lens of Value Stream Management, we see the real key to success: AI's effectiveness depends entirely on understanding how you manage your value streams.

DORA insights: Where is AI really driving developer productivity?

Discover valuable insights from the 2024 Accelerate State of DevOps Report and learn how you can harness AI to maximize team performance and innovation.

Register for the webinar

Key takeaways
  • A strategic approach to AI involves linking it to value streams, ensuring AI is implemented precisely where constraints exist for optimal value delivery.
  • Transitioning from project-based to flow-based thinking enhances team alignment and effectiveness in AI implementations.
  • AI should be seamlessly integrated within a Value Stream Management approach to effectively eliminate inefficiencies and support organizational goals.