Outputs, not just features
An AI-Native ART produces four kinds of output. Experiments and prototypes exist to learn: they conclude with the uncertainty they reduced. Features and enablers exist to deliver value: they release with verification evidence. Every output hangs off an outcome and a metric, and stopping one is as easy as advancing it: discard from any stage, with a reason.
The pipeline and its gates
| Stage | Gate to leave it |
|---|---|
| discover | an outcome and a metric are linked |
| specify | a specification exists, the intent record is clean (no pending review) and the metric is set |
| build | a verification owner is named |
| validate | value items need verification evidence; learning items record the uncertainty reduced |
| release | released, or concluded for learning items |
Intent, specification, context
Three records carry what an AI system is meant to do. The intent record states the current and future state and the measure. Specifications and context records derive from it. When an intent changes, every derived record is marked for review and the item is held at Specify until each review is cleared; the evidence log keeps the trail.
Curated data
- Complete: People and agents can reach all the knowledge a task requires.
- Accurate: Verified, trustworthy, lineage traceable to source.
- Securely accessed: The right people and agents reach the right data, and no more.
- Maintained: The data stays current.
Team capability coverage
- Product: Defines the why, anchors to customer needs, synthesises feedback.
- Builder: Creates the solution.
- Domain expert: Deep subject knowledge.
- AI: Named as a team capability, not a tool; humans review its output for quality and bias.
The twelve success factors
| Shift | Factors |
|---|---|
| Outcome-driven flow | Outcome-driven product development; Portfolio outcomes and investment strategy; Curated data |
| Cross-functional, AI-augmented teams | AI-Native ARTs; AI-Native Teams; AI Value Architect |
| Iterative learning and rapid experimentation | Product vision and roadmap; PI Outcome Planning; Sense and Respond |
| Development and innovation at scale | Intent, specifications and context; AI Innovation Pipeline; AI governance and ethics |
Sense and Respond
Every iteration the ART answers three questions from data, not slides: what did we learn, what moved, what should we stop. Trainyard pulls the agenda from the records: key result deltas, what released, discard candidates (no metric linked, no movement since release), the review queue and AI cost variance by workflow.
Updated 2026-10-22. SAFe® is a registered trademark of Scaled Agile, Inc. All guides.