Agents that work inside the instance, not beside it. We design, build and govern AI agents on Now Assist and AI Agent Studio, wired to your ITSM, HR and customer workflows through Flow Designer and IntegrationHub, and handed over as a scoped app with ATF tests your platform team can sign off.
From Pilot to Production: ServiceNow AI Agents Across Your Instance
Six agent types covering the queues that actually consume your service desk. Each one is built in a scoped application, reads only what you granted it, and writes back to the record so the work is auditable.
Agent Types
01 · Agent type
ITSM Incident & Request Agents
Categorise, prioritise and assign on arrival
Resolve password, access and how-to tickets unattended
Pull CI, user and entitlement context from the CMDB
Raise the change or catalogue task the fix needs
Escalate to the right assignment group with the work shown
What Do You Actually Get From a ServiceNow AI Agent?
A chatbot answers. An agent finishes. Inside ServiceNow that difference shows up on the record: state changed, task created, approval requested, work notes written. Four things that change in the first quarter.
Queues that stop growing
The tickets nobody wants are the ones an agent is best at: password resets, access requests, the same how-to question asked forty times a week. They close on arrival instead of ageing in a queue.
Tier-one tickets resolved unattended
Overnight queues cleared before the shift starts
People see exceptions, not the whole pile
Resolution time that actually moves
Most of an incident's life is waiting: for triage, for context, for the right group. An agent does that reading and routing the moment the record is created.
Categorised and assigned on arrival
CMDB and entitlement context already attached
Fewer reassignments between groups
Capacity without more headcount
Demand rises with the business. Headcount rises with the budget cycle. An agent absorbs the gap and gives you the volume numbers to prove it.
Peak and seasonal load absorbed
Cover outside business hours
Deflection measured per workflow, not guessed
Work your platform team can govern
Everything the agent does lands on the record, inside your ACLs, under a service account you control. That is the difference between a pilot and something audit will accept.
Every prompt and tool call traced
Approvals on anything consequential
Scoped app, update sets, ATF tests
ServiceNow Agents That Close the Record
A demo that answers a question is not the same as an agent that finishes a ticket. These six disciplines are how we get from one to the other, backed by our AI development and ServiceNow engineering teams. Need conversational delivery too? See AI chatbot development.
[1]Now Assist Skills & Prompting
Skills built in Skill Kit against your own records and knowledge, with prompts held in source control and a test set that says whether a change made them better or worse.
[2]AI Agent Studio Orchestration
One agent rarely covers a process. Studio lets several specialise, hand work to each other and stop at the same approval gates, with the orchestrator deciding which one owns the next step.
[3]Grounded Retrieval Over Knowledge & CMDB
Answers pulled from your knowledge base, catalogue items, CI records and past resolutions, cited back to the article or record, so the agent cannot invent a policy that was never published.
[4]Flow Designer & IntegrationHub Actions
The agent acts through the same actions and spokes your platform team already maintains, typed and validated, rather than through a side door nobody can see in the flow history.
[5]Virtual Agent & Channel Delivery
Portal, Teams, Slack, email and the mobile app, all reaching the same agent, with a clean handover to a person that carries the conversation and the record with it.
[6]Evaluation, ACLs & Guardrails
A held-back set of real tickets scores every release, ACLs bound what the agent can see, and anything consequential waits for a named approver before it happens.
Technologies
The ServiceNow AI Stack We Build On
What separates an agent that survives a real service desk from one that impresses in a demo. Pick one to see where it earns its place in a build.
Now Assist & Skill Kit
Out-of-the-box skills where they fit, custom skills in Skill Kit where they do not. Prompts live in the scoped app alongside the code, so a change is reviewable and reversible rather than something someone edited in a form.
AI Agent Studio & Orchestrator
Agents defined with an explicit goal, a tool list and a stopping condition, and an orchestrator that decides which one takes the next step. Multi-agent only where a single agent genuinely cannot carry the process.
Model Choice
Now LLM where the licensing works, Azure OpenAI, OpenAI, Anthropic or an open-weight model in your own tenancy where residency or cost says otherwise. We benchmark on your own tickets before anyone commits.
Predictive Intelligence
Classification, similarity and clustering trained on your historical records. Often cheaper and more accurate than a language model for pure routing, and we will say so when that is the case.
Flow Designer & IntegrationHub
Every action the agent can take is a flow action or a spoke, so it appears in the flow execution history and your team can read exactly what ran, in the tooling they already use.
AI Search & Knowledge
Retrieval tuned over knowledge bases, catalogue items and resolved records, with the gaps in your knowledge reported back so the agent gets better as the content does.
ACLs, Domains & Residency
A dedicated service account with the narrowest roles that still work, domain separation respected, and instance data kept where your contract says it has to stay.
Not Sure Which Queue to Automate? Get a Free Agent Audit.
One call about your busiest queues, your modules and what your platform team will allow tells you which workflow is worth automating first, what it will cost to build, and where a ServiceNow agent is the wrong answer.
Nine sectors where agents are already taking work off a ServiceNow queue. See our fintech and enterprise software practices for the systems either side of the platform.
[01]Manufacturing
Plants where a stopped line is a cost per minute and the service desk sits between IT and the floor.
Delivered since 2018 across AI, mobile, cloud and enterprise builds.
2-3 wk
Pilot to live
Typical time from a signed scope to an agent running in your sub-production instance.
CSA/CIS
Certified engineers
ServiceNow certifications held across ITSM, ITOM, CSM and HRSD, maintained on current releases.
15+
Countries served
Client operations running on Appther-built platforms worldwide.
Case Studies
Agent Work We've Shipped
Agents from our wider AI practice rather than ServiceNow builds, shown because the engineering is the same: a voice agent answering every patient call, a WhatsApp assistant carrying frontline support, and a routing engine trained on live fleet data.
Plenty of firms can demo Now Assist. Fewer will run your own ticket history through an evaluation set in week one and tell you plainly which queue is worth automating and which is not. We cover the platform work and the model work in one team, alongside our cloud infrastructure practice.
01
Platform-native, not bolted on
The agent lives in a scoped application, acts through Flow Designer actions and IntegrationHub spokes, and writes to the record. Nothing runs in a shadow system your platform team cannot see, and nothing lands in global. Backed by our ServiceNow engineers.
02
Certified engineers who have survived upgrades
CSA as a baseline and Certified Implementation Specialist credentials across ITSM, ITOM, CSM and HRSD, maintained on current releases. ATF tests ship with the build so the next family upgrade is a test run, not a gamble.
03
Evaluated, not demoed
Every agent ships with a held-back set of your own tickets and a baseline to beat. If a routing rule or Predictive Intelligence already does the job, we will tell you before you pay for a language model.
04
You own the scoped app
Source, prompts, skills, flows, test suites and documentation are yours, in your instance, from the first update set. There is nothing to unwind if you decide to carry on without us.
Security & Compliance
How Does Appther Keep a ServiceNow Agent Safe?
An agent that can change a record is an agent that can change the wrong record. These are the controls we build in, and only what we can honestly claim.
Prompts and traces carry personal data more often than teams expect. We design around minimisation first, so compliance is a property of the architecture rather than a later retrofit.
GDPR & UK GDPRLawful basis mapped per record type, subject access answered from the trace table rather than from memory.
Data residencyInstance data stays in the region your ServiceNow contract specifies, and any external model call honours the same boundary.
MinimisationThe agent is sent the fields it needs to decide, not the whole record, which also keeps token cost down.
RetentionPrompts, responses and traces age out on a schedule you set, not whenever someone remembers to clear them.
An agent is only as safe as its loosest tool. Identity, permissions and action limits are treated as one system, not three separate features.
Least-privilege service accountA dedicated account with the narrowest roles that still let the agent work, reviewed with your platform team before go-live.
Domain separationWhere you run domain separation, the agent respects it, so a managed-service tenant can never see another tenant's records.
Prompt-injection defenceTicket text and attachments are treated as data, never as instructions, and every tool call is validated against a schema.
Secrets and keysCredentials live in the credential store and the connection alias, never in a prompt, a script include or a property.
The controls below describe how we build and review code. They are engineering practices we apply on every project, not certifications we hold.
Scoped applicationEverything ships inside a scoped app with its own roles and tables, so nothing is written to global by accident.
ATF coverageAutomated Test Framework tests cover the flows the agent triggers, and they run before every promotion and upgrade.
Peer reviewNo update set is promoted without a second engineer reading it, and all code is held in source control alongside the prompts.
Instance scanInstance Scan rules run against the app so it stays within your own platform standards, not just ours.
To be explicit: Appther is not a certification body, and we are not claiming a ServiceNow partner tier. Our engineers hold ServiceNow certifications; formal ISO or SOC 2 attestation of your instance sits with your compliance team and its auditor.
Regulated sectors change the architecture, not just the paperwork. We scope these constraints during discovery so they never surface late.
Approval gatesAnything that spends money, emails a customer or closes a record waits for a named approver, recorded on the task.
Change managementChanges travel by update set or your existing pipeline, on your release calendar, reviewed by your own CAB.
Audit trailEvery prompt, tool call, result and decision is written to a trace table your auditors can query directly.
Model transparencyWhere a decision affects a person, the inputs and the model's reasoning are stored with the record so it can be explained.
Agent interfaces fail review for predictable reasons. We build against the rules that trip up AI features specifically.
Disclosure of AI usePeople are told when they are dealing with an agent, on every channel, before it does anything for them.
Human escalationEvery agent has a documented route to a person, and it takes that route when confidence drops rather than guessing.
WCAG 2.2 AAPortal and Virtual Agent surfaces are tested for contrast, focus order and screen-reader behaviour.
Failure statesWhen the agent is unsure it says so, shows its sources, and leaves the record for a human rather than closing it.
Tech Stack
Platforms, Models & Integrations
The stack we build ServiceNow agents on, from the Now Platform capabilities themselves through to the models, spokes and channels either side of them, with performance work once the agent is carrying real volume.
From the skill that answers to the flow that acts, grouped by the layer it belongs to.
Now Platform
Now Assist
AI Agent Studio
AI Agent Orchestrator
Virtual Agent
Now Assist Skill Kit
Predictive Intelligence
AI Search
Flow Designer
Now Platform
Now Assist
AI Agent Studio
AI Agent Orchestrator
Virtual Agent
Now Assist Skill Kit
Predictive Intelligence
AI Search
Flow Designer
OpenAI
Anthropic Claude
Azure OpenAI
Google Vertex AI
AWS Bedrock
Meta Llama
Mistral
Hugging Face
Google Gemini
OpenAI
Anthropic Claude
Azure OpenAI
Google Vertex AI
AWS Bedrock
Meta Llama
Mistral
Hugging Face
Google Gemini
IntegrationHub
Slack
Jira
Salesforce
Microsoft 365
SQL sources
MongoDB
REST & SOAP
MID Server
IntegrationHub
Slack
Jira
Salesforce
Microsoft 365
SQL sources
MongoDB
REST & SOAP
MID Server
Python
Node.js
React
Docker
Terraform
Kubernetes
PostgreSQL
AWS
Python
Node.js
React
Docker
Terraform
Kubernetes
PostgreSQL
AWS
Feature Set
Key Features We Deliver in ServiceNow Agent Builds
The building blocks that show up in most ServiceNow agent builds, from the first skill through to the trace table your auditors read.
Now Assist Skill Kit Skills
Multi-Step Task Planning
Flow Designer Action Calling
Retry & Rollback on Failure
Grounded Retrieval with Citations
Run Trace & Audit Table
PII Redaction Before the Model
Scoped Service Account per Agent
Virtual Agent, Teams, Slack & Email
IntegrationHub Spokes & REST
ACLs & Role-Based Access
Approval Gates & Handover Rules
Prompt & Skill Versioning
Attachment & Document Understanding
Deflection & Cost Analytics
ATF Test Coverage
Architecture
ServiceNow Agent Architecture, Layer by Layer
A production agent is ten layers that have to agree with each other. Get one wrong and the symptom appears somewhere else, usually as a ticket closed that should not have been, which is why we design them together.
Channels
Where the request arrives: the service portal, Virtual Agent, Teams, Slack, email or the mobile app. All of them reach the same agent.
01
Intent & Routing
Predictive Intelligence or a light model decides what this is and which agent owns it, before any expensive reasoning happens.
02
Agent Orchestrator
AI Agent Studio decides which agent takes the next step, when to hand over and when the goal has actually been met.
03
Skills & Planner
The agent breaks the goal into ordered steps and picks the skill for each one, re-planning when a step comes back empty.
04
Tool Layer
Flow Designer actions and IntegrationHub spokes, typed and validated, so every call the agent makes shows up in the flow history.
05
Knowledge & CMDB
Retrieval across knowledge articles, catalogue items, CI records and resolved tickets, cited back to the source record.
06
Guardrails
ACLs, allow-listed actions, redaction before anything leaves the instance, and refusal handling when a request is out of scope.
07
Human Review
Approval gates on consequential actions and clean assignment to a group, with the agent's reasoning attached to the task.
08
Trace & Eval
Every prompt, tool call and result written to a trace table, scored against a held-back set of real tickets on each release.
09
Model Layer
Now LLM, Azure OpenAI, Anthropic or an open-weight model in your own tenancy, chosen per skill on cost, latency and accuracy.
10
Channels
Where the request arrives: the service portal, Virtual Agent, Teams, Slack, email or the mobile app. All of them reach the same agent.
01
Intent & Routing
Predictive Intelligence or a light model decides what this is and which agent owns it, before any expensive reasoning happens.
02
Agent Orchestrator
AI Agent Studio decides which agent takes the next step, when to hand over and when the goal has actually been met.
03
Skills & Planner
The agent breaks the goal into ordered steps and picks the skill for each one, re-planning when a step comes back empty.
04
Tool Layer
Flow Designer actions and IntegrationHub spokes, typed and validated, so every call the agent makes shows up in the flow history.
05
Knowledge & CMDB
Retrieval across knowledge articles, catalogue items, CI records and resolved tickets, cited back to the source record.
06
Guardrails
ACLs, allow-listed actions, redaction before anything leaves the instance, and refusal handling when a request is out of scope.
07
Human Review
Approval gates on consequential actions and clean assignment to a group, with the agent's reasoning attached to the task.
08
Trace & Eval
Every prompt, tool call and result written to a trace table, scored against a held-back set of real tickets on each release.
09
Model Layer
Now LLM, Azure OpenAI, Anthropic or an open-weight model in your own tenancy, chosen per skill on cost, latency and accuracy.
10
Protocols we build on
Now AssistSkill KitAI Agent StudioPredictive IntelligenceRAGFlow actionsIntegrationHubATF
Which layer does the work is decided by your modules, your licensing and what your platform team will govern, not by what is newest. See our AI development services for the model work either side of the platform.
Our Process
End-to-End ServiceNow Agent Delivery
Thirteen stages take an agent from the first conversation to something your platform team runs without us. Agent work punishes guesswork, so the risky parts are proven against your own tickets early and every stage ends in something you can inspect.
Week 1
Discovery
Which queue, whose group, and what “working” has to mean before anyone builds.
1
What we doSit with the people working the queue, read a month of real tickets, and map the task end to end: what arrives, what gets looked up, what gets decided.
What you getA written problem statement, the constraint list from your platform team, and the metric the agent will be judged on.
Week 1–2
Business Case
What the agent has to be worth before it is worth building.
2
What we doModel the hours the queue consumes today, size the licensing and build cost, and agree the deflection number we will be measured against.
What you getA business case with the arithmetic shown, and a scope you can hold us to.
Week 2–3
Task Mapping
Writing down what a good resolution actually looks like.
3
What we doTurn the process into explicit steps, decisions and stopping conditions, and mark which steps a person must still approve.
What you getA task map your service owner recognises, with the approval points already agreed.
Week 3–8
Instance & Data Access
What the agent may read, and under whose account.
4
What we doAgree the service account, the roles, the tables and the knowledge bases, and check domain separation and residency constraints.
What you getA least-privilege account, documented, reviewed by your platform team before a line of code is written.
Week 3–6
Skill & Model Selection
Choosing the cheapest thing that clears the bar.
5
What we doBenchmark Predictive Intelligence, out-of-the-box Now Assist skills and custom skills on your own ticket history, with cost and latency per call.
What you getA documented choice with the numbers behind it, including where a rule beats a model.
Week 5–12
Agent Build
The agent itself, inside a scoped application.
6
What we doBuild skills in Skill Kit, define goals and tools in AI Agent Studio, and hold prompts in source control alongside the app.
What you getA working agent in your sub-production instance, in a scoped app with its own roles and tables.
Week 7–13
Flow & Integration Layer
The actions the agent is allowed to take.
7
What we doWire every action through Flow Designer and IntegrationHub, typed and validated, with credentials in the credential store.
What you getActions that appear in the flow execution history, so your team can read exactly what ran.
Week 9–16
Channels & Handover
Where people meet the agent, and how they get past it.
8
What we doDeliver on portal, Virtual Agent, Teams, Slack or email, and design the handover so the conversation and the record travel together.
What you getChannels live, and an escalation path a person can take without repeating themselves.
Week 10–17
Trace & Ops Console
Somewhere to look when it does something odd.
9
What we doWrite every prompt, tool call and result to a trace table, with a dashboard for volume, deflection, cost and escalation rate.
What you getAn ops view your service owner reads weekly and your auditors can query directly.
Week 12–19
Guardrails & ATF
Proving it behaves before anyone depends on it.
10
What we doApproval gates, redaction, refusal handling and prompt-injection tests, plus Automated Test Framework coverage on every flow the agent triggers.
What you getA documented guardrail set and a test suite that runs before every promotion.
Week 14–21
Shadow Running
Watching it work without letting it act.
11
What we doThe agent runs alongside your team for a fortnight, recording what it would have done, with no write access to the record.
What you getA side-by-side comparison against what your people actually did, and a go or no-go you can defend.
Week 20–22
Phased Rollout
Turning it on one group at a time.
12
What we doPromote by update set on your release calendar, start with one assignment group, and widen only once the numbers hold.
What you getA live agent, rolled out at your pace, with a rollback that is one update set away.
Ongoing
Support & Tuning
The part that lasts longer than the build.
13
What we doRe-run the evaluation set as your knowledge and ticket mix change, and test on a clone before every family upgrade.
What you getQuality that holds through upgrades, and a monthly report your service owner can take to the business.
Pricing
Fixed Price, Agreed Before We Start
Round numbers so you can budget today. The exact figure is confirmed in writing after scoping and does not move unless the scope does. ServiceNow licensing is separate and billed by ServiceNow, not by us.
Agent Pilot
Prove it works on one queue before committing to a programme. One agent, one workflow, running in your sub-production instance.
All tiers include the scoped application source, prompts, skills, flows, ATF tests and documentation, handed over in your instance. ServiceNow licensing for Now Assist and AI Agent Studio is purchased from ServiceNow directly; we will tell you when a custom scoped app works out cheaper for your volume.
Client Voices
Trusted By Teams Who Ship
“Appther built our student competition app exactly the way we envisioned it: easy registration, smooth judging, and real-time results that keep participants engaged.”
Dionna MilemOwner · My BOB Team LLC
“Building a mental health app means every detail has to be handled with care, and Appther got that from day one. The product feels warm, simple, and genuinely helpful.”
Bob SinghalProfessor of Joy · Joyscore Inc
“Appther exceeded every expectation, on time and on budget, and the final product has been instrumental in scaling our business. A rare engineering partner.”
Johnathan PefferCEO · Bull's Eye Technologies Inc
“Implementing an ERP across our operations felt daunting at first, but Appther made the transition seamless. Every department now works off a single source of truth.”
Pushpender KumarGeneral Manager · A.B. Sugars Ltd
Latest Insights
Notes on Building Enterprise AI Agents
Field notes from real agent builds. No hype, just what worked and what did not.
Where agents take work off an enterprise queue, and where a person should still approve.
Free 30-Minute Agent Strategy Call
A ServiceNow Agent Starts With One Queue. Let's Pick Yours.
Walk away from one call with the workflow worth automating first, a realistic build cost, an honest read on the licensing, and a delivery timeline. No pitch deck, no obligation.
Virtual Agent holds a conversation and can run a topic flow you scripted in advance. An AI agent reasons about a goal, decides which steps it needs, and calls tools to finish the work: reading the CMDB, updating an incident, requesting an approval, then writing back what it did. Virtual Agent is often the front door; the agent is what happens behind it.
For agents built on Now Assist and AI Agent Studio, yes, those are licensed ServiceNow capabilities and you buy them from ServiceNow, not from us. If the licensing does not stack up for your use case, we can build the same outcome with a custom scoped app calling your own model endpoint. We will tell you honestly which of the two is cheaper for your volume.
A single-workflow pilot is $6,900 and goes live in two to three weeks. A production agent across a full process is $14,900. A multi-agent platform spanning several modules starts at $29,900. Those are our build fees, confirmed in writing after scoping; ServiceNow licensing is separate and billed by ServiceNow.
We work in your sub-production instance, in a scoped application so nothing leaks into global. Changes travel by update set or via your existing CI pipeline, with Automated Test Framework tests attached, so your platform team reviews and promotes on their own schedule. We never push to production ourselves unless you ask us to.
Scoped apps and ATF tests are the reason we build this way. Before a family upgrade we run the test suite on a cloned instance, fix what the release changed, and hand you the results. Prompts and skills are versioned alongside the code, so you can see exactly what changed and why.
Three layers. ACLs and a dedicated service account mean the agent can only see and touch what you granted it. Every consequential action, anything that spends money, emails a customer or closes a record, waits behind an approval. And every prompt, tool call and result is written to a trace table you can query, so an auditor gets an answer rather than a shrug.
Most often ITSM and HRSD, then CSM, ITOM and Security Operations. The pattern is the same wherever a queue exists: something arrives, a person reads it, classifies it, looks something up and acts. If your process lives in a custom scoped app rather than a ServiceNow module, that works too.
That is what we recommend. Pick the noisiest queue, give the agent one workflow and read access to what it needs, and run it in shadow mode alongside your team for a fortnight. You compare what it would have done against what your people did, and only then decide whether to let it act.
A ServiceNow Engineer Replies Within One Business Day
Let's Build Your ServiceNow Agent
Tell us which queue is loudest and which modules you run. No sales pitch, just a straight answer on approach, scope, licensing and cost. Prefer email? Use our contact page.
Free 30-min scoping callAn honest read on which workflow to automate first.
Fixed-price build quoteDetailed scope and price after discovery, with licensing called out separately.
NDA on requestYour instance, your data and your roadmap stay yours.
