Agentic HR Playbook
How HR leaders build with AI, step by step
At the 2026 Charter Workplace Summit in New York, Clara Shih shared that she believes the HR team will be the most critical team for AI transformation in the coming years. A big statement, and one I stand behind wholeheartedly. Yet so many messages to HR right now feel diminutive: implement AI, so you can do more with less.
We think about things differently at Leapsome. We believe in unleashing HR teams from the administration and bureaucracy, and enabling our expertise to thrive in Talent Management, Recruitment Strategy, and Organizational Design work. Agentic HR means handing the repeatable parts of HR work to AI agents (the policy questions, the document chasing, the workflows that just need to run) so your time goes to the work that genuinely needs a person: judgment, relationships, the hard conversations.
This playbook shows you what Agentic HR looks like in practice: real builds from HR leaders using AI in their daily work right now, broken down step by step. What they built, why they built it, and exactly how, so you can take any one of these and run it yourself this week.
And that's my one ask of you: play. The only way to learn this is by doing it. Don't overthink your first attempts to learn, and don't worry about getting it wrong. Nobody is really an expert in this yet (truly, nobody), so the advantage goes to the people willing to try, fail, and try again. Start somewhere imperfect. Enjoy your work. It's why we love what we do so much.
Featuring
Emily Mabie
Melanie Naranjo
Sonja Grabner
Adam Horne
Paulina Sophie Augustin
Jonathan Durnford-Smith
Agentic People Ops
Featuring Emily Mabie (Zapier), Melanie Naranjo (Ethena) & Sonja Grabner (BEAT81)
Predictive People Insights
Featuring Adam Horne (Open Org), Jessica Zwaan (Leapsome) & Paulina Sophie Augustin (Kayzen)
Coming soon
Future-Proof Recruiting
Featuring Jonathan Durnford-Smith (Synthesia)
Coming soon
Automation
Agentic People Ops
You've explained the PTO carryover policy to the 77th person this year. You've copy-pasted the same onboarding contract into Gmail eighteen times this week. Somewhere in there, you were supposed to find an hour to hash out a new comp philosophy with your CEO. Instead, you spent it on admin a system could have handled without you.
That's the whole premise behind Agentic People Ops: agents handle the stuff that just needs to run (onboarding, document chasing, the same three questions on repeat), so the people team is left with the parts only a person can do. "Agents," "MCPs," "agentic AI-powered automations," the vocabulary feels like it was built for engineers, not HR. The idea underneath is simpler: the busywork moves onto autopilot.
In this chapter, three HR leaders show you exactly how they build with AI: a Slack helpdesk that answers HR questions on its own, a custom bot that stops vague performance reviews before they go out, and an onboarding workflow in Leapsome that runs with no manual setup.
1. Build an HR helpdesk that answers all the small questions (so you don't have to)
Meet Emily, AI Innovation Lead for People at Zapier.
Emily is arguably the queen of HR automations, heading up this topic at Zapier (a platform that is literally all about automation). Emily's perspective is that anyone can be technical, even if they don't write code, and she really earned her AI chops by trying stuff out, experimenting, getting stuff wrong, and sharing what she learned. So we could think of no better person to speak to for our first AI build. Here we'll talk you through an automation she set up for recurring HR requests in Slack.
Emily MabieAI Innovation Lead for People
You might have the most up-to-date and in-depth HR policy document in the world, but surely all your employees have not read that 15-page document, nor have they memorized it. So when an HR question comes up, be it on parental leave or development budget, the same questions will pop up again and again. To take this off her team's plate, Emily created a virtual HR helper with Slack (but you can apply this same setup wherever your HR questions live) and Zapier. Here's how to build it.
Read transcript
Picking and testing the trigger
"When a new message comes in in our Slack channel, then I want those people to get their needs met. So the trigger is the 'when' part of the sentence. And if you're thinking, 'Emily, we don't use Slack, we use something else,' great news, Zapier has over 9,000 different apps to choose from. I'm choosing Slack, but you could explore a different app your team uses. I'll say: new message posted to channel. This will wake up our automation whenever there's a new message posted in that channel. For you guys, I'm using the test webinar channel, but if I were to turn this on internally, I'd use the HR channel, wherever most people ask their questions."
Writing the prompt
"Our trigger wakes up the automation, and then any actions we put after it are going to do the things I want in order to meet the goal. I'm using this thing called AI by Zapier. At the top, it has a prompt, instructions we write for AI to take action. In this case, we're going to give it tools so it can go do things we want. Number one, I want it to read the whole thread of messages, because at Zapier a lot of times there'll be seven or eight messages of people adding context and I don't want it to just read the top message. I want it to use knowledge sources, because we've got documentation in HR, our problem isn't needing more documentation, it's needing people to get what they need out of it. I want it to give the actual answer if it's in the documentation. I don't want it to guess or overwhelm the user. If something is sensitive, somebody comes into the channel and says 'my colleague punched me at an offsite, what should I do?' that doesn't belong in the channel. I want to reroute them to the right person. And if they have an answer, put it in the Slack thread, don't make the channel messy."
Adding the tools
"The last thing you need to know about using this tool is it cannot do any of those things unless we add in the tools to allow it to reach into our apps. I asked it to read the whole thread of Slack messages, we didn't give it a tool for that, so we have to give it the Slack tool. We also have to give it knowledge sources, wherever we've documented our HR policies. And I also need to give it the ability to reply in Slack. Just because we gave it permission to read a thread doesn't mean it has permission to send a message, and vice versa. The last thing I want to do is make sure it has access to our HR documentation, I'm not going to add one more tool, I'm going to add a knowledge source, point the agent at wherever I keep my documentation."
The live test
"I'll go ahead and choose our laptop provisioning guide, point it there, and click preview to see it work. Remember, I posted in the Slack channel: 'Hey, I'm a new hire, my laptop hasn't arrived yet, is there something I should be doing?' My agent build is kicking off, reading the whole thread, searching our knowledge source. [Result shown:] 'Hey Emily, great question. You're in the UK. It usually takes up to two weeks to get your laptop. In the meantime, please don't use your personal computer for work. If it's been more than two weeks, let us know.' It's not that I'm replacing an HR team, we never had a whole HR team whose only job was to monitor Slack. This just augments their time so they have more time to do the human stuff. We're not replacing, we're just augmenting."
- Start where the repeat questions already live
Wherever the requests are landing, whether it's a Slack channel, an email chain, or a Notion request board, this should be your starting point. For Emily, that's the Slack channel where HR questions pile up, so in this use case, she used the "new message posted to channel" trigger and pointed it there. Zapier integrates with a ton of different platforms, so you can be pretty flexible here.
- Add an AI step that checks your policy docs before it answers
Right after the trigger, add an AI step, and tell it what to check before it replies: read the full thread for context, then check it against your HR policy documentation, wherever that lives, Notion, Google Drive, an internal wiki. If the answer is already written down, it answers directly. If the question touches something that needs a judgment call, like a specific leave situation or a performance issue, it skips answering and routes it to a person instead.
- Give it permission to do something
Right now the AI step can think, but it can't act. In Zapier, add three connections: one that lets it read the Slack thread, one that lets it reply in the thread, and a knowledge source pointing at your HR documentation (for Emily, a laptop provisioning guide). Skip any of these and it can draft a perfectly good answer that never reaches anyone.
- Test it on the question you're dreading, not a fake one
Emily tested hers on a real question already sitting in the queue: a new hire asking where their laptop was. The bot pulled the answer from the provisioning guide, replied correctly, and held up on a follow-up without losing the thread. Don't test yours on something you already know the answer to. Point it at whatever's unresolved right now.
Context is what makes this work, Emily says: the AI only answers as well as the documentation it's pointed at. "We're not replacing," she says, "we're just augmenting." The hours that used to go into finding and re-sharing the same links now go back to the team.
How to do this in Leapsome
The same answers, already inside Leapsome
Emily built her helpdesk because the answers lived in a policy doc nobody opened. Leapsome's HR Helpdesk Agent does the same job, built in: load your carry-over rules, parental leave policy, and expense guidelines into the Knowledge hub once, and when someone asks Leapsome AI how many days they can carry over, the answer comes from what's written down, inside the platform they already use. Anything that needs a judgment call goes to a person.
Read transcript
One extra day every week
"What would your HR team do with one extra day every week? For most teams, that day disappears into admin work, summarizing survey comments, chasing review drafts, writing meeting notes, and building development plans one person at a time. That's what Leapsome's AI agents take off your plate. They work across the entire platform: in surveys, goals, meetings, reviews, and everything in between. In this walkthrough, I'll show you how."
One connected data model
"An AI agent is only as good as what it knows. Most HR AI data sits on top of a single tool and sees a single slice of your people. Leapsome agents work on one connected data model, HRIS, recruitment, performance, engagement, goals, learning, and compensation. So when an agent drafts a review, suggests a goal, or builds a learning path, it already has the full picture. Access rights and anonymity stay in place the whole time."
Engagement survey
"To show you what it looks like, let's follow one quarter in the life of an HR team. We can start with our latest engagement survey. I've got dozens of comments and a leadership meeting on Thursday. I do not have time to read them all. Thankfully, the assistant does it for me. It can group my comments by themes, show where sentiment shifted since the last survey, and point to the teams that need special attention. It also suggests what to do next, which topics managers should raise with their team, and where HR should step in."
Turning it into a goal
"The survey told us where to focus. Now let's turn that into a goal for the next quarter. The assistant drafts the goal and the key results. From there, I'm able to edit the goal so it perfectly fits my needs."
The weekly one-on-one
"Next up, the weekly one-on-one. Before the meeting, the assistant suggested talking points based on goals, recent feedback, and past meetings. I'm able to pick up the conversation exactly where the last one ended. Because I'm transcribing within Leapsome and taking notes, afterwards I'll get a summary giving me insights into how the conversation went and how I can improve for next time. It'll also pull out various action items or other things that need to be done in Leapsome automatically. Because the agent serves as both an assistant and a coach, I can ensure that my one-on-one meetings are more effective moving forward."
Review time
"It's now the end of the cycle, which means review time. This is where managers often get stuck. The review assistant pulls in context from the whole quarter: goal progress, feedback, and meeting notes. It helps phrase observations with concrete examples and suggests growth-oriented next steps, and the manager stays in charge. The assistant suggests, and you ultimately decide what goes into the review."
From feedback to a learning path
"The review is done, and now let's turn that feedback into growth. Based on the review, the assistant suggests a learning path, with content, milestones, and timing. I can discard the content that doesn't work for me or ask it to regenerate. Progress is tracked automatically and feeds straight into the next review."
One connected loop, then workflows
"Surveys, goals, one-on-ones, reviews, and learning paths. One quarter, one connected loop. At every step, an agent did the heavy lifting while people made the decisions. And because every agent works in the same data, you can take it one step further. With workflows, all this context starts working on its own. A review closes and a learning path is assigned automatically. Engagement drops on a team, and the manager gets a one-on-one talking point and a check-in task. I can upload documentation I've already built to the workflow assistant, and it will build out the baseline for me. Like everything else, I can always go in and adjust the steps to perfectly fit my needs."
A dashboard for leadership
"And when leadership asks how the company is really doing, you don't need to spend a week pulling reports. Describe what you need, and Leapsome builds a dashboard in seconds. Survey themes, goal progress, review results, and learning activity all in one view. CEO-ready, filterable, and linked back to the underlying data."
Wrap-up
"That's a look at agentic HR with Leapsome, AI agents that work as an extension of your HR team across the entire employee lifecycle. Human review by default, clear permissions, and outputs you can track back to data. Your agents draft, summarize, and suggest, so you have the capacity to lead. Thank you so much."
2. Create a reviews GPT that asks for the receipts
Meet Melanie, Chief People Officer at Ethena.
Melanie heads up the People org at Ethena, a compliance and ethics software company.
Melanie's issue was that she had a streamlined performance review process that was carefully thought out and implemented, yet she was finding that managers were often showing up with vague and ineffective feedback in review cycles. Without good feedback, the whole effort was essentially pointless. It was not scalable to have HR manually check and review feedback, so they needed a scalable way to increase the standard across the board. So Melanie built a custom GPT to handle this.
Melanie NaranjoChief People Officer, Ethena
To build something similar, here's how.
Read transcript
Live demo
"This is a custom GPT that I built for performance reviews, you'll see I put in just the worst possible prompt so you can see the power. I said: 'My direct report is struggling to meet their deadlines and isn't a great communicator, can you help me start drafting their performance review?' And rightfully, it said, 'Please provide me the job description before I begin,' because it knows you can't give good feedback if you don't know what metrics you're measuring. I copied and pasted one, and it said, 'Thanks for providing the job description and the salary band, since we are missing several critical details required by the performance review standards, I've set up the official template layout below with the necessary placeholders.' I think sometimes managers struggle with, 'I don't know where to start, I know what I have to say but I don't know how to say it, I don't know how to format it.' The reality is, it doesn't matter to me if you're a good writer or good at putting things into a template, what matters is whether you're providing the important substance. It will ask you for things that are missing: can you provide two to three specific instances where a deadline was missed, because this happens a lot with managers, they'll say 'you're missing deadlines,' and the honest answer when pushed is often 'well, not really.' This just helps them really think. All of this is taking it off my plate so they don't have to talk to me."
"Nobody's an AI expert"
"Don't be scared, don't be intimidated, just open up an app and start asking questions, you will be surprised at how easy it is."
- Set up a saved assistant, not a one-off prompt
Build it once, with its own instructions, so you or your managers can reopen and reuse it every time instead of re-explaining what you want from scratch. Melanie built hers as a custom GPT in ChatGPT, but most AI tools let you save an assistant like this. Give it a name like "Review Assistant," and you're ready to add instructions. With the Leapsome MCP connected, you can set this up in any LLM that supports MCP and let it pull the employee's details from Leapsome instead of pasting them in (there is also an AI review assistant baked directly into the Leapsome platform).
- Set it up to ask for the job description first
In your assistant's instructions, tell it not to draft anything until it has the employee's job description. Feed it a complaint with no context, and it should stop you before it drafts anything real.
- Have it build the review template and flag the gaps
Once it has the job description, tell it to build out your official review template around it, and flag anything missing, accomplishments, a salary band evaluation, whatever your template requires. Try testing it with a deliberately thin prompt, something like "My direct report is struggling to meet their deadlines and isn't a great communicator, can you help me start drafting their performance review?" Even with next to nothing to go on, it should catch exactly what's missing before it lets you move on.
- Have it push back on vague feedback
A complaint about missed deadlines isn't specific enough for a review, and managers know it when they're pushed on it: ask for an example of a missed deadline, and the honest answer is often "well, not really." Tell your assistant to ask for two or three concrete instances instead, the same follow-up question a manager should ask themselves, and to coach toward what better looks like instead of just naming what's wrong.
- Reuse it every review cycle
Once it's built, managers just describe the issue in plain language, even a rough voice memo works, Melanie says, and the same assistant handles the formatting and the follow-up questions every time, for every review.
Melanie NaranjoChief People Officer, Ethena
How to do this in Leapsome
AI review assistant in Leapsome that has all the context
Melanie's GPT won't write a word until it has the job description. Leapsome's AI Review Assistant starts from the same rule, without the setup: it already knows the employee's role, level, the competencies they're assessed on, and their goal progress and feedback from the cycle. Hit Start, answer its questions by typing or talking, and it pushes back when an answer is vague or doesn't match what's expected at that level, helping managers write better feedback at scale. Every manager gets a draft to edit and confirm, nothing is submitted automatically.
Read transcript
Shared expectations
"Strong performance starts with shared expectations. In Leapsome, competencies define what success looks like across roles and levels, and then reviews turn those expectations into structured, useful conversations."
Drafting a competency framework
"Let's begin with a competency framework. Instead of starting with a blank page, you can describe your organization, roles, values, and behaviors that you want to see from your team. Leapsome AI then drafts a complete framework that fits that context."
Building the review process
"Once the framework is in place, you can bring the same expectations directly into performance reviews in addition to any other questions you might be looking to add. From there, you can then build the process that your organization really needs, whether that's simply a manager review or a full 360. Leapsome Reviews are highly customizable to ensure that you're getting the perfect fit for your organization."
AI help while writing the review
"During the review, AI helps managers turn their observations into clear, constructive feedback. The review assistant can improve phrasing, strengthen examples, and help the reviewer check whether the feedback is both specific and actionable. They can also use relevant Leapsome context, such as goals and previous feedback, to ensure that the manager is including everything in the current scope of the review. The manager still reviews and edits every suggestion before submitting it, but these AI tools really help to keep the conversation on track."
Analysis and calibration
"After assessments are complete, Leapsome brings the results together for analysis and for calibration. AI-generated calibration pre-reads compare the rating in your reviews alongside goals, feedback, competencies, and historical context before the calibration meeting begins. The pre-read flags can be worth discussing, including managers who consistently rate higher or lower than their peers, or ratings that don't seem to align with feedback that the employee was given. These all allow for prompts for better conversation and ultimately give managers and calibration committees the power to be making the best-informed decision about their teams."
From feedback to growth
"That process doesn't end when the review is shared. Leapsome can turn feedback into suggested next steps, including personalized learning paths that help employees act on development priorities noted in their review. With competencies, reviews, goals, and learning all connected, performance management becomes a continuous system. People know what's expected, managers get better context, and HR can support fair decisions at scale. I hope you enjoyed this look into the Leapsome Reviews module today. If you'd like to see more, we'd love to show you. Bye-bye."
3. Set up an onboarding workflow that runs itself in Leapsome
Meet Sonja, People & Culture Lead at BEAT81.
BEAT81 is a Berlin-based fitness company that runs high-intensity workouts in studios across Germany.
Sonja and her team need to onboard over 15 new freelance workout trainers per week across fitness studios in Germany. That's a lot of new people to process. It meant she and her team were bogged down with a ton of repetitive work: sending out contracts and getting people set up on the system, over and over, studio by studio. Sonja worked directly in Leapsome, BEAT81's HRIS, to set up an automated onboarding process instead.
Sonja GrabnerPeople & Culture Lead, BEAT81
Hear Sonja's story
Read transcript
One tool for HRIS, feedback and reviews
"Because we already had Leapsome in place for feedback and development talks, we decided to, instead of launching a third or fourth tool, move our HRIS to Leapsome as well. And now we have that in one tool. Having feedback and the HRIS part connected allows us to automate certain kickoffs. So for example, now that Leapsome basically knows when this person joined and has their contract information in the HRIS part, the review part knows when to set up a new cycle."
Migration and time saved
"The data migration itself was only around three days, so that was actually super quick. It's probably a minimum of 10 hours a month that we save due to having those workflows set up."
Advice to other leaders
"I would tell other leaders to not be scared of it. It was easier than I thought."
- Build the track once per role
Onboarding tracks combine checklists, tasks, and learning modules, and you set them up per role or department instead of rebuilding one from scratch for every hire. Sonja set hers up so a new coach and a new payroll hire get genuinely different onboarding, not the same generic form.
- Let the system assign the work, not you
Tasks get assigned automatically to whoever owns them, the manager, IT, HR, with reminders going out to keep things on schedule. That's the part that used to mean chasing someone down in Slack.
- Start before day one
Pre-boarding shares welcome messages, documents, and role-specific info before the start date, so new hires show up already oriented instead of filling out paperwork on day one.
- Check that it's working
Dynamic onboarding surveys measure how each path is landing, the real version of what Sonja called a pulse check, a real signal three months in, not just a guess.
How to do this in Leapsome
Describe the workflow. Leapsome builds it.
Sonja set up BEAT81's onboarding tracks step by step. Leapsome's PeopleOps Partner Agent can do that groundwork for you: describe the process instead, something like "Onboard a new coach. Contract before day one, studio intro in week one, feedback talk at day 30," and it drafts the steps, owners, reminders, and approval gates, then lays them out on a timeline you can edit. Nothing runs until you switch it on, and every task lands with the right person, so nobody is chasing anyone in Slack.
Read transcript
From admin to building
"Where does your HR team actually spend its time? For a lot of teams, it's still chasing task owners, copying data between tools, and managing processes that could run themselves. That's what Leapsome Workflows fixes. The best companies expect their people teams to build with AI. They've shifted their people work from admin to strategic talent work. With Workflows, teams working with Leapsome make exactly that shift. They can describe the process they're looking to build in plain language, review the draft, and let it run. HR builds instead of administering. Here's an example."
Uploading an existing onboarding checklist
"Let's say you recently moved to a new HR platform like Leapsome. You can take a look at our recommended best practice library built by HR leaders just like you. Or if you have your own best practices or processes that have worked well for you before, you can upload those directly into Leapsome. I've got an onboarding checklist for our marketing team that I want to move over, so we'll do that together now. All I need to do is upload my existing PDF, and Leapsome's workflow assistant will take care of the rest. And here we are. If we accept these settings and zoom in, we can see what needs to happen both before and after the employee's first day."
Reaching the whole employee lifecycle
"Workflows are only as good as what they can reach. Leapsome's Workflows reach the entire employee lifecycle because the platform covers it in such depth. So I can assign my new hire their first learning path on first steps with Leapsome, update employee data, generate the contract, and create an IT ticket to order all of their new equipment. In a fragmented stack, that would take at least four tools and a spreadsheet keeping track of all of them. You review every step, adjust what you want, and activate it once it's ready. Your next HR workflow really is one upload or even just one sentence away."
Beyond onboarding
"You can do so much more than just onboarding with Workflows. You can do offboarding, policy rollouts, probation period conversations, absence follow-up, you name it. Since Workflows are connected to the entire employee lifecycle, all it takes is one click to have a more connected and cohesive employee journey. Thanks for taking the time to watch today, and if you're interested in learning more about Leapsome's Workflows, please don't hesitate to reach out to us. Bye-bye."
reclaimed a month once onboarding moved into automated workflows in Leapsome, Sonja Grabner, People & Culture Lead, BEAT81
Who's behind this
About the contributors
The HR leaders featured in this playbook, in their own words and their own builds.
Jessica Zwaan
VP People Strategy and Operations, Leapsome
Jessica Zwaan is VP People Strategy and Operations at Leapsome. She's spent her career in senior people and talent roles, including COO at Talentful and Whereby, and is the author of two books on people strategy, Built for People and Purpose and Work.
Emily Mabie
AI Innovation Lead for People, Zapier
Emily Mabie is AI Innovation Lead for People at Zapier, where she was the company's first AI automation engineer hired specifically for HR. She builds AI-powered automations and agents for Zapier's own HR team, including the helpdesk build featured in this chapter.
Melanie Naranjo
Chief People Officer, Ethena
Melanie Naranjo is Chief People Officer at Ethena, where she's worked since 2021. Before Ethena, she led People & Culture at Insider Inc., and she writes and speaks regularly on HR leadership, including for TroopHR and the Future of HR podcast.
Sonja Grabner
People & Culture Lead, BEAT81
Sonja Grabner is People & Culture Lead at BEAT81, the Berlin-based fitness company behind its high-intensity training studios across Germany. She's worked in People & Culture at BEAT81 since 2022, and is a certified mediator through the Institut Berlin Mediation.
FAQ
Agentic HR, answered
What is Agentic HR?
Agentic HR means handing the repeatable parts of HR work, like policy questions, document chasing, and workflows that just need to run, to AI agents, so HR teams can focus on judgment, relationships, and the hard conversations: the work that actually needs a person. Agents handle the busywork. People handle everything else.
How is Agentic HR different from automation?
Automation moves data from one app to another on fixed rules a person sets up ahead of time. Agentic AI goes a step further: you give it a goal, and it works out which steps and tools get you there. Most of what gets called "agentic" in HR today sits somewhere on that spectrum, and the builds in this playbook show both ends of it.
Do you need to be technical to build these?
No. None of the builds in this playbook require code. Every one comes with exact, step-by-step instructions, and the HR leaders who built them did it without a technical background.
What's an example of Agentic HR in practice?
A Slack bot that answers policy questions directly from your HR documentation instead of routing every question to a person. A custom AI reviewer that won't let a manager submit vague performance feedback. An onboarding workflow that assigns tasks, sends reminders, and tracks progress with no manual setup. All three are covered step by step in this playbook.
What's the Leapsome Agentic HR Accelerator?
A six-month guided program led by Jessica Zwaan, VP People Strategy and Operations at Leapsome. Over 12 weeks, a cohort builds agentic workflows and their own people analytics dashboards, plus one more use case suited to their business, on Leapsome's full platform.
Do you need to switch to Leapsome to use the ideas in this playbook?
No. Every build here works with tools you already use: Slack, ChatGPT, Zapier, whatever your HR team runs on. Leapsome has AI Agents built directly into the platform too (see the "How to do this in Leapsome" sections throughout), but nothing here requires switching platforms first.