Practical AI with ServiceNow — from zero

1. Prerequisites — before you open Agent Studio

2. Direct URLs — bookmark these

Replace {BASE} with your instance (use the bar at top after page load). All links open in a new tab once applied.

3. Lab — one read-only agent (leave or incidents)

Pick one track. Both use the same pattern: Script Include tool → register in Agent Studio → data access = invoking user → GlideRecordSecure so ACLs apply as the human who asked.

3.1 Script Include — Track A (leave, read-only)

var LabAIReadTools = Class.create();
LabAIReadTools.prototype = {
    initialize: function() {
        this.T_REQ = 'x_786686_hr_leave_req_v2';
        this.T_EMP = 'x_786686_hr_leave_employee_v2';
    },

    _employeeSysIdForUser: function(userId) {
        var gr = new GlideRecordSecure(this.T_EMP);
        gr.addQuery('user', userId);
        gr.setLimit(1);
        gr.query();
        return gr.next() ? gr.getUniqueValue() : null;
    },

    /** AI Agent tool — list leave requests visible to invoking user via ACLs */
    listLeaveForUser: function() {
        var userId = gs.getUserID();
        var empId = this._employeeSysIdForUser(userId);
        if (!empId) return { ok: true, rows: [], note: 'no_employee_profile' };

        var out = [];
        var req = new GlideRecordSecure(this.T_REQ);
        req.addQuery('u_employee_ref', empId);
        req.orderByDesc('sys_created_on');
        req.setLimit(25);
        req.query();
        while (req.next()) {
            out.push({
                number: req.getValue('number'),
                state: req.getDisplayValue('u_state'),
                type: req.getDisplayValue('u_leave_type'),
                start: req.getValue('u_start_date'),
                end: req.getValue('u_end_date'),
                days: req.getValue('u_days')
            });
        }
        return { ok: true, count: out.length, rows: out };
    },

    type: 'LabAIReadTools'
};

3.2 Script Include — Track B (incidents, read-only)

var LabAIReadTools = Class.create();
LabAIReadTools.prototype = {
    initialize: function() {},

    /** AI Agent tool — open incidents the invoking user may read (ACL-enforced) */
    listOpenIncidents: function(limit) {
        var max = Math.min(parseInt(limit, 10) || 10, 25);
        var out = [];
        var inc = new GlideRecordSecure('incident');
        inc.addActiveQuery();
        inc.orderByDesc('opened_at');
        inc.setLimit(max);
        inc.query();
        while (inc.next()) {
            out.push({
                number: inc.getValue('number'),
                short_description: inc.getValue('short_description'),
                state: inc.getDisplayValue('state'),
                priority: inc.getDisplayValue('priority'),
                assignment_group: inc.getDisplayValue('assignment_group')
            });
        }
        return { ok: true, count: out.length, rows: out };
    },

    getIncidentByNumber: function(number) {
        if (!number) return { ok: false, error: 'number_required' };
        var inc = new GlideRecordSecure('incident');
        inc.addQuery('number', number);
        inc.setLimit(1);
        inc.query();
        if (!inc.next()) return { ok: false, error: 'not_found_or_denied' };
        return {
            ok: true,
            number: inc.getValue('number'),
            description: inc.getValue('description'),
            state: inc.getDisplayValue('state'),
            opened_at: inc.getValue('opened_at')
        };
    },

    type: 'LabAIReadTools'
};

4. Agentic workflow vs single agent

Start with a single agent. Promote to agentic workflow when you can draw the steps on a whiteboard and name which agent owns each step.

5. RAG — practical steps (high level)

Retrieval Augmented Generation = give the model relevant text at question time; do not rely on training memory. On ServiceNow you usually implement retrieval first, generation second.

Deep implementation (three patterns A/B/C): AI Full Demo Master — Part 7 (RAG).

6. MCP — practical pointer

Model Context Protocol connects AI clients (Cursor, Claude Desktop, agent frameworks) to tools through a standard interface. For ServiceNow labs, the security boundary stays on the instance: allow-listed Script Includes or Scripted REST, not ad-hoc SQL from a laptop.

7. Define data access — invoking user vs AI User

8. Best practices — do / don’t

9. ServiceNow SDK — what it CAN and CANNOT do (AI agents)

From SDK 4.4+, Fluent APIs include AiAgent, agent tools, triggers, and agentic workflow artefacts — same governance tables as UI (sn_aia_*), expressed as TypeScript in your repo.

// Conceptual Fluent shape — see building-ai-agents-guide in SDK docs
import { AiAgent } from '@servicenow/sdk/core'

export const leaveReadAgent = AiAgent({
  name: 'Leave Read Assistant (lab)',
  description: 'Read-only leave visibility for invoking user',
  // tools, instructions, triggers: defined alongside Script Include APIs in repo
})

Full SDK + MCP pairing: Full Demo Part 10 · Local SDK walkthrough: 16 SDK from scratch.

10. Related guides — go deeper