Game On: A Playbook for Choosing the Right Agentic AI Use Cases - Knowledge 2026 Session Recap

[Mrini]

Enterprise teams consistently struggle with the same problem: too many AI ideas, no structured way to evaluate them, and no clear path from idea to implementation. The usual outputs — long use case lists, gut-feel prioritization, endless debates — rarely lead to action.

This session set out to change that with three concrete goals:

01  Understand Agentic AI Get a clear, working definition of what Agentic AI is — and what makes it meaningfully different from Generative and Autonomous AI.

02  Identify & Prioritize Learn a practical, repeatable framework for spotting high-value Agentic AI opportunities and sequencing them by readiness and impact.

03  Apply the learnings Walk away with tools you can use on Monday — in a process review, a backlog session, or a conversation with leadership.

What is Agentic AI?

Agentic AI refers to AI systems that can autonomously plan, reason, and execute multi-step tasks to achieve a defined goal — with minimal human intervention per step.

The key distinction is autonomy over a chain of actions — not just the ability to generate a response.

Type	Badge	What it does	Human in the loop
💬 Generative AI	RESPONDS	Generates content in response to prompts	HIGH — human must act on output
⚙️ Agentic AI	EXECUTES	Plans and executes multi-step workflows using tools and memory	MEDIUM — oversight at key checkpoints
🚀 Autonomous AI	SELF-DIRECTS	Operates independently across extended periods with minimal oversight	LOW — minimal human involvement

How to Identify an Agentic AI Use Case

Look for processes that hit most of these five signals:

• High frequency, repeatable work
• Multi-step across systems
• Heavy context or data synthesis required
• Clear action AI can take — not just suggest
• Human oversight points are identifiable

If a use case hits two or three, it is worth developing. If it only hits one, it is more likely a feature enhancement than an agent.

How to Prioritize: Now, Next, or Later

Once you have a shortlist, sequence by readiness and impact using this three-bucket framework:

NOW — Deploy immediately •      High value & high trust •      Clear business impact •      Feasible with current tools •      Human oversight defined

NEXT — Near-term with prep •      High potential, needs validation •      Strong value, moderate risk •      Requires guardrails or pilots •      Policy checks needed

LATER — Future exploration •      Low confidence or high risk •      Trust or compliance gaps •      Low ROI or unclear ownership •      Better solved without AI

Pro tip: Revisit your 'Later' and 'Never' categories periodically. Organizational readiness and technology capabilities shift faster than you expect.

Key Takeaways

The core shift: stop asking "Where can we use AI?" and start asking "Where in my workflow is a human only there to check, route, or notify?" That is the signal.

1	🎯  Spot an Agentic AI opportunity 3+ sequential steps + system check + human handoff = agent candidate. This pattern repeats in IT routing, HR onboarding, legal review, and ops dispatch.

2	⚙️  Separate augmentation from agentic Augmentation = AI helps a human. Agentic = AI runs the task end-to-end and only escalates when genuinely stuck. Smarter autocomplete is a feature, not an agent.

3	🔍  Use the 3 unlocking questions daily ① Where does someone manually hand something off?  ② Where does work queue for a human?  ③ Where do things fall through the cracks between teams?  Two of three = a use case worth pitching.

4	📊  Score before you build Business value + feasibility = your two-axis filter. High on both: 2-week discovery sprint. High value, low feasibility: research spike. Low on both: park it.

5	🗺️  Map airport stations to your workflow Security = compliance gate. Lounge = approval bottleneck. Gate = routing logic. Boarding Call = notification cascade. Takeoff = closure. Where does YOUR work stall?

How to Apply This Next Week

🔄  In a process review

Walk one workflow you own through the 8 airport stations. For each: is this step currently manual? If yes → candidate. If it also checks a system or notifies multiple people → shortlist.

📝  Writing a use case brief

Use this 3-part structure: Pain point (what's the manual step + its cost) → Agent design (tools, decisions, escalation rules) → Value case (time saved, errors reduced, experience improved).

📋  Prioritizing a backlog

Plot candidates on value vs feasibility. High on both → 2-week discovery sprint. High value, low feasibility → research spike. Low on both → park it. Stop debating, start plotting.

📣  Socializing upward

Lead with the pain, not the tech. 'Our L2 engineers spend 40% of their time routing tickets that should never reach them' lands better than 'we want to build an AI agent.' Pain first. Agent second.

Resource: Airport to Your World — Field Guide

When you land on a station, ask: "Where in MY work does this same pattern exist?"

Airport Station	Pattern	IT / ITSM	HR	Legal	Operations	Question to Ask
🎫 Booking Confirmation	Request intake & validation	IT ticket logged, category verified	Job req submitted, headcount approved	Contract request logged, party checked	Work order raised, scope defined	Where does someone manually validate a request before it moves forward?
🧳 Check-in & Baggage Drop	Intake processing & triage	Ticket enriched, priority assigned	Candidate screened, docs collected	Matter scoped, conflicts checked	Order kitted, materials staged	Where does work pile up waiting to be sorted and assigned?
🛂 Security Screening	Compliance & access control	Change approval, access validation	Background check, right-to-work	Regulatory review, risk flagged	Quality inspection, safety check	Where does every request need a human to check compliance?
🏛 Lounge / Hold Area	Waiting on dependency	Awaiting approvals, blocked by vendor	Offer pending, counter-sign out	External counsel review, SLA wait	Parts on backorder, upstream delay	Where does work stall waiting on another team or approval?
🚪 Gate Assignment	Routing to right resource	Ticket assigned to right team/tier	Role matched to hiring manager	Matter routed to specialist	Task dispatched to right crew	Where does work go to the wrong person or team first?
📢 Boarding Call	Notification & mobilization	Alert fired, SLA breach warning	Interview invite, confirmations sent	Deadline reminder, stakeholder nudge	Crew notified, shift briefed	Where do people chase others manually for updates?
✈ Takeoff / Departure	Resolution & closure	Incident resolved, change deployed	Offer accepted, onboarding starts	Contract signed, matter closed	Work order done, delivery confirmed	Where does closing a task trigger a dozen manual next steps?
🚨 Delay / Crisis Card	Exception handling	P1 incident, rollback triggered	Offer rescinded, urgent backfill	Injunction, emergency redline	Outage, supply chain disruption	Where do exceptions derail an otherwise smooth process?

3 UNLOCKING QUESTIONS:  ①  Where does someone manually hand something off?   ②  Where does work sit in a queue waiting for a human?   ③  Where do things fall through the cracks between teams?

Session Resources

Download the materials from this session to use in your own team workshops:

🖨️	Printable Reference Card A4 landscape handout with the full field guide and the 3 unlocking questions — leave it on the table during any AI planning session

"Find AI use cases by looking for places where a human is only there to check, route, or notify — then ask whether an AI agent could do that faster, more consistently, and at scale."

About the authors

Mrini Gorla is a Sr. Staff UX Researcher at ServiceNow focused on Agentic AI, ITSM, and DEX.  mrini.gorla@servicenow.com

Aditya Dabral is a Staff UX Researcher at ServiceNow.  aditya.dabral@servicenow.com

Questions or want to run this with your team? Drop a comment in the ServiceNow Community or reach out directly.