AI in HR means using software to do the administrative half of people work: answering policy questions from your own handbook, running the onboarding and offboarding checklist, keeping employee records complete, and chasing the paperwork that has a legal deadline. Used well, it frees whoever handles HR to spend their time on the half that needs a person, which is hiring decisions, pay, conflict and anything that touches someone's livelihood.
Most of what gets written about AI in human resources assumes you have an HR department. It talks about talent intelligence platforms, workforce analytics and enterprise suites with a six-month rollout. Most companies don't look like that. At twenty or forty people, HR is usually an operations manager, an office manager or the owner, doing it on top of their real job. That person doesn't need a strategy. They need the leave question from Tuesday answered without them, and the new starter's laptop ordered before Monday.
This post is for that person. It covers what HR actually does, which parts AI can take today, which parts it should never touch, and what the law already says. At Sanaf this work is the AI HR assistant job, one of the jobs on the roster you can give Sanaf. Where we quote its job description we say so. Outside facts come from the source they are linked to, read on 2 October 2026.
What does HR do, when there's no HR department?
The US Bureau of Labor Statistics describes the professional version of the role in one line: "Human resources specialists recruit, screen, and interview job applicants and place newly hired workers in jobs." It puts their median pay at $75,940 a year in May 2025 and projects the occupation to grow 6 percent from 2025 to 2035, faster than average, according to its occupational outlook page.
That definition is about hiring, and hiring is the visible part. In a company without a dedicated HR person, the day-to-day load looks more like this:
- Answering questions about leave, expenses, holidays and benefits
- Onboarding new starters: forms, accounts, equipment, the first-week plan
- Offboarding leavers: access removed, equipment back, final paperwork
- Keeping employee records complete and current
- Meeting paperwork deadlines, such as the I-9 for every US hire
- Writing and updating HR policies and the employee handbook
- Scheduling interviews, reviews and training
- Handling conflict, discipline, pay and performance
Look down that list and a split shows up. The first seven items are administration. They follow a document or a checklist, they repeat, and when they go wrong it is because nobody had the time. The last item is judgment. It is personal, it is sometimes legal, and it is the part where being handled by software is itself the harm.
That split is the whole idea behind using AI in HR sensibly. Hand over the administration. Keep the judgment.
Why HR admin goes wrong in a growing team
The AI HR assistant job page has a line that sums up the problem: "In most companies without an HR department, HR admin belongs to whoever has the least room for it."
You can usually spot it from three signs, which the job description lists:
- Onboarding is different every time because it lives in someone's head.
- Employee paperwork is incomplete and nobody knows which parts.
- The same policy questions reach you every week.
None of these happen because people are careless. They happen because HR admin is interruption work. Each question takes two minutes, each form takes five, and the person doing it has a full job besides. So the two-minute jobs get batched for a quiet afternoon that never comes, and the gaps only show up later as a missing form, an account nobody set up, or a leaver who still has access to the shared drive three months on.
What AI in HR can take on first
Here are the five jobs worth handing over, roughly in the order we would start.
1. Policy questions, answered from your own documents
"How many days of leave do I have left?" "Is the day after Thanksgiving a holiday?" "What's the limit on a client lunch?" These questions all have answers, and the answers are written down in a PTO policy, an expenses policy or the employee handbook that nobody reads. So people ask the busiest person in the building instead.
This is the safest place to start with AI for HR, because the answer already exists. The job here is retrieval and a plain reply, with a link to the paragraph it came from. Someone asks in Slack or Microsoft Teams, gets the answer in a minute, and can check the source themselves.
The important rule is what happens when the policy doesn't cover the question. The job page is clear about it: it "says the policy does not cover it and gets a person. Inventing an answer to an HR question is a particularly expensive mistake, because the person asking will reasonably treat the answer as the company speaking." Any HR AI software you look at should behave the same way. Ask it a question your handbook doesn't answer and see what it does. If it improvises, don't buy it.
A side benefit: every question the policy couldn't answer is a gap in the policy. After a month you have a list of what your handbook should say and doesn't.
2. The new hire onboarding checklist, run the same way every time
A new hire onboarding checklist exists in most companies. It just exists in one person's memory, and what they remember depends on how busy that week was. Two people hired a month apart get two different first weeks.
Written down, a decent employee onboarding checklist for a US hire covers four stretches:
Before day one. Offer letter signed and filed. Laptop and accounts requested. Email address created. Calendar invites sent for the first week. Manager reminded to plan the first task.
Day one. Form I-9 Section 1 completed by the employee. Tax forms collected. Handbook shared and acknowledged. Accounts working. Someone assigned to take them to lunch.
Within three business days. I-9 Section 2 completed. This one has a federal deadline. The USCIS employer handbook says: "Within three business days of the date employment begins, you or your authorized representative must complete Section 2," and gives the example that someone who starts on a Monday needs Section 2 done "on or before Thursday of that week." It is exactly the kind of step that slips when it lives in a head (USCIS M-274, section 4.0).
First month. Benefits enrollment done before its window closes. Training booked. A thirty-day check-in on the manager's calendar.
An AI HR assistant runs that list as a list. It sends the requests, chases the forms, books the meetings and tells you on Wednesday afternoon that Section 2 is still open. It works from the records and folders you already keep, which for most teams means Google Drive, Gmail and a spreadsheet, or an HR system like BambooHR if you have one. You don't need to buy an HR platform first.
3. The employee offboarding checklist, which is where the risk is
Onboarding mistakes are annoying. Offboarding mistakes are expensive. A leaver who keeps access to the CRM, a laptop that never comes back, a final paycheck that misses a state deadline: these are the ones that turn into problems months later.
An employee offboarding checklist covers access removal across every app, equipment return, final pay handled by whoever runs payroll, benefits and COBRA notices where they apply, the exit conversation, and updating the records. Offboarding usually happens fast and under some stress, which is why the job page says it runs offboarding "so the steps people forget under pressure get done."
What it does not do is the conversation itself. Telling someone they are leaving is a person's job, every time.
4. Employee records that are actually complete
Ask most companies which employee files are missing a signed handbook acknowledgment, an emergency contact or a current address, and the honest answer is "we'd have to check." The job description puts it plainly: it "keeps staff records complete and says which parts are missing."
That second half is the useful part. A monthly note that says "four people have no emergency contact on file and one I-9 is missing its Section 2 signature" is worth more than any dashboard, because it tells you what to fix.
5. HR policies and the employee handbook
Drafting is where generative AI in HR is genuinely good. A first draft of a PTO policy, a remote work policy or an expenses policy takes an AI a few minutes, and it can match the tone of the handbook you already have.
But a draft is a draft. Employment law differs by state and changes often, so every policy should be read by someone who knows the rules where your people work, and anything with legal weight should be checked by an employment lawyer. Use the AI to get from a blank page to something worth reviewing. Don't use it as the review.
What should never be handed to AI
The AI HR assistant job has one boundary that matters more than any other on the roster. Its job description says: "You do not handle grievances, discipline, pay decisions, or anything sensitive. Those go directly to a person, and you say so to the employee."
Pay attention to the last clause. If an employee writes in about a complaint, the reply is not a policy summary or a form. It tells them their message has gone to a named person, and it goes to that person straight away. Nobody should find out later that their grievance sat in an AI's queue.
The same line applies to:
- Pay and promotion decisions. Data can inform them. A person makes them.
- Discipline and termination. No exceptions.
- Performance judgments. An AI can collect what was agreed in last quarter's review. It does not rate anyone.
- Hiring decisions. AI can schedule interviews and organize candidates. A person decides who gets the job.
Lawmakers have started drawing the same line.
What the law already says about AI in hiring
If you use AI anywhere near hiring, two rules are worth knowing even if you aren't directly covered by them, because they show where regulation is heading.
New York City Local Law 144. Since 5 July 2023, employers and employment agencies in New York City using an automated employment decision tool must have it "subject to a bias audit within one year of the use of the tool," make information about the audit publicly available, and give candidates notice, according to the city's Department of Consumer and Worker Protection. NYC Local Law 144 is the first law of its kind in the US, and it's a good checklist for anyone choosing AI recruiting software.
The EU AI Act. Annex III classes as high-risk any AI system "intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates," along with systems used "to make decisions affecting terms of work-related relationships, the promotion or termination of work-related contractual relationships" or to "monitor and evaluate the performance and behaviour of persons." High-risk systems carry heavy obligations.
Read those two together and you see what regulators worry about: AI that makes decisions about people, such as screening, ranking, promotion, termination and monitoring. The things we suggested starting with, policy answers, checklists, records and paperwork, sit outside that line. We picked them for that reason. The risk there is low and the time saved is high.
None of this is legal advice. If you are in New York City, the EU or any other place with rules on AI in hiring, check with an employment lawyer before you use AI on candidates.
Will AI replace HR?
Not the HR people, going by the numbers. The BLS projects 6 percent growth for HR specialists over the next decade, which is faster than average. What AI is replacing is the version of HR where a capable person spends half their week answering the same leave question and chasing the same missing form.
The bigger change is for companies that never had HR at all. A business with thirty people rarely has a full-time HR specialist, so the work falls to someone with another job. For that company, AI in HR means the onboarding checklist finally gets run properly, and the person who used to squeeze it in at 9 p.m. gets their evenings back.
Examples of AI in HR you can start this week
If you want to try it without a project, here is what a first week looks like with Sanaf's HR assistant job.
- Connect the folder where your policies live, and your email and calendar. That's enough to begin.
- Point it at your handbook and tell it, in plain words, who handles what: "Anything about pay, complaints or someone's performance goes to Dana, today."
- Ask it the five questions people ask you most. Check the answers against the document. Correct anything wrong.
- Give it your next new starter. Hand it the start date and the manager, and let it run the onboarding checklist. Watch which steps it chases.
- Ask for a records check. "Which employee files are missing something?" The answer will probably surprise you.
Everything it sends starts as a draft until you decide otherwise, and it asks before anything it can't undo. Every action is kept in a record you can read. Our security page explains how connections work and how to revoke them one app at a time.
AI for HR professionals vs. AI for teams without HR
The search results for AI tools for HR mix two different buyers.
HR professionals in companies with an HR team want HR AI software that sits inside their HR information system: screening at volume, analytics, an employee self-service chatbot for a thousand people. Those products are real and often worth it at that size, and they come with that size's price and rollout time.
Teams without HR want the admin done. They don't have an HRIS to plug into, and they don't need workforce analytics. They need someone to own the checklist. That is the gap an AI HR assistant fills, and it's why it works from the files and email you already have instead of asking you to buy a platform first.
If hiring is your bigger problem, the recruiting side is a separate job: the AI recruiter schedules interviews and keeps candidates informed, and our guide to the best AI recruiting software compares the dedicated tools. Policy and training that carry a legal deadline belong to the compliance job, and the handbook itself, kept findable and current, sits with knowledge management.
Questions people ask about AI in HR
What is AI in HR?
It is the use of AI to handle human resources work. In practice, the safe and useful part is administration: answering policy questions from your documents, running onboarding and offboarding checklists, keeping records complete and drafting policies for a person to review.
What are the benefits of AI in HR?
Faster answers for employees, onboarding that is the same every time, fewer missed paperwork deadlines and records you can trust. The less obvious benefit is that the person who handled HR on the side gets their time back for the conversations that need them.
Is it safe to use AI for HR?
For administration, yes, with two conditions. It should answer only from your own documents and hand anything it can't answer to a person. And it must never handle grievances, discipline, pay or hiring decisions. Ask any vendor how their product handles both.
Can AI make hiring decisions?
It shouldn't, and in some places the law constrains it. New York City requires bias audits and candidate notice for automated employment decision tools, and the EU AI Act treats recruitment AI as high-risk. Use AI to organize hiring. Let people decide.
Do I need HR software before I use AI for HR?
No. Sanaf's HR assistant job works from the records and policy documents you already keep, often a shared folder and a spreadsheet. If you already run an HR system, it can work with that too.
How much does it cost?
Sanaf starts free with $120 of credit and no card. After that you buy credits as you need them and pay for the work done, with no seat fee. The pricing page has the details.
Where to start
If one of these sounds familiar, it's worth an afternoon: the same leave question reaches you every week, your last two hires had different first weeks, or you aren't sure every I-9 is complete.
Read the full AI HR assistant job, see every job Sanaf can hold, or start free and point it at your handbook.