AI Innovation Pipeline

I think the Mantra move fast and break things got a bad rep because, you know, it broke things. I think the better Mantra is move fast and be responsible.

— Andrew Ng, [1]

What is the AI Innovation Pipeline?

Outcome-driven product development, at the heart of AI-Native SAFe, is a process that connects intended outcomes to customer and business results and harnesses the resulting feedback to guide future cycles. Each AI-Native ART builds and manages an AI Innovation Pipeline to power the execution that converts these intended outcomes into realized value through a series of outputs, as shown in Figure 1.

Although AI-Native Teams are free to generate as many types of output as needed to help them manage their work, at a minimum, four types of work items are recommended, as these encourage effective prioritization across the key categories of learning-oriented (experiments and prototypes) and value-oriented (features and enablers). The AI Innovation Pipeline supports AI-Native Teams by accelerating the flow of their work from discovery to release and by incorporating the feedback that work generates, thereby amplifying the customer and business value it produces. 

No single lifecycle fits every organization, ART, or product, but the journey from planned outcome to realized value, shown in Figure 1, can be generalized by the following stages:

• Discover refines the intent behind the work: the problem it addresses and the outcomes it is meant to produce, whether it tests a hypothesis through an experiment or delivers value directly via a feature.
• Specify elaborates that intent into specifications, the ‘what and how’, along with a context, that will be utilized by the team and AI agents.
• Build produces the output defined by the specification, ranging from a lightweight probe for a learning output to an integrated, release-ready candidate for a value-capture output.
• Validate tests the output against its purpose. For a learning output such as a prototype, this means checking the assumptions and hypotheses that prompted it; for a value-oriented output such as a feature, this means determining whether it is useful, appropriate, and ready to release.
• Release delivers features to production and into users' hands, turning real use into value and the associated telemetry into learning. Experiments and prototypes usually conclude at the validation stage; however, techniques such as A/B testing may also result in the release of experiments.

The pipeline’s coverage of the entire lifecycle also provides a rich surface for visual management of the work as it progresses. Teams and ARTs often find it useful to create Kanban systems, such as the simple example in Figure 2 below. The combination of this type of visualization with meaningful flow metrics provides critical feedback as the ART seeks to bring SAFe Principle 6 to life and “make value flow without interruptions.”

What are the components of the AI Innovation Pipeline?