One of the most common mistakes in AI projects is choosing the technology before understanding the problem. Agentic AI is powerful, but it is not always the right answer, and using it where a standard workflow would do the job adds cost and complexity for no gain.
The decision tree
Question 1: Is the process rule-based and deterministic?
If the same input always produces the same output and the logic can be fully mapped upfront, use workflow automation.
- OOTB ServiceNow flows, business rules, and approvals handle this well
- Lowest cost, most predictable, easiest to audit
- Examples: ticket routing by category, approval chains, standard notifications
Common mistake: Building a Now Assist skill to summarise and route an incident when a simple flow with assignment rules would do it in milliseconds at zero AI cost.
Question 2: Does it need generative AI to augment human judgment?
If the output varies, language is involved, or a human still makes the final call, generative AI is the right layer.
- Now Assist covers this: summarisation, drafting, recommendations, search
- Examples: summarising an incident for a new agent, drafting a change request justification
Common mistake: Building an autonomous agent to draft and send customer communications when a Now Assist skill with human review before send would be faster, safer, and cheaper to run.
Question 3: Does it need autonomous multistep reasoning?
If the process involves variable steps, requires the system to plan and adapt based on intermediate results, and a human should not be in every decision loop, agentic AI earns its place.
- AI Agents and Now Assist Skills with orchestration
- Human oversight at key checkpoints, not every step
- Examples: autonomous triage and resolution of L1 incidents with escalation logic; supplier risk assessment pulling from multiple data sources
Common mistake: Scoping an agentic workflow for a process that has three fixed steps and never deviates. The agent adds latency, cost, and unpredictability to something that was working fine as a flow.
The rule
Start at question 1. Stop as soon as you have a yes. The first tool that fits is the right tool.
AI is not a hierarchy where agentic is better than Now Assist, which is better than workflow automation. They solve different problems. The best outcome is often a combination: a standard flow handles the deterministic parts, Now Assist handles the language layer, and an agent handles the steps that genuinely require autonomous reasoning.
What this looks like in practice
A common employee onboarding use case might break down like this:
- Workflow automation: create accounts, assign equipment, trigger tasks across departments
- Now Assist: draft the welcome email, summarise onboarding status for the manager
- Agentic AI: handle the exceptions autonomously, with escalation when needed
Three tools. One process. Each doing what it is actually built for.
Sylvain Hauser is an AI Architect at ServiceNow, based in Australia. He holds Certified Master Architect (CMA) and Certified Technical Architect (CTA) credentials with 18+ years of ServiceNow expertise.
