AI in HR: the five jobs it does today, and how each one looks in PeopleForce

Somebody on your team asked an AI a work question today. Maybe about a policy, maybe about a colleague's availability, maybe "rewrite this feedback so it sounds kinder." AI is already part of your HR work. What's still open is whether it runs inside systems you govern or in a browser tab you'll never see.

This article walks each job and shows how it looks in PeopleForce today, including what is in early access or available on request, then closes with the security principles and the five questions to ask any vendor. The examples come from PeopleForce, our own platform. The five jobs, the principles, and the test apply to any HR platform you evaluate.

The AI your team already uses without asking you

Start with the uncomfortable number, and note where it comes from. Cyberhaven Labs' 2026 AI Adoption & Risk Report found that 39.7% of employees' AI interactions involve sensitive data, and that about a third of employees reach AI tools through personal accounts: outside single sign-on, outside central logging, and outside your retention rules.

That is the working definition of shadow AI: work AI use that IT and HR never see.

For HR this is not an abstract governance worry. The questions people take to a public chatbot are often HR questions ("how do I phrase this feedback," "what does parental leave usually look like"), and the paste that goes with the question can contain a name or a salary. People paste because it's the fastest way to get an answer. The real risk is HR data that's unreachable where people work, or reachable only by pasting it somewhere it shouldn't go.

That's the frame for everything below: the job of AI in HR is to give people a governed way to do what they're already doing ungoverned.

AI in HR crossed from pilot to practice

Adoption is now measurable, and it's smaller than vendor marketing suggests. In SHRM's State of AI in HR 2026:

Adoption is narrow: most usage is a handful of specific chores. That's exactly the right expectation to carry into the five jobs.

Job 1: Answering the questions that fill HR's inbox

What it is

A conversational assistant that answers employees' and managers' questions from the company's own knowledge base and people records: what the travel policy says, who's away next week, how much leave is left, who reports to whom. It replaces the daily trickle of questions that each take a minute of HR's attention and arrive by the dozen.

What good looks like

Three tests separate a real answering assistant from a chat-shaped search box:

In PeopleForce

This job is our AI chatbot, built into the platform. It:

It works on the web and in the PeopleForce mobile app. Every answer is limited by the asker's own field-level permissions.

Job 2: Connecting HR data to the AI your team already uses

What it is

One answer to shadow AI. Instead of pretending people will abandon the assistant they already work in, the platform exposes HR data to that assistant: governed, permissioned, revocable. In the industry this runs on the Model Context Protocol (MCP), an open standard that lets AI assistants connect to business systems; a growing number of HR vendors now offer MCP integrations.

What good looks like

Per-user authorization, so each person connects their own assistant and can revoke it. The same permission model as the platform itself, so the assistant sees only what its user may see. And a capability list you can read out loud: a named, bounded set rather than "your agent can do anything."

In PeopleForce

This job is PeopleForce MCP, our own MCP server. Over it, an assistant can look up who is away, how much leave someone has left, and who works where and reports to whom. Every lookup runs as the person asking: for current employees, each answer is bounded by that person's role, directory visibility and field-level restrictions, just as in PeopleForce.

Your data stays governed here and reaches their assistant on your terms. What the assistant receives becomes part of the conversation there, under that assistant's own data settings. That gives the tools people already use a governed feed of HR data, so the same data doesn't have to leave through a personal account.

Job 3: Screening and parsing candidates

What it is

One of the longest-running AI jobs in HR, and the most common one in SHRM's survey: reading CVs so recruiters don't have to do the first pass by hand. The pressure keeps rising: 84% of job seekers used AI in their search, and the median number of applicants per filled job rose to 69 from 58 a year earlier (Workday Global Workforce Report, October 2026). Parsing turns an uploaded resume into structured fields; scoring compares candidates against a vacancy's stated criteria; generation drafts the screening criteria from the job description itself.

What good looks like

Hiring is where AI carries real regulatory and fairness weight. The EU AI Act classifies hiring-related AI as high-risk, and while the Digital Omnibus (Regulation (EU) 2026/1744) moved those obligations to 2 December 2027, the obligations are still coming. So the bar is explicit criteria the humans wrote, visible reasoning, and a human decision at the end.

In PeopleForce

PeopleForce runs this job through three recruiting features.

Job 4: Visualizing data on demand

What it is

Conversational analytics means typing a question about your data and getting a chart back, with no report builder involved. The trend started in BI tools and has reached people data. Its real effect is organizational. In many companies, only one or two people know how to get answers out of the HR system; conversational analytics makes the data answerable by everyone who's allowed to see it.

What good looks like

The chart should render where the question was asked, and it should draw only on data the asker is permitted to see. It should also say what each number counts, so the reader can check the chart before acting on it.

In PeopleForce

In PeopleForce, this is the AI chatbot's chart answering: ask for "people per department as a bar chart" and it draws the chart inline (bar, pie or line, up to 50 data points) and formats multi-attribute answers as tables, inside the same permission bounds as everything else it does. Today the charts are people counts, grouped by fields such as department, location, position or hire year; averages, trends and turnover still live in your reports.

Job 5: Doing a share of the work

What it is

The newest job, and the one many vendors are now pitching: not an assistant that answers questions about the work, but an agent that produces some of it, pitched across the market as "handling the first 80%" of a process, or as supervisor agents coordinating worker agents underneath. The category is moving from answering to contributing. SAP, for example, released new agents for its Performance and Goals Assistant in Q3 2026 and is rolling out general availability in phases: its Goal Creation Agent writes SMART performance goals from company strategy and the employee's role, and its Calibration Equity Agent flags inconsistent ratings before or during calibration.

What good looks like

The honest version of this job isn't an agent that decides. It's an agent that contributes work a human owns, and three things make that difference:

Where those three are missing, you don't have a contributor. You have an unattributed opinion in your records.

The principles that separate governed AI from a leak

The five jobs are the what; these principles are the how. They apply to any HR platform, ours included. If you forward one screen of this article to whoever signs off on your AI plan, make it this list. Accountability is the gap to close first: in a Gartner survey of 297 CHROs published in October 2026, only 24% said their leadership team agreed on who is accountable for AI agents.

Five questions to ask any HR AI before you roll it out

Skip the feature list; run this test. It's five questions, and the first one takes thirty seconds in any demo.

1. Ask it "who's out this week?" in your own words, unscripted. A real assistant answers from live data; a chat-shaped search box changes the subject.

2. "What can it see, and whose permissions does it use?" The only acceptable answer: the asker's own, everywhere, with no AI-specific side door.

3. "Can it take actions, and if so, what stands between a prompt and a change in my records?" An assistant that only answers is a legitimate answer. So is action with explicit human approval, or an agent whose output lands as an attributed contribution rather than a silent edit. "It just does it" is not.

4. "Where do answers come from, and can the reader check?" Look for cited sources on policy answers and explicit criteria on scores.

5. "Who turns it on, and how granularly?" Per-account, per-role, per-feature switches mean you can adopt at your pace; one big AI toggle means you can't.

We sell an HR platform with AI in it, so we aren't neutral here. Ask us these five questions in a demo too, and hold our answers to the same standard.

The data foundation under all five jobs

Every job in this article depends on one unglamorous thing: the data underneath it is complete, current and permissioned in one place. An assistant is only as good as the records it answers from; a score is only as fair as the criteria and data behind it. Gartner's research makes the point (October 2026):

That's the real order of operations for AI in HR: govern the data first, then let the AI at it. If your people data already lives in one governed home, the five jobs are much easier to switch on. If it doesn't, that's the project, and it pays off far beyond AI.

FAQ

Is AI going to replace the HR team?

The five jobs in this article are chores: answering repeated questions, moving data safely to the tools people use, first-pass screening, drawing charts, and producing a first draft somebody else owns. Each one frees time for judgment work. The decisions (hire, promote, act on survey results) should stay with people, and in PeopleForce they do.

What's the safest first AI use in HR?

The answering job. It touches no decisions, its errors are easy to catch (a policy answer cites the article it came from, so a wrong summary is one click from the source), and it removes the most daily friction. Screening comes second, with explicit criteria and human review.

How do we deal with employees already using public chatbots for work questions?

Give them a governed alternative for the HR half of it, and clear guidance for the rest. Bans rarely hold against convenience; a faster answer inside the platform gives people a reason to switch. That's the design logic behind our AI chatbot and PeopleForce MCP alike.

What does PeopleForce actually have today?

The main ones:

• The AI chatbot, in early access: answers, people lookups, who's-out, leave balances and requests, your Inbox, 1:1s and objectives, inline charts, on the web and in the mobile app.

• PeopleForce MCP for external AI assistants.

• Recruiting AI: resume parsing, on by default, plus criteria generation and CV scoring, available on request, with review disclaimers.

• Performance AI, available on request: a writing coach for review drafts, AI insights on completed reviews, and an AI reviewer you can add to a review cycle.

Each has its own switch, and the AI chatbot is also granted role by role.