What a human resources AI agent actually does
Answers an employee's question about PTO balance, benefits or company policy directly in Slack or Teams, pulling from the handbook and HRIS instead of making them file a ticket.
A human resources AI agent sits in Slack, Teams or an intranet chat and answers the questions that otherwise clog an HR inbox: how many vacation days do I have left, when does open enrollment end, what's the policy on parental leave. It reads from the employee handbook and, for account-specific answers, calls a read-only tool against the HRIS to check an actual PTO balance or enrollment status. The better tools route anything that isn't a documented policy question, a grievance, a request for an accommodation, a question with legal exposure, straight to a human HR generalist instead of guessing.
This category consolidated around a few large, expensive platforms through 2025-2026 rather than staying open to smaller players. Workday relaunched its AI as Sana from Workday in March 2026 after roughly $3B in AI-related acquisitions (Sana, Paradox, HiredScore, Evisort), and ServiceNow completed its acquisition of Moveworks on December 15, 2025 for a final consideration of about $2.4B, folding Moveworks' employee-support agent into ServiceNow's existing roadmap. Both moves pushed HR agent capability further into suites built for companies already running enterprise HRIS platforms.
Where these agents still fail: a policy question that sounds routine but actually touches a protected leave category (FMLA, ADA accommodation) where a wrong answer carries real legal risk, and any question where the real answer is it depends, talk to HR rather than a clean policy lookup. An agent that confidently answers an accommodation or disciplinary question from general policy text, instead of recognizing it needs a human with full context, is the single biggest failure mode in this category.
AI capabilities involved
Answering a policy question by retrieving from the employee handbook
Looking up an employee's own PTO or benefits status through an HRIS tool call
Recognizing when a question needs to escalate to a human HR generalist
Guiding a new hire through an onboarding checklist
Who uses this
- HR teams at companies of roughly 100-2,000 employees fielding repetitive policy questions
- People operations leads trying to free up generalist time from ticket triage
- Companies with a distributed or remote workforce needing 24/7 answers to basic HR questions
- Mid-market employers who already run an HRIS but find its native help features too thin
What the HR agent handles (and what it shouldn't)
| Task | Agent handles it? | Systems it touches | Human role |
|---|---|---|---|
| Answering a PTO balance or how-to-request-leave question | Fully | HRIS, employee handbook | None |
| Explaining benefits enrollment options and deadlines | Fully | Benefits platform, handbook | None |
| Guiding a new hire through an onboarding checklist | Fully | HRIS, onboarding tool | Checks in personally during the first week |
| A simple HRIS update, like an address or emergency contact change | With approval | HRIS | Confirms the change before it's finalized |
| Explaining a standard company policy from the handbook | Fully | Employee handbook | None |
| A formal complaint, grievance or harassment report | No | None | Takes the report over personally and starts the HR process |
| A leave request that may touch FMLA or an ADA accommodation | No | HRIS, legal/compliance | Handles the determination personally, with legal input if needed |
| Discipline, termination or performance-related questions | No | HRIS | Handles personally |
Off-the-shelf AI agents on the market
Real products you can sign up for today — with current 2026 pricing, honest pros and cons.
Leena AI
Enterprises with 1,000+ employees that need HR, IT and Finance support unified in one assistant
Demo only
Contact sales; reportedly around $150/employee/yr with a 1,000-seat minimum
Custom enterprise contracts
Pros
- +Agentic assistant spans HR, IT and Finance questions in one Teams or Slack interface
- +Connects to a wide range of HRIS and ITSM systems through pre-built integrations
- +Purpose-built for high-volume employee self-service, not retrofitted from a general chatbot
Cons
- −The reported 1,000-seat minimum puts the effective floor around $150K a year regardless of actual usage
- −The vendor's own pricing page does not publicly resolve, so every number requires third-party confirmation
Sana from Workday
Companies already running Workday HCM as their system of record
Not available
Sana Enterprise reportedly from $30/user/mo plus usage-based Flex Credits
A reported $45K-$65K AI transformation fee on top of per-user costs; Workday publishes no list price
Pros
- +The Self-Service Agent sits directly on top of your actual Workday HCM data, not a separate copy
- +Agent System of Record extends governance to third-party agents too, not just Workday's own
- +Backed by roughly $3B in AI-focused acquisitions (Sana, Paradox, HiredScore, Evisort) consolidated into one roadmap
Cons
- −Requires an existing Workday HCM tenant, so it isn't a standalone purchase
- −Pricing is entirely opaque until a sales conversation, including a reported five-figure setup fee
Moveworks (now part of ServiceNow)
Enterprises already evaluating or running ServiceNow who want employee support folded into the same platform
Not available
Contact sales
ServiceNow completed its roughly $2.4B acquisition of Moveworks on December 15, 2025; pricing now follows ServiceNow's enterprise contract model
Pros
- +Employee-support agent spans HR, IT and Finance questions with over 100 pre-built connectors
- +Strong enterprise search layer for finding the right policy document, not just answering from a fixed script
- +Now backed by ServiceNow's broader platform and resources
Cons
- −Its product roadmap is now bundled inside ServiceNow's priorities following the acquisition, with less certainty about Moveworks as a standalone product going forward
- −No public pricing; every deal requires a sales conversation
The AI stack
A production HR agent needs a reasoning model, a grounded and current handbook, a scoped read-only HRIS connection, and a hard rule for what it should never answer on its own.
Reasoning model
Reads the question, retrieves the relevant policy, and decides whether it's safe to answer directly
Claude Haiku 5.5
$0.10/M input tokens (up to 100K), $0.50/M outputHigh-volume, routine policy Q&A
Claude Sonnet 5.5
$2/M input, $10/M outputThe main question-answering and escalation-decision loop
GPT-6.1 Sol
$2/M input, $10/M outputTeams already standardized on OpenAI
Our pick: Claude Sonnet 5.5 for the main loop, since the escalation decision matters more here than raw speed, with Haiku 5.5 handling clearly routine lookups.
Knowledge & retrieval
Grounds every policy answer in your actual, current handbook instead of generic HR knowledge
Vendor-native knowledge base (Leena, Workday, Moveworks)
Included on the enterprise contractThe buy-SaaS path
Custom RAG over your handbook and policy docs
Roughly $0.15/mo on Supabase pgvector for a small corpusCustom builds
Our pick: A custom RAG pipeline over your actual handbook is the practical default outside enterprise scale, since it reaches the same accuracy without the platform floor.
HRIS integration
Lets the agent check a real PTO balance or benefits status instead of giving a generic policy answer
Native HRIS connector (within an enterprise platform)
Included on the enterprise contractBuy-SaaS
Read-only API call to your HRIS
Developer timeCustom builds and n8n
Our pick: A scoped, read-only API connection to your existing HRIS reaches the same practical outcome as a native enterprise connector for a fraction of the cost.
Observability & evals
Catches a wrong or legally risky answer before it's repeated to another employee
Vendor dashboards
Included on the enterprise contractBuy-SaaS
Langfuse
Free up to 50K units/mo, then $29+/moCustom builds
Our pick: Review a sample of real transcripts weekly on any path, with particular attention to anything that touched a leave, accommodation or complaint-adjacent question.
Reference architecture
The hardest engineering problem is distinguishing a routine policy lookup from a question that only sounds routine but actually carries legal exposure, since a leave or accommodation question answered confidently from general policy text, instead of escalated to a human with full context, is where real liability lives.
Question arrives
Slack, Teams or intranet chatAn employee asks a question in natural language
Classify the question
Reasoning modelThe model determines whether this is a routine policy lookup, an account-specific check, or something sensitive
Retrieve and look up
RAG plus HRIS tool callThe model pulls the relevant handbook section or calls the HRIS for an account-specific fact like a PTO balance
Escalation check
Guardrail layerAnything touching a grievance, accommodation, discipline or legal gray area routes straight to a human HR generalist
Respond or escalate
Chat interfaceThe agent answers directly for routine cases, or hands off with a summary for sensitive ones
Human follow-up
HR generalistEscalated cases get the full conversation history and the agent's reasoning for context
Logging and eval
Observability toolEvery interaction is logged, and a sample is reviewed weekly for accuracy and for escalation rule drift
Estimated cost per request
~$0.001-$0.02 per employee question answered in model costs alone, before any vendor platform fee
Cost calculator
Drag the sliders to model your actual usage. The numbers update in real time so you can stress-test economics before writing a single line of code.
Estimate your monthly cost at your own employee headcount and question volume.
Estimated monthly cost
$6.00
≈ $72.00 per year
Calculator notes
- This calculator reflects a custom build or DIY path; the three major enterprise vendors don't publish a per-employee or per-question rate that scales down to this volume.
- Model cost covers question classification, handbook retrieval and any HRIS lookup, and applies on every custom or DIY path.
- At 200 employees, Leena AI's reported 1,000-seat minimum would still apply in full, making its effective per-employee cost roughly five times higher than the headcount shown here.
- A custom build's cost barely moves with employee count, since the model cost scales with questions asked, not headcount alone.
Build it yourself with vibe-coding tools
n8n can run a working HR policy assistant in an afternoon: a Slack trigger, an AI Agent node with a RAG tool over your handbook and a read-only HRIS lookup, and a hard escalation rule for anything sensitive.
Time to MVP
1-3 days for a first version, 1-2 weeks to get escalation rules production-ready
Total cost to MVP
$0-$70 (n8n Cloud Starter at roughly 20 EUR/mo plus model API costs)
You'll need
Starter prompt
Set up a Slack trigger (a slash command or a mention in a dedicated HR-questions channel) feeding into an AI Agent node using Claude Sonnet 5.5 as the model, with a Vector Store tool indexing your employee handbook and an HTTP Request tool for read-only HRIS lookups by employee ID. Give the AI Agent node this system prompt: 'You are an HR assistant for {Business Name}. Answer employee questions about PTO balances, benefits enrollment and company policy using only the handbook search tool and the HRIS lookup tool. Never answer from general knowledge about typical company policies; if the handbook doesn't cover it, say so and escalate. Always escalate instead of answering: any question involving a formal complaint, harassment, discrimination or grievance; any leave request that might involve FMLA or an ADA accommodation; anything about discipline, performance or termination; and anything you are not confident is answered correctly by the handbook. When escalating, post a one-sentence summary to the HR review channel and tell the employee a person will follow up, without speculating on an answer yourself.' Add a Switch node after the AI Agent: route escalations to a private HR Slack channel, and route confident, routine answers back to the employee directly. Log every question and answer to a Google Sheet so HR can review a sample weekly for accuracy.
Paste this into n8n
Follow-up prompts (run in order)
- 1
Add a step that checks for recent handbook updates and flags any answer that might be based on an outdated policy version
- 2
Add a monthly digest to HR leadership summarizing the most common questions asked, to spot a gap in the handbook
- 3
Add a multi-language version for a distributed workforce that doesn't all work in the same language
- 4
Add a confirmation step before any HRIS write action, such as updating an address or emergency contact
Expected output
A working Slack assistant that answers routine PTO and policy questions directly and escalates anything legally sensitive to HR with a clear summary, instead of guessing.
Known gotchas
- !The AI Agent node will confidently answer from general knowledge about typical company policies instead of your actual handbook unless explicitly told not to
- !A handbook update that isn't re-indexed means the agent keeps answering from a stale policy version
- !Leave and accommodation questions often don't announce themselves clearly; an employee asking about 'extra time off for a health thing' is an escalation, not a routine PTO question
- !Giving the agent write access to the HRIS instead of read-only access is a real risk if a generated update goes wrong
- !There's no built-in eval harness: you have to manually sample transcripts or wire in Langfuse yourself
Compliance & risk reality check
An HR agent limited to policy Q&A and HRIS lookups carries far less regulatory weight than one that scores or decides, but the data it touches and the topics it brushes against still matter.
Keep the agent to answering, not deciding
Illinois HB 3773 (effective January 1, 2026) requires notice to employees whenever AI is used in hiring, promotion or discipline, and similar algorithmic-decision rules (Colorado SB 26-189, effective 2027; EU AI Act high-risk obligations, December 2027) target AI that makes or materially influences a consequential employment decision.
Mitigation: Scope the agent strictly to answering policy questions and looking up an employee's own data; never let it score, rank or recommend a decision about promotion, discipline or termination.
Sensitive HR and health-adjacent data
PTO, leave and benefits questions routinely brush against FMLA and ADA accommodation territory, where a wrong or incomplete answer carries real legal exposure for the employer.
Mitigation: Hard-code an escalation rule for any leave request that might involve a medical, family or disability accommodation, and never let the agent make that determination itself.
Employee data privacy
A read-only HRIS connection still exposes the agent, and whichever vendor hosts it, to personal data including salary, benefits elections and sometimes dependent information.
Mitigation: Scope the HRIS credential to only the fields the agent actually needs, log every lookup, and check the vendor's data retention and training policy before connecting anything.
Formal complaints and grievances
A harassment, discrimination or retaliation complaint routed to an AI agent instead of a human risks both a legally inadequate response and real harm to the employee reporting it.
Mitigation: Hard-code an immediate human handoff for anything resembling a formal complaint, with no AI-generated response in between.
Build vs buy: the real math
2-4 weeks for a RapidDev build; 1-3 days for a basic n8n version
Custom build time
$8K-$15K for a single-channel policy assistant (up to $20K wired into a full HRIS); $0-$70/mo for DIY
One-time investment
Never, in practical terms, below roughly 1,000 employees, which is where Leena AI's reported seat minimum starts to make sense on a per-employee basis
Breakeven vs buying
Leena AI's reported $150/employee/yr pricing with a 1,000-seat minimum means a 200-person company would still pay the full $150K/yr floor, working out to $750 per employee, five times the headline rate. A $15K RapidDev build amortized over a year costs about $1,250/mo, or $15,000/yr total, roughly 10% of that floor for the same core job of answering policy and PTO questions. Workday Sana's reported $45K-$65K AI transformation fee plus per-user costs tells a similar story: it only makes sense once you're already a Workday customer at real enterprise scale. Below that scale, a scoped custom build is simply the more honest math.
Skip the DIY — RapidDev builds the production version
A Lovable MVP gets you a demo. Production needs auth that doesn't leak data, AI calls that don't bankrupt you, observability when models drift, and code you can audit. That's what we ship through our AI app development service.
Discovery call (free)
30 minWe map your exact human resources AI agent use case: who uses it, target volume, AI model choice, integrations, compliance scope. You get a detailed scope document and fixed-price quote within 48 hours.
AI-accelerated build
2-4 weeks for a RapidDev build; 1-3 days for a basic n8n versionOur engineers use Claude Code, Lovable, and custom tooling to ship 3–5x faster than agencies. You see weekly progress in a staging environment — not a black box.
Launch + handoff
1 weekWe deploy to your infrastructure, transfer the GitHub repo, set up CI/CD and monitoring, and train your team. You own 100% of the source code, prompts, and model configurations.
What you get
Timeline
2-4 weeks for a RapidDev build; 1-3 days for a basic n8n version
Investment
$8K-$15K for a single-channel policy assistant (up to $20K wired into a full HRIS); $0-$70/mo for DIY
vs SaaS
ROI in Never, in practical terms, below roughly 1,000 employees, which is where Leena AI's reported seat minimum starts to make sense on a per-employee basis
30-min call. Fixed-price quote within 48 hours. No commitment.
Frequently asked questions
How much does an AI agent for HR cost?
Leena AI is reported at around $150/employee/yr with a 1,000-seat minimum, roughly a $150K/yr floor, and Workday Sana adds a reported $45K-$65K AI transformation fee on top of per-user costs. A custom RapidDev build runs $8K-$15K upfront (up to $20K for a full HRIS integration), plus $50-$250/mo in model and infra costs. A DIY n8n version costs $0-$70/mo but takes your own time to build and maintain.
How long does it take to set up?
The major enterprise platforms take months and typically require an existing Workday or ServiceNow tenant. A custom agency build takes 2-4 weeks because it has to be wired into your handbook and HRIS. A basic n8n version can work within 1-3 days, though getting escalation rules production-ready takes 1-2 weeks.
Can RapidDev build this for us?
Yes. RapidDev has shipped 600+ apps and offers a free 30-minute consultation to scope your handbook, HRIS and escalation rules before you commit to anything.
Is a $150K/yr tool really the only option?
No. Leena AI's reported 1,000-seat minimum and Workday's enterprise pricing only make financial sense at real enterprise scale. A scoped custom build reaches the same policy-Q&A and PTO-lookup job for $8K-$20K upfront, a small fraction of that floor, for any company below roughly 1,000 employees.
Can it handle a harassment complaint or a grievance?
No, and it shouldn't. A well-built agent escalates any formal complaint, harassment report or grievance straight to a human HR generalist without attempting an AI-generated response in between.
Will it know my company's specific policies, or just generic HR advice?
Only if it's grounded in your actual handbook. A well-built agent retrieves from your current written policies rather than answering from general knowledge about what companies typically do, and it should say so when a question isn't covered.
Is employee data safe with an AI HR agent?
Only with the right scoping. Use a read-only HRIS credential limited to the fields the agent actually needs, log every lookup, and check any vendor's data retention and training policy before connecting sensitive employee data.
What happens with a leave or accommodation question?
A well-built agent treats any question that might touch FMLA or an ADA accommodation as an automatic escalation to a human, since the legal exposure from a wrong automated answer is too high to risk.
Want the production version?
- Delivered in 2-4 weeks for a RapidDev build; 1-3 days for a basic n8n version
- You own 100% of the code
- AI cost monitoring built in
30-min call. No commitment.
