A Reader's Guide to the ServiceNow AI Agent Well-Architected Review
The Well-Architected Review for AI Agents is a 56-page document, and how you navigate it matters as much as what you take from it. This guide orients you before you dive in.
Who it's for:
The framework is written for multiple audiences: architects and developers building agents, security and compliance teams reviewing them, operations teams running them in production, and business leaders accountable for AI outcomes. Not every pillar will be equally relevant to you, and that's by design.
What's inside:
There are six pillars in the framework:
- Agent Design & Architecture
- Security & Compliance
- Reliability
- Operational Excellence
- Cost Optimization
- User Experience
Each pillar follows the same structure. It starts with Design Principles which provide foundational knowledge to answer the Assessment Questions. The Assessment Questions are the spine of the framework so answer truthfully. The Leading Practices are where the implementation depth lives and should be used to dive deep or during remediation activities.
A note on Leading Practices:
The practices included in this framework are intentionally generalized – they're a strong foundation, not an exhaustive list. You'll likely need to adapt them to your organization's specific context, risk tolerance, and platform maturity. If you're looking for greater depth on a topic, Appendix B (Further Reading & Resources) is a good next stop.
When to use it:
The framework supports three usage modes: pre-build design guidance, post-build assessment, and ongoing governance. These modes are explained in the "Running and Scoring the Framework Assessment" section just before the appendices. This section may be worth reading first if you're unsure where to start.
How to navigate it:
Several topics intentionally span multiple pillars. A quick pass through the document before diving deep will give you a feel for where topics live and where to go when you want more detail.
How to handle N/A:
Throughout the assessment questions you'll encounter conditional callouts instructing you to mark questions N/A if a feature doesn't apply to your workflow. N/A excludes a question from both numerator and denominator in your score calculation — it doesn't count against you. Mark N/A honestly based on your actual workflow configuration.
How to score it:
Don't calculate by hand. An interactive scoring tool is available through the ServiceNow AI Center of Excellence – contact your Account Team or Impact Team to request access. If you do score manually, all methodology, maturity level thresholds, and critical pillar requirements live in Appendix A, not within the pillar sections themselves. A high overall score does not override critical pillar failures for Security or Reliability.
Get the latest document:
This framework was written with the Yokohama and Zurich releases in mind and will be updated periodically as the platform evolves. Check back for improvements, which will be documented in the version history log.
