From Pilot to Platform: A Maturity Model for Scaling AI Agents Across the Enterprise
Most orgs are stuck with pilots that never become a program. A five-stage maturity model for scaling agents, and the trap of scaling before governance.
#deep-dive
23 articlesPrompt Patterns That Actually Work in Now Assist Skills (and the Ones That Quietly Fail)
Prompting a platform skill is a different craft from prompting a chatbot. The patterns that hold up, schema-anchored, grounded, single-purpose, and the anti-patterns that quietly fail.
Testing AI You Cannot Fully Predict: Adapting ATF and QA for Agentic Workflows
Generative AI breaks the assumption ATF is built on: same input, same output. Testing non-deterministic workflows with property checks, evaluation sets, and adversarial guardrail tests.
Governing AI to the EU AI Act and NIST Without Slowing Your Team Down
Compliance stalls agentic programs when it is a gate at the end instead of a property built in. Operationalizing NIST and EU AI Act controls with AI Control Tower, without slowing down.
RaptorDB and the Unsexy Truth About AI Performance at Scale
Agents change the shape of your database load from a human trickle to a machine flood. Why data-tier performance becomes a first-class constraint on scaling autonomy.
HRSD's AI Moment: From Policy Lookups to Autonomous Onboarding Orchestration
HR service delivery offers a clean maturity ladder from policy lookups to cross-department onboarding orchestration. Climb it in order, and skip a rung at your peril.
AI Voice Agents in CSM: What Changes Technically When the Customer Talks to the Machine
A voice agent is not a chatbot with a microphone. The three constraints that reshape the design, latency, no undo, and turn-taking, and why graceful escalation beats heroics.
The SecOps Wrap-Up Agent: AI Inside the Incident Response Loop
Security is where the audit bar is highest. How SecOps AI agents accelerate summarization, correlation, and case wrap-up while the analyst keeps the judgment.
Token Economics: Cutting the Cost and Latency of Your Now Assist Skills
At production scale, tokens are a line item and latency is felt on every call. Where the tokens go, matching the model to the task, and measuring cost per resolution, not per call.
Self-Healing IT: How Predictive AIOps Moved From Alerting You to Fixing It
AIOps used to stop at a smarter alarm. The technical chain from ingestion to anomaly detection to autonomous remediation, and the boundary discipline that makes self-healing safe.
Evaluating a Skill Like an Engineer: Building a Test Set and Measuring Quality
It looked good in the demo is not QA. Building a representative test set, scoring groundedness explicitly, and regression-testing skills after every change, no data-science team required.
Workflow Data Fabric: Why the Boring Data Layer Is the Real AI Unlock
Everyone talks about agents and nobody talks about the data plumbing that decides whether they are brilliant or useless. Why Workflow Data Fabric, not the agents, is the real moat.
Adaptive Thresholds Beat Static Alerts: The Quiet ML Win Most IT Teams Ignore
Most of your alerts are noise because the threshold is static. How learned baselines and anomaly detection cut the noise, and the tuning period that is the honest catch.
Building an MCP-Exposed Action Without Opening a Security Hole
The GA MCP Server lets external agents take governed actions in ServiceNow. Expose one narrow action, scope the identity, log everything, and avoid building a remotely-callable backdoor.
Instrumenting Agents for Observability: What to Log Before You Scale Past Ten
An agent you cannot observe is one you cannot debug, trust, or govern. What to log per agent run, why selection traces matter most, and connecting to AI Control Tower before you scale past ten.
Data Readiness for Agents: Making Your CMDB and Knowledge Base Safe to Ground On
Your agents are only as good as the data they ground on, and most CMDBs are part fiction. The unglamorous, highest-impact prep: accuracy, freshness, ACL hygiene, and structure for retrieval.
Guardrails as Code: Configuring AI Guardian and Approval Gates for Autonomous Agents
Guardrails are not a slide, they are configuration. Placing every action on the autonomy spectrum, enforcing prohibitions with AI Guardian, and gating the risky five percent.
Triggers, Not Chat: Event-Driven AI Agents with Flow Designer
The agents that move your metrics fire on platform events, not chat. Wiring event-driven agents in Flow Designer: precise triggers, tight context, deliberate branching, real error handling.
Designing Tools for Agents: The Difference Between a Tool an Agent Uses and One It Abuses
An agent is only as good as its tools, and a tool is judged by whether the agent calls it correctly. Descriptions as prompts, input contracts, idempotency, and testing tool selection.
Single Agent vs. Agent Team: When to Reach for the AI Agent Orchestrator
When one agent with a pile of tools starts picking wrong and looping, it is time to decompose. A field guide to the AI Agent Orchestrator, the patterns that work, and the costs nobody mentions.
Under the Hood of Similarity: How doc2vec Actually Powers ServiceNow Matching
Open up the similarity model: doc2vec turns messy record text into vectors that match meaning, not keywords. Why matches degrade, and how to tune with the word corpus.
RAG Inside a Custom Skill: Wiring AI Search Retrievers the Right Way
RAG on the Now Platform is a concrete component, a retriever wired into a custom skill. Where retrieval quality leaks, and how to test it separately from generation.
Predictive Intelligence or Now Assist? A Real Decision Framework for the Two AI Engines
Two distinct AI engines, and reaching for the LLM when a cheap classifier would win is a design error. A real decision framework for Predictive Intelligence versus Now Assist.